1 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
4 * final.c (rest_of_clean_state): Also look for calls inside sequences.
6 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
9 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
11 (ix86_output_indirect_function_return): New prototype.
12 (ix86_split_simple_return_pop_internal): Likewise.
13 * config/i386/i386.c (indirect_return_via_cx): New.
14 (indirect_return_via_cx_bnd): Likewise.
15 (indirect_thunk_name): Handle return va CX_REG.
16 (output_indirect_thunk_function): Create alias for
17 __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
18 (ix86_output_indirect_jmp): Remove the bool argument.
19 (ix86_output_indirect_function_return): New function.
20 (ix86_split_simple_return_pop_internal): Likewise.
21 * config/i386/i386.md (*indirect_jump): Don't pass false
22 to ix86_output_indirect_jmp.
23 (*tablejump_1): Likewise.
24 (simple_return_pop_internal): Change it to define_insn_and_split.
25 Call ix86_split_simple_return_pop_internal to split it for
27 (simple_return_indirect_internal): Call
28 ix86_output_indirect_function_return instead of
29 ix86_output_indirect_jmp.
31 2018-02-26 Jakub Jelinek <jakub@redhat.com>
34 * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
35 memset and value initialization afterwards.
37 2018-02-26 Christophe Lyon <christophe.lyon@linaro.org>
39 * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
41 2018-02-26 Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
44 * common/config/aarch64/aarch64-common.c
45 (aarch_option_optimization_table[]): Switch
46 off fomit-frame-pointer
48 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
49 Chung-Ju Wu <jasonwucj@gmail.com>
51 * config/nds32/nds32-multiple.md (load_multiple): Disallow
53 (store_multiple): Ditto.
55 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
57 * config.gcc: Add --with-cpu support for nds32 target.
58 * config/nds32/nds32-opts.h (nds32_cpu_type): New.
59 * config/nds32/nds32.opt: Add -mcpu= option.
61 2018-02-25 Segher Boessenkool <segher@kernel.crashing.org>
63 * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
64 isel=yes): Warn for these deprecated options.
66 2018-02-23 David Edelsohn <dje.gcc@gmail.com>
68 * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
69 ISA_2_5_MASKS_EMBEDDED.
71 2018-02-23 Jakub Jelinek <jakub@redhat.com>
73 * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
74 p->max as pointers rather than using iterative_hash_expr.
76 2018-02-23 Carl Love <cel@us.ibm.com>
78 * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
79 macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
81 * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
82 P8V_BUILTIN_VEC_VSIGNED2. Change VSX_BUILTIN_VEC_VUNSIGNED2 to
83 P8V_BUILTIN_VEC_VUNSIGNED2.
85 2018-02-22 Vladimir Makarov <vmakarov@redhat.com>
88 * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
89 * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
91 * lra-constraints.c (curr_insn_transform): Set up
92 LRA_NON_CLOBBERED_ALT for moves processed on the fast path. Use
94 (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
95 * lra-eliminations.c (spill_pseudos): Ditto.
96 (process_insn_for_elimination): Ditto.
97 * lra-lives.c (reg_early_clobber_p): Use the new macros.
98 * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
99 LRA_NON_CLOBBERED_ALT.
101 2018-02-22 Martin Sebor <msebor@redhat.com>
103 PR tree-optimization/84480
104 * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
105 to maybe_diag_stxncpy_trunc. Call it.
106 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
107 from gimple_fold_builtin_strcpy. Print inlining stack.
108 (handle_builtin_stxncpy): Print inlining stack.
109 * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
111 2018-02-22 H.J. Lu <hongjiu.lu@intel.com>
114 * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
115 error when -mindirect-branch=thunk-extern, -fcf-protection=branch
116 and -fcheck-pointer-bounds are used together.
117 (indirect_thunk_prefix): New enum.
118 (indirect_thunk_need_prefix): New function.
119 (indirect_thunk_name): Replace need_bnd_p with need_prefix. Use
120 "_nt" instead of "_bnd" for NOTRACK prefix.
121 (output_indirect_thunk): Replace need_bnd_p with need_prefix.
122 (output_indirect_thunk_function): Likewise.
124 (ix86_code_end): Update output_indirect_thunk_function calls.
125 (ix86_output_indirect_branch_via_reg): Replace
126 ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
127 (ix86_output_indirect_branch_via_push): Likewise.
128 (ix86_output_function_return): Likewise.
129 * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
130 incompatible with -fcf-protection=branch and
131 -fcheck-pointer-bounds.
133 2018-02-22 Steve Ellcey <sellcey@cavium.com>
136 * config/aarch64/aarch64.c (aarch64_print_address_internal):
137 Change gcc_assert call to output_operand_lossage.
139 2018-02-22 Steve Ellcey <sellcey@cavium.com>
141 * doc/extend.texi (__builtin_extend_pointer): Document builtin.
143 2018-02-22 DJ Delorie <dj@redhat.com>
144 Sebastian Perta <sebastian.perta@renesas.com>
145 Oleg Endo <olegendo@gcc.gnu.org>
147 * config/rx/rx.c (rx_rtx_costs): New function.
148 (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
150 2018-02-22 Thomas Preud'homme <thomas.preudhomme@arm.com>
152 * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
154 2018-02-22 Martin Liska <mliska@suse.cz>
157 * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
158 Add "native" as a possible value.
160 2018-02-22 Martin Liska <mliska@suse.cz>
163 * config/i386/i386.c (ix86_option_override_internal):
164 Add "native" as a possible value for -march and -mtune.
166 2018-02-22 Jakub Jelinek <jakub@redhat.com>
169 * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
170 to all type variants.
172 PR tree-optimization/84503
173 * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
174 width as info->bitpos + info->bitsize - start.
175 (merged_store_group::merge_overlapping): Simplify width computation.
176 (check_no_overlap): New function.
177 (imm_store_chain_info::try_coalesce_bswap): Compute expected
178 start + width and last_order of the group, fail if check_no_overlap
180 (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
181 to group if check_no_overlap fails.
183 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
185 * config/rs6000/altivec.md: Delete contraint arguments to
186 define_expand, define_split, and define_peephole2, and in
187 define_insn_and_split if always unused.
188 * config/rs6000/darwin.md: Ditto.
189 * config/rs6000/dfp.md: Ditto.
190 * config/rs6000/rs6000.md: Ditto.
191 * config/rs6000/sync.md: Ditto.
192 * config/rs6000/vector.md: Ditto.
193 * config/rs6000/vsx.md: Ditto.
195 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
197 * config/rs6000/altivec.md: Write output control strings as braced
198 blocks instead of double-quoted strings.
199 * config/rs6000/darwin.md: Ditto.
200 * config/rs6000/rs6000.md: Ditto.
201 * config/rs6000/vector.md: Ditto.
202 * config/rs6000/vsx.md: Ditto.
204 2018-02-21 Jason Merrill <jason@redhat.com>
206 PR c++/84314 - ICE with templates and fastcall attribute.
207 * attribs.c (build_type_attribute_qual_variant): Remove assert.
209 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
211 * ipa-cp.c (determine_versionability): Fix comment typos.
213 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
216 * ipa-cp.c (determine_versionability): Do not version functions caling
219 2018-02-21 Martin Liska <mliska@suse.cz>
222 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
223 Add "native" as a possible value.
224 * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT): Define
225 the macro when native cpu detection is available.
227 2018-02-21 Martin Liska <mliska@suse.cz>
230 * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
231 Add "native" as a possible value.
232 * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
233 when native cpu detection is available.
235 2018-02-21 Jakub Jelinek <jakub@redhat.com>
236 Martin Sebor <msebor@redhat.com>
238 PR tree-optimization/84478
239 * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
241 * gimple-fold.c (get_range_strlen): Make minlen const and assume it
242 can't be NULL. Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
243 support which is conservatively correct, for 2 only stay conservative
244 for maxlen. Formatting and comment capitalization fixes. Add STRICT
245 argument to the 2 argument get_range_strlen, adjust 6 arg
246 get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
248 (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
249 (gimple_fold_builtin_strlen): Pass true as last argument to
252 2018-02-20 Martin Sebor <msebor@redhat.com>
255 * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
256 (builtin_memref::set_base_and_offset): Same. Handle inner references.
257 (builtin_memref::builtin_memref): Factor out parts into
258 set_base_and_offset and call it.
260 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
263 * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
264 is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
265 greater precision. If to_mode is a MODE_PARTIAL_INT, stop the
266 search at the associated MODE_INT.
268 2018-02-20 Jeff Law <law@redhat.com>
271 PR tree-optimization/81592
273 * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
274 for range data rather than using global data.
275 * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
276 range data rather than using global data.
277 * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
278 pass it to children as needed.
279 (struct directive::fmtresult): Similarly.
280 (struct directive::set_width): Similarly.
281 (struct directive::set_precision): Similarly.
282 (format_integer, format_directive, parse_directive): Similarly.
283 (format_none): Accept unnamed vr_values parameter.
284 (format_percent, format_floating, format_character): Similarly.
285 (format_string, format_plain): Similarly.
286 * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
287 the EVRP range analyzer for range data rather than using global data.
288 * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
289 gimple-ssa-evrp-analyze.h
290 (class sprintf_dom_walker): Add after_dom_children member function.
291 Add evrp_range_analyzer member.
292 (sprintf_dom_walker::before_dom_children): Call into the EVRP
293 range analyzer as needed.
294 (sprintf_dom_walker::after_dom_children): New member function.
295 * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
297 (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
298 (evrp_range_analyzer::pop_to_marker): Likewise.
300 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
302 PR tree-optimization/84419
303 * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
304 with the required type if its current type is compatible but
307 2018-02-20 Jakub Jelinek <jakub@redhat.com>
310 * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
313 2018-02-20 Martin Liska <mliska@suse.cz>
316 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
317 possible values if we don't have a hint.
319 2018-02-20 Martin Liska <mliska@suse.cz>
323 * final.c (shorten_branches): Build align_tab array with one
325 * opts.c (finish_options): Add alignment option limit check.
326 (MAX_CODE_ALIGN): Likewise.
327 (MAX_CODE_ALIGN_VALUE): Likewise.
328 * doc/invoke.texi: Document maximum allowed option value for
329 all -falign-* options.
331 2018-02-19 Jakub Jelinek <jakub@redhat.com>
334 * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
335 * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
336 * var-tracking.c (emit_note_insn_var_location): Remove all references
337 to NOTE_INSN_CALL_ARG_LOCATION.
338 (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
339 the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
340 Use copy_rtx_if_shared.
341 * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
342 NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
343 (dwarf2out_var_location): Remove handling of
344 NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
346 * final.c (final_scan_insn): Remove all references to
347 NOTE_INSN_CALL_ARG_LOCATION.
348 (rest_of_clean_state): Likewise. Remove REG_CALL_ARG_LOCATION notes
349 before dumping final insns.
350 * except.c (emit_note_eh_region_end): Remove all references to
351 NOTE_INSN_CALL_ARG_LOCATION.
352 * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
353 * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
354 * config/arc/arc.c (hwloop_optimize): Likewise.
355 * config/arm/arm.c (create_fix_barrier): Likewise.
356 * config/s390/s390.c (s390_chunkify_start): Likewise.
357 * config/sh/sh.c (find_barrier): Likewise.
358 * config/i386/i386.c (rest_of_insert_endbranch,
359 ix86_seh_fixup_eh_fallthru): Likewise.
360 * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
361 * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
362 * config/frv/frv.c (frv_function_prologue): Likewise.
363 * emit-rtl.c (try_split): Likewise. Copy over REG_CALL_ARG_LOCATION
365 (note_outside_basic_block_p): Remove all references to
366 NOTE_INSN_CALL_ARG_LOCATION.
367 * gengtype.c (adjust_field_rtx_def): Likewise.
368 * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
370 * jump.c (cleanup_barriers, delete_related_insns): Likewise.
371 * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
374 * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
377 PR tree-optimization/84452
378 * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
379 expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
382 2018-02-19 Martin Liska <mliska@suse.cz>
385 * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
387 2018-02-19 Martin Liska <mliska@suse.cz>
388 Richard Sandiford <richard.sandiford@linaro.org>
390 PR tree-optimization/82491
391 * gimple-fold.c (get_base_constructor): Make earlier bail out
394 2018-02-19 Carl Love <cel@us.ibm.com>
396 * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
397 BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
399 * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
402 2018-02-19 Sebastian Perta <sebastian.perta@renesas.com>
404 * config/rl78/rl78.md (movdf): New define expand.
406 2018-02-19 Martin Liska <mliska@suse.cz>
409 * doc/invoke.texi: Fix typo.
410 * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
412 2018-02-18 Segher Boessenkool <segher@kernel.crashing.org>
414 * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
415 handle rs6000_single_float and rs6000_double_float specially for
418 2018-02-16 Jeff Law <law@redhat.com>
420 * config/rx/rx.c (add_pop_cfi_notes): New function.;
423 2018-02-16 Jakub Jelinek <jakub@redhat.com>
426 * ipa-inline.c (inline_small_functions): Fix a typo.
428 2018-02-16 Nathan Sidwell <nathan@acm.org>
430 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
432 2018-02-16 Carl Love <cel@us.ibm.com>
434 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
435 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
436 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
437 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
438 expansion to P8V_BUILTIN_VEC_FLOAT2.
440 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
442 PR rtl-optimization/70023
443 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
444 src_regno into account.
446 2018-02-16 Carl Love <cel@us.ibm.com>
448 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
449 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
450 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
451 * config/rs6000/rs6000.c: Remove case statements for
452 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
453 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
454 and P9V_BUILTIN_VEC_VINSERT4B.
455 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
456 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
457 * config/rs6000/vsx.md:
458 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
461 2018-02-16 Carl Love <cel@us.ibm.com>
463 * config/rs6000/altivec.h: Add builtin names vec_extract4b
465 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
467 * config/rs6000/rs6000-c.c: Add the definitions for
468 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
469 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
470 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
471 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
472 definition for insert4b and define insn *insert3b_internal.
473 * doc/extend.texi: Add documentation for vec_extract4b.
475 2018-02-16 Nathan Sidwell <nathan@acm.org>
477 * doc/extend.texi (Backwards Compatibility): Mention friend
478 injection. Note for-scope is deprecated.
479 * doc/invoke.texi (-ffriend-injection): Deprecate.
481 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
483 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
484 that moved to I2, also allow destinations that are a paradoxical
485 subreg (instead of a normal reg).
487 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
490 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
493 2018-02-16 Richard Biener <rguenther@suse.de>
495 PR tree-optimization/84037
496 PR tree-optimization/84016
498 * config/i386/i386.c (ix86_builtin_vectorization_cost):
499 Adjust vec_construct for the fact we need additional higher latency
500 128bit inserts for AVX256 and AVX512 vector builds.
501 (ix86_add_stmt_cost): Scale vector construction cost for
504 2018-02-16 Richard Biener <rguenther@suse.de>
506 PR tree-optimization/84417
507 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
508 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
509 (non_rewritable_lvalue_p): Likewise, use poly-ints.
511 2018-02-16 Martin Liska <mliska@suse.cz>
514 * internal-fn.def (ASAN_CHECK): Set proper flags.
515 (ASAN_MARK): Likewise.
517 2018-02-16 Julia Koval <julia.koval@intel.com>
519 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
522 2018-02-16 Jakub Jelinek <jakub@redhat.com>
525 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
526 Use ++iter rather than iter++ for std::list iterators.
527 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
528 defer deleting them until all nodes in the forest are processed. Do
529 free even leaf nodes. Change to_process into auto_vec.
532 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
533 * vec.h (vec_default_construct): Use memset instead of placement new
534 if BROKEN_VALUE_INITIALIZATION is defined.
535 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
536 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
539 PR rtl-optimization/83723
540 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
541 * lra.c (lra_substitute_pseudo): Likewise. If true, use
542 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
544 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
546 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
548 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
550 PR rtl-optimization/81443
551 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
552 from inner REGs to paradoxical SUBREGs.
554 2018-02-16 Richard Biener <rguenther@suse.de>
556 PR tree-optimization/84399
557 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
558 For operands we can analyze at their definition make sure we can
559 analyze them at each use as well.
561 2018-02-16 Richard Biener <rguenther@suse.de>
563 PR tree-optimization/84190
564 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
565 volatile accesses if the decl isn't volatile.
567 2018-02-15 Jason Merrill <jason@redhat.com>
569 PR c++/84314 - ICE with templates and fastcall attribute.
570 * attribs.c (build_type_attribute_qual_variant): Don't clobber
571 TYPE_CANONICAL on an existing type.
573 2018-02-15 Jakub Jelinek <jakub@redhat.com>
575 PR tree-optimization/84383
576 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
577 dstoff nor call operand_equal_p if dstbase is NULL.
579 PR tree-optimization/84334
580 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
581 also a CONSTANT_CLASS_P, punt.
583 2018-02-14 Jim Wilson <jimw@sifive.com>
585 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
586 first SMALL_OPERAND check. New local min_second_step. Move assert
587 to where locals are set. Add TARGET_RVC support.
588 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
590 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
592 * doc/invoke.texi: Correct -Wformat-overflow code sample.
594 2018-02-14 Martin Sebor <msebor@redhat.com>
596 PR tree-optimization/83698
597 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
598 arrays constrain the offset range to their bounds.
599 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
600 (builtin_access::overlap): Avoid setting the size of overlap if it's
602 (maybe_diag_overlap): Also consider arrays when deciding what values
603 of offsets to include in diagnostics.
605 2018-02-14 Martin Sebor <msebor@redhat.com>
608 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
609 that correspond to the kind of a declaration.
611 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
614 * config/pa/pa.md: Load address of PIC label using the linkage table
615 if the label is nonlocal.
617 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
619 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
620 warning message if user requests -maltivec=be.
621 * doc/invoke.texi: Document deprecation of -maltivec=be.
623 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
626 * config/rs6000/rs6000-c.c: Update definitions for
627 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
628 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
630 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
633 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
634 add _get_ssp intrinsics. Remove argument from
635 __builtin_ia32_rdssp[d|q].
636 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
637 * config/i386/i386-builtin.def: Remove argument from
638 __builtin_ia32_rdssp[d|q].
639 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
640 ix86_expand_special_args_builtin for _rdssp[d|q].
641 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
642 Clear register before usage.
643 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
644 Add documentation for new _get_ssp and _inc_ssp intrinsics.
646 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
648 PR tree-optimization/84357
649 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
650 operand 1 of an ARRAY_REF too.
652 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
655 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
656 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
658 (set_of_reg): New struct.
659 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
660 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
661 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
663 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
664 Split into bitclr, bitset, bitinvert patterns if appropriate.
665 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
666 use rx_fuse_in_memory_bitop.
667 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
668 to named insn, correct maximum insn length.
670 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
673 * machmode.def: Define a complex mode for PARTIAL_INT.
674 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
676 * doc/rtl.texi: Document CSPImode.
677 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
679 (msp430_hard_regno_nregs_with_padding): Likewise.
681 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
684 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
686 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
688 PR rtl-optimization/84169
689 * combine.c (try_combine): New variable split_i2i3. Set it to true if
690 we generated a parallel as new i3 and we split that to new i2 and i3
691 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
692 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
693 those to i2, not i1. Partially rewrite this scan code.
695 2018-02-13 Jakub Jelinek <jakub@redhat.com>
698 * stor-layout.c (place_field): For variable length fields, adjust
699 offset_align afterwards not just based on the field's alignment,
700 but also on the size.
703 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
704 of exps and logs in the use_exp2 case.
706 2018-02-13 Jeff Law <law@redhat.com>
708 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
711 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
714 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
718 * common.opt (gas-loc-support, gas-locview-support): New.
719 (ginline-points, ginternal-reset-location-views): New.
720 * doc/invoke.texi: Document them. Use @itemx where intended.
721 (gvariable-location-views): Adjust.
722 * target.def (reset_location_view): New.
723 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
724 (TARGET_RESET_LOCATION_VIEW): New.
725 * doc/tm.texi: Rebuilt.
726 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
727 (dwarf2out_default_as_locview_support): New.
728 (output_asm_line_debug_info): Use option variables.
729 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
730 (output_loc_list): Likewise.
731 (add_high_low_attributes): Check option variables.
732 Don't output entry view attribute in strict mode.
733 (gen_inlined_subroutine_die): Check option variables.
734 (dwarf2out_inline_entry): Likewise.
735 (init_sections_and_labels): Likewise.
736 (dwarf2out_early_finish): Likewise.
737 (maybe_reset_location_view): New, from...
738 (dwarf2out_var_location): ... here. Call it.
739 * debug.h (dwarf2out_default_as_loc_support): Declare.
740 (dwarf2out_default_as_locview_support): Declare.
741 * hooks.c (hook_int_rtx_insn_0): New.
742 * hooks.h (hook_int_rtx_insn_0): Declare.
743 * toplev.c (process_options): Take -gas-loc-support and
744 -gas-locview-support from dwarf2out. Enable
745 -gvariable-location-views by default only with locview
746 assembler support. Enable -ginternal-reset-location-views by
747 default only if the target defines the corresponding hook.
748 Enable -ginline-points by default if location views are
749 enabled; force it disabled if statement frontiers are
751 * tree-inline.c (expand_call_inline): Check option variables.
752 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
754 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
756 PR tree-optimization/84321
757 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
758 handling. Also check whether the anti-range contains any values
759 that satisfy the mask; switch to a VR_RANGE if not.
761 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
764 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
766 2018-02-13 Martin Jambor <mjambor@suse.cz>
769 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
770 of call statements, also set location of a load to a temporary.
772 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
774 * config/rl78/rl78.c (add_vector_labels): New function.
775 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
776 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
777 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
778 which checks that no arguments are passed.
779 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
780 * doc/extend.texi: Documentation for the new attribute.
782 2018-02-13 Andreas Schwab <schwab@suse.de>
784 * config/riscv/linux.h (CPP_SPEC): Define.
786 2018-02-13 Jakub Jelinek <jakub@redhat.com>
789 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
790 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
791 OPTION_MASK_ISA_AES as first argument to def_builtin_const
792 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
793 instead of OPTION_MASK_ISA_PCLMUL as first argument to
794 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
795 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
796 temporarily for AES and PCLMUL builtins.
798 PR tree-optimization/84339
799 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
800 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
804 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
805 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
806 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
808 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
810 * omp-simd-clone.h: New file.
811 * omp-simd-clone.c: Include omp-simd-clone.h.
812 (expand_simd_clones): No longer static.
813 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
814 cgraph.h and omp-simd-clone.h.
815 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
816 (vect_recog_widen_shift_pattern): Formatting fix.
817 (vect_pattern_recog_1): Don't check optab for calls.
820 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
821 operands[2] into a REG before using gen_lowpart on it.
823 2018-02-12 Jeff Law <law@redhat.com>
826 * config/sh/sh.c (find_barrier): Consider a sibling call
829 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
830 successfully back substituting a reg.
832 2018-02-12 Richard Biener <rguenther@suse.de>
834 PR tree-optimization/84037
835 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
836 parameter, move visited init to caller.
837 (vect_slp_analyze_operations): Separate cost from validity
838 check, initialize visited once for all instances.
839 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
841 * tree-vect-stmts.c (vect_model_simple_cost): Make early
843 (vect_model_promotion_demotion_cost): Likewise.
844 (vectorizable_bswap): Guard cost modeling with !slp_node
845 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
847 (vectorizable_call): Likewise.
848 (vectorizable_conversion): Likewise.
849 (vectorizable_assignment): Likewise.
850 (vectorizable_shift): Likewise.
851 (vectorizable_operation): Likewise.
852 (vectorizable_store): Likewise.
853 (vectorizable_load): Likewise.
854 (vectorizable_condition): Likewise.
855 (vectorizable_comparison): Likewise.
857 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
860 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
861 (ASAN_MARK): Fix fnspec to account for return value, change pointer
862 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
864 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
867 * params.def (inline-min-speedup): Increase from 8 to 15.
868 (max-inline-insns-auto): Decrease from 40 to 30.
869 * ipa-split.c (consider_split): Add some buffer for function to
870 be considered inlining candidate.
871 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
874 2018-02-12 Richard Biener <rguenther@suse.de>
876 PR tree-optimization/84037
877 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
878 matched stmts if we cannot swap the non-matched ones.
880 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
882 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
883 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
884 _mm_maskz_scalef_round_ss): New intrinsics.
885 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
886 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
887 __builtin_ia32_scalefss_round): Remove.
888 (__builtin_ia32_scalefsd_mask_round,
889 __builtin_ia32_scalefss_mask_round): New intrinsics.
890 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
891 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
892 ((match_operand:VF_128 2 "<round_nimm_predicate>"
893 "<round_constraint>")): Changed to ...
894 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
895 "<round_scalar_constraint>")): ... this.
896 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
897 %0, %1, %2<round_op3>}"): Changed to ...
898 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
899 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
900 %2<round_scalar_mask_op3>}"): ... this.
901 * config/i386/subst.md (round_scalar_nimm_predicate): New.
903 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
905 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
906 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
907 (_mm_maskz_sqrt_round_ss): New intrinsics.
908 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
909 (__builtin_ia32_sqrtsd_mask_round)
910 (__builtin_ia32_sqrtss_mask_round): New builtins.
911 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
912 (__builtin_ia32_sqrtss_round): Remove.
913 (__builtin_ia32_sqrtsd_mask_round)
914 (__builtin_ia32_sqrtss_mask_round): New builtins.
915 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
916 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
917 ((match_operand:VF_128 1 "vector_operand"
918 "xBm,<round_constraint>")): Changed to ...
919 ((match_operand:VF_128 1 "vector_operand"
920 "xBm,<round_scalar_constraint>")): ... this.
921 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
922 %0, %2, %<iptr>1<round_op3>}): Changed to ...
923 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
924 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
925 %<iptr>1<round_scalar_mask_op3>}): ... this.
926 ((set_attr "prefix" "<round_prefix>")): Changed to ...
927 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
929 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
932 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
933 Cast vec_cmpeq result to correct type.
934 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
935 Cast vec_cmpgt result to correct type.
937 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
939 * final.c (final_scan_insn_1): Renamed from...
940 (final_scan_insn): ... this. New wrapper, to recover
941 seen from the outermost call in recursive ones.
942 * config/sparc/sparc.c (output_return): Drop seen from call.
943 (output_sibcall): Likewise.
944 * config/visium/visium.c (output_branch): Likewise.
946 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
948 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
951 2018-02-10 Alan Modra <amodra@gmail.com>
954 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
955 Specify LR as an input.
957 2018-02-10 Jakub Jelinek <jakub@redhat.com>
960 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
961 remove_member_access_dummy_vars): New functions.
962 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
963 lower_omp_1, execute_lower_omp): Use them.
965 PR rtl-optimization/84308
966 * shrink-wrap.c (spread_components): Release todo vector.
968 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
970 PR rtl-optimization/57193
971 * ira-color.c (struct allocno_color_data): Add member
972 conflict_allocno_hard_prefs.
973 (update_conflict_allocno_hard_prefs): New.
974 (bucket_allocno_compare_func): Add a preference based on
975 conflict_allocno_hard_prefs.
976 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
977 (color_allocnos): Remove a dead code. Initiate
978 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
980 2018-02-09 Jakub Jelinek <jakub@redhat.com>
983 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
984 constraint from =wa to wa. Avoid a subreg on the output operand,
985 instead use a pseudo and subreg it in a move.
986 (p9_xxbrd_<mode>): Changed to ...
987 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
988 (p9_xxbrd_v2df): New expander.
989 (p9_xxbrw_<mode>): Changed to ...
990 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
991 (p9_xxbrw_v4sf): New expander.
993 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
995 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
997 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
1000 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
1001 multiply in 32-bit mode.
1002 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
1003 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
1006 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1008 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
1009 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
1010 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
1011 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
1013 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
1016 * dwarf2out.c (is_trivial_indirect_ref): New function.
1017 (dwarf2out_late_global_decl): Do not generate a location
1018 attribute for variables that have a non-trivial DECL_VALUE_EXPR
1019 and that are not defined in the current unit.
1021 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1023 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
1024 instead of a libcall for UNORDERED.
1026 2018-02-09 Tamar Christina <tamar.christina@arm.com>
1029 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
1030 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
1032 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1035 * config/s390/s390.c (s390_set_current_function): Invoke
1036 s390_indirect_branch_settings also if fndecl didn't change.
1038 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1040 * config/rs6000/rs6000.md (blockage): Set length to zero.
1042 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1044 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
1046 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1049 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
1050 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
1054 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
1055 PARALLEL incoming that failed vt_get_decl_and_offset check.
1058 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
1059 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
1061 (get_variable_section): For decls in named .bss* sections pass true as
1062 second argument to bss_initializer_p.
1064 2018-02-09 Marek Polacek <polacek@redhat.com>
1065 Jakub Jelinek <jakub@redhat.com>
1068 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
1069 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
1070 Sync some changes from cxx_fold_indirect_ref.
1072 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1074 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
1076 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
1077 (BLOCK_INLINE_ENTRY_LABEL): New.
1078 (dwarf2out_var_location): Disregard inline entry markers.
1079 (inline_entry_data): New struct.
1080 (inline_entry_data_hasher): New hashtable type.
1081 (inline_entry_data_hasher::hash): New.
1082 (inline_entry_data_hasher::equal): New.
1083 (inline_entry_data_table): New variable.
1084 (add_high_low_attributes): Add DW_AT_entry_pc and
1085 DW_AT_GNU_entry_view attributes if a pending entry is found
1086 in inline_entry_data_table. Add old entry_pc attribute only
1087 if debug nonbinding markers are disabled.
1088 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
1089 markers are enabled.
1090 (block_within_block_p, dwarf2out_inline_entry): New.
1091 (dwarf2out_finish): Check that no entries remained in
1092 inline_entry_data_table.
1093 * final.c (reemit_insn_block_notes): Handle inline entry notes.
1094 (final_scan_insn, notice_source_line): Likewise.
1095 (rest_of_clean_state): Skip inline entry markers.
1096 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
1098 * gimple.c (gimple_build_debug_inline_entry): New.
1099 * gimple.h (enum gimple_debug_subcode): Add
1100 GIMPLE_DEBUG_INLINE_ENTRY.
1101 (gimple_build_debug_inline_entry): Declare.
1102 (gimple_debug_inline_entry_p): New.
1103 (gimple_debug_nonbind_marker_p): Adjust.
1104 * insn-notes.def (INLINE_ENTRY): New.
1105 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
1106 inline entry marker notes.
1107 (print_insn): Likewise.
1108 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
1109 (INSN_DEBUG_MARKER_KIND): Likewise.
1110 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
1111 * tree-inline.c (expand_call_inline): Build and insert
1112 debug_inline_entry stmt.
1113 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
1114 inline entry blocks early, if nonbind markers are enabled.
1115 (dump_scope_block): Dump fragment info.
1116 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
1117 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
1118 (gimple_build_debug_inline_entry): New.
1119 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
1120 Enable/disable inline entry points too.
1121 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
1122 (DEBUG_INSN): Describe inline entry markers.
1124 * common.opt (gvariable-location-views): New.
1125 (gvariable-location-views=incompat5): New.
1126 * config.in: Rebuilt.
1127 * configure: Rebuilt.
1128 * configure.ac: Test assembler for view support.
1129 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
1130 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
1131 * dwarf2out.c (var_loc_view): New typedef.
1132 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
1133 (dwarf2out_locviews_in_attribute): New.
1134 (dwarf2out_locviews_in_loclist): New.
1135 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
1136 (enum dw_line_info_opcode): Add LI_adv_address.
1137 (struct dw_line_info_table): Add view.
1138 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
1139 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
1140 (zero_view_p): New variable.
1141 (ZERO_VIEW_P): New macro.
1142 (output_asm_line_debug_info): New.
1143 (struct var_loc_node): Add view.
1144 (add_AT_view_list, AT_loc_list): New.
1145 (add_var_loc_to_decl): Add view param. Test it against last.
1146 (new_loc_list): Add view params. Record them.
1147 (AT_loc_list_ptr): Handle loc and view lists.
1148 (view_list_to_loc_list_val_node): New.
1149 (print_dw_val): Handle dw_val_class_view_list.
1150 (size_of_die): Likewise.
1151 (value_format): Likewise.
1152 (loc_list_has_views): New.
1153 (gen_llsym): Set vl_symbol too.
1154 (maybe_gen_llsym, skip_loc_list_entry): New.
1155 (dwarf2out_maybe_output_loclist_view_pair): New.
1156 (output_loc_list): Output view list or entries too.
1157 (output_view_list_offset): New.
1158 (output_die): Handle dw_val_class_view_list.
1159 (output_dwarf_version): New.
1160 (output_compilation_unit_header): Use it.
1161 (output_skeleton_debug_sections): Likewise.
1162 (output_rnglists, output_line_info): Likewise.
1163 (output_pubnames, output_aranges): Update version comments.
1164 (output_one_line_info_table): Output view numbers in asm comments.
1165 (dw_loc_list): Determine current endview, pass it to new_loc_list.
1166 Call maybe_gen_llsym.
1167 (loc_list_from_tree_1): Adjust.
1168 (add_AT_location_description): Create view list attribute if
1169 needed, check it's absent otherwise.
1170 (convert_cfa_to_fb_loc_list): Adjust.
1171 (maybe_emit_file): Call output_asm_line_debug_info for test.
1172 (dwarf2out_var_location): Reset views as needed. Precompute
1173 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
1174 attribute. Set view.
1175 (new_line_info_table): Reset next view.
1176 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
1177 (dwarf2out_source_line): Likewise. Output view resets and labels to
1178 the assembler, or select appropriate line info opcodes.
1179 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
1180 (optimize_string_length): Catch it. Adjust.
1181 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
1182 dw_val_class_view_list, and remove it if no longer needed.
1183 (hash_loc_list): Hash view numbers.
1184 (loc_list_hasher::equal): Compare them.
1185 (optimize_location_lists): Check whether a view list symbol is
1186 needed, and whether the locview attribute is present, and
1187 whether they match. Remove the locview attribute if no longer
1189 (index_location_lists): Call skip_loc_list_entry for test.
1190 (dwarf2out_finish): Call output_asm_line_debug_info for test.
1191 Use output_dwarf_version.
1192 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
1193 (struct dw_val_node): Add val_view_list.
1194 * final.c (SEEN_NEXT_VIEW): New.
1195 (set_next_view_needed): New.
1196 (clear_next_view_needed): New.
1197 (maybe_output_next_view): New.
1198 (final_start_function): Rename to...
1199 (final_start_function_1): ... this. Take pointer to FIRST,
1200 add SEEN parameter. Emit param bindings in the initial view.
1201 (final_start_function): Reintroduce SEEN-less interface.
1202 (final): Rename to...
1203 (final_1): ... this. Take SEEN parameter. Output final pending
1204 next view at the end.
1205 (final): Reintroduce seen-less interface.
1206 (final_scan_insn): Output pending next view before switching
1207 sections or ending a block. Mark the next view as needed when
1208 outputting variable locations. Notify debug backend of section
1209 changes, and of location view changes.
1210 (rest_of_handle_final): Adjust.
1211 * toplev.c (process_options): Autodetect value for debug variable
1212 location views option. Warn on incompat5 without -gdwarf-5.
1213 * doc/invoke.texi (gvariable-location-views): New.
1214 (gvariable-location-views=incompat5): New.
1215 (gno-variable-location-views): New.
1217 2018-02-08 David Malcolm <dmalcolm@redhat.com>
1219 PR tree-optimization/84136
1220 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
1221 that the result of find_edge is non-NULL.
1223 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1226 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
1227 storing integer register in SImode. Fix cost of 256 and 512
1228 byte aligned SSE register store.
1230 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1232 * config/i386/i386.c (ix86_multiplication_cost): Fix
1233 multiplication cost for TARGET_AVX512DQ.
1235 2018-02-08 Marek Polacek <polacek@redhat.com>
1237 PR tree-optimization/84238
1238 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
1241 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1243 PR tree-optimization/84265
1244 * tree-vect-stmts.c (vectorizable_store): Don't treat
1245 VMAT_CONTIGUOUS accesses as grouped.
1246 (vectorizable_load): Likewise.
1248 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1250 PR tree-optimization/81635
1251 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
1252 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
1253 (test_round_for_mask): New functions.
1254 (wide_int_cc_tests): Call test_round_for_mask.
1255 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
1256 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
1257 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
1258 range returned by get_range_info.
1260 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1263 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
1264 * symtab.c: Include builtins.h
1265 (symtab_node::output_to_lto_symbol_table_p): Move here
1266 from lto-streamer-out.c:output_symbol_p.
1267 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
1268 (output_symbol_p): Move all logic to symtab.c
1269 (produce_symtab): Update.
1271 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1273 * config/s390/s390-opts.h (enum indirect_branch): Define.
1274 * config/s390/s390-protos.h (s390_return_addr_from_memory)
1275 (s390_indirect_branch_via_thunk)
1276 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
1277 (enum s390_indirect_branch_type): Define.
1278 * config/s390/s390.c (struct s390_frame_layout, struct
1279 machine_function): Remove.
1280 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
1281 (indirect_branch_table_label_no, indirect_branch_table_name):
1283 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
1284 (enum s390_indirect_branch_option): Define.
1285 (s390_return_addr_from_memory): New function.
1286 (s390_handle_string_attribute): New function.
1287 (s390_attribute_table): Add new attribute handler.
1288 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
1289 (s390_indirect_branch_via_thunk): New function.
1290 (s390_indirect_branch_via_inline_thunk): New function.
1291 (s390_function_ok_for_sibcall): When jumping via thunk disallow
1292 sibling call optimization for non z10 compiles.
1293 (s390_emit_call): Force indirect branch target to be a single
1294 register. Add r1 clobber for non-z10 compiles.
1295 (s390_emit_epilogue): Emit return jump via return_use expander.
1296 (s390_reorg): Handle JUMP_INSNs as execute targets.
1297 (s390_option_override_internal): Perform validity checks for the
1298 new command line options.
1299 (s390_indirect_branch_attrvalue): New function.
1300 (s390_indirect_branch_settings): New function.
1301 (s390_set_current_function): Invoke s390_indirect_branch_settings.
1302 (s390_output_indirect_thunk_function): New function.
1303 (s390_code_end): Implement target hook.
1304 (s390_case_values_threshold): Implement target hook.
1305 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
1307 * config/s390/s390.h (struct s390_frame_layout)
1308 (struct machine_function): Move here from s390.c.
1309 (TARGET_INDIRECT_BRANCH_NOBP_RET)
1310 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
1311 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
1312 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
1313 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
1314 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
1315 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
1316 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
1317 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
1318 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
1319 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
1320 (mnemonic attribute): Add values which aren't recognized
1322 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
1323 pattern for branch conversion. Fix mnemonic attribute.
1324 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
1325 indirect branch via thunk if requested.
1326 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
1327 ("*indirect_jump"): Disable for branch conversion using out of
1329 ("indirect_jump_via_thunk<mode>_z10")
1330 ("indirect_jump_via_thunk<mode>")
1331 ("indirect_jump_via_inlinethunk<mode>_z10")
1332 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
1333 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
1334 ("casesi_jump_via_inlinethunk<mode>_z10")
1335 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
1336 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
1337 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
1338 ("*indirect2_jump"): Disable for branch conversion.
1339 ("casesi_jump"): Turn into expander and expand patterns for branch
1341 ("return_use"): New expander.
1342 ("*return"): Emit return via thunk and rename it to ...
1343 ("*return<mode>"): ... this one.
1344 * config/s390/s390.opt: Add new options and and enum for the
1347 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1349 * lra-constraints.c (match_reload): Unconditionally use
1350 gen_lowpart_SUBREG, rather than selecting between that
1351 and equivalent gen_rtx_SUBREG code.
1353 2018-02-08 Richard Biener <rguenther@suse.de>
1355 PR tree-optimization/84233
1356 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
1357 changed flag instead of boguously re-using phi_inserted.
1359 2018-02-08 Martin Jambor <mjambor@suse.cz>
1361 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
1362 static local variables.
1364 2018-02-08 Richard Biener <rguenther@suse.de>
1366 PR tree-optimization/84278
1367 * tree-vect-stmts.c (vectorizable_store): When looking for
1368 smaller vector types to perform grouped strided loads/stores
1369 make sure the mode is supported by the target.
1370 (vectorizable_load): Likewise.
1372 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1374 * config/aarch64/aarch64.c (aarch64_components_for_bb):
1375 Increase LDP/STP opportunities by adding adjacent callee-saves.
1377 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1379 PR rtl-optimization/84068
1380 PR rtl-optimization/83459
1381 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
1383 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
1385 PR tree-optimization/84224
1386 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
1387 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
1390 2018-02-07 Iain Sandoe <iain@codesourcery.com>
1393 * config/rs6000/altivec.md (*restore_world): Remove LR use.
1394 * config/rs6000/predicates.md (restore_world_operation): Adjust op
1395 count, remove one USE.
1397 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
1399 * doc/install.texi (Configuration): Document the
1400 --with-long-double-format={ibm,ieee} PowerPC configuration
1404 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
1405 Convert from define_expand to be define_insn_and_split. Rework
1406 float/double/_Float128 conversions to QI/HI/SImode to work with
1407 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
1408 conversions to QI/HImode types did a store and then a load to
1409 truncate the value. For conversions to VSX registers, don't split
1410 the insn, instead emit the code directly. Use the code iterator
1411 any_fix to combine signed and unsigned conversions.
1412 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
1413 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
1414 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1415 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
1416 (fix_<mode>di2_hw): Likewise.
1417 (fixuns_<mode>di2_hw): Likewise.
1418 (fix_<mode>si2_hw): Likewise.
1419 (fixuns_<mode>si2_hw): Likewise.
1420 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
1421 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1422 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
1423 fix<uns>_trunc<SFDF:mode>si2_p8.
1424 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
1426 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1427 (fix<uns>_<mode>_mem): Likewise.
1428 (fctiw<u>z_<mode>_mem): Likewise.
1429 (fix<uns>_<mode>_mem): Likewise.
1430 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1431 the register allocator from doing a direct move to the GPRs to do
1432 a store, and instead use the ISA 3.0 store byte/half-word from
1433 vector register instruction. For IEEE 128-bit floating point,
1434 also optimize stores of 32-bit ints.
1435 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1437 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1439 * genextract.c (push_pathstr_operand): New function to support
1441 (walk_rtx): Call push_pathstr_operand.
1442 (print_path): Support [a-zA-Z].
1444 2018-02-07 Richard Biener <rguenther@suse.de>
1446 PR tree-optimization/84037
1447 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1448 (cse_and_gimplify_to_preheader): Declare.
1449 (vect_get_place_in_interleaving_chain): Likewise.
1450 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1452 (_loop_vec_info::~_loop_vec_info): Delete it.
1453 (cse_and_gimplify_to_preheader): New function.
1454 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1455 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1456 (vectorizable_load): Likewise. For grouped stores always base
1457 the IV on the first element.
1458 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1459 condition before gimplifying.
1461 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1463 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1464 *DIV_EXPR and *MOD_EXPR.
1466 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1469 * config/i386/i386.c (ix86_option_override_internal): Mask out
1470 the CF_SET bit when checking -fcf-protection.
1472 2018-02-07 Tom de Vries <tom@codesourcery.com>
1475 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1478 2018-02-07 Richard Biener <rguenther@suse.de>
1480 PR tree-optimization/84204
1481 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1484 PR tree-optimization/84205
1485 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1486 special-case isl_ast_op_zdiv_r.
1488 PR tree-optimization/84223
1489 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1490 Only add conditions from within the region.
1491 (gather_bbs::after_dom_children): Adjust.
1493 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1496 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1497 * config/avr/avr.md: Only post-reload split REG-REG moves if
1498 either register is REGERAL_REG_P.
1500 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1502 PR tree-optimization/84235
1503 * tree-ssa-scopedtables.c
1504 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1505 if the subtraction is performed in floating point type where NaNs are
1506 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1507 build 1. Formatting fix.
1509 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1512 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1513 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1514 and skip it regardless of bb boundaries. Use CALL_P macro,
1515 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1516 unnecessarily, formatting fix.
1518 2018-02-06 Michael Collison <michael.collison@arm.com>
1520 * config/arm/thumb2.md:
1521 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1522 (*thumb_mov_notscc): Ditto.
1524 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1527 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1528 unsigned_fix, not "s".
1530 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1532 * configure.ac (gcc_fn_eh_frame_ro): New function.
1533 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1534 correct .eh_frame permissions.
1535 * configure: Regenerate.
1537 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1539 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1542 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1544 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1545 Display warning message for -mno-speculate-indirect-jumps.
1547 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1549 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1551 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1553 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1555 PR tree-optimization/84225
1556 * tree-eh.c (find_trapping_overflow): Only call
1557 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1559 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1562 * config/i386/i386.c: Reimplement the check of possible options
1563 -mibt/-mshstk conbination. Change error messages.
1564 * doc/invoke.texi: Fix a typo: remove extra '='.
1566 2018-02-06 Marek Polacek <polacek@redhat.com>
1568 PR tree-optimization/84228
1569 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1571 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1574 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1575 emitted arch directives.
1576 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1577 __ARM_FEATURE_COPROC before changing architectures.
1579 2018-02-06 Richard Biener <rguenther@suse.de>
1581 * config/i386/i386.c (print_reg): Fix typo.
1582 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1584 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1586 * configure: Regenerate.
1588 2018-02-05 Martin Sebor <msebor@redhat.com>
1590 PR tree-optimization/83369
1591 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1594 2018-02-05 Martin Liska <mliska@suse.cz>
1596 * doc/invoke.texi: Cherry-pick upstream r323995.
1598 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1600 * ira.c (ira_init_register_move_cost): Adjust comment.
1602 2018-02-05 Martin Liska <mliska@suse.cz>
1604 PR gcov-profile/84137
1605 * doc/gcov.texi: Fix typo in documentation.
1607 2018-02-05 Martin Liska <mliska@suse.cz>
1609 PR gcov-profile/83879
1610 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1611 using dlopen functionality.
1613 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1615 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1616 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1617 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1618 _mm_maskz_range_round_ss): New intrinsics.
1619 (__builtin_ia32_rangesd128_round)
1620 (__builtin_ia32_rangess128_round): Remove.
1621 (__builtin_ia32_rangesd128_mask_round,
1622 __builtin_ia32_rangess128_mask_round): New builtins.
1623 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1624 __builtin_ia32_rangess128_round): Remove.
1625 (__builtin_ia32_rangesd128_mask_round,
1626 __builtin_ia32_rangess128_mask_round): New builtins.
1627 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1628 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1629 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1630 "<round_saeonly_constraint>")): Changed to ...
1631 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1632 "<round_saeonly_scalar_constraint>")): ... this.
1633 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1634 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1635 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1636 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1637 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1639 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1641 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1643 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1644 Remove all values except native, 8540 and 8548.
1646 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1648 * config/i386/i386.c (ix86_output_function_return): Pass
1649 INVALID_REGNUM, instead of -1, as invalid register number to
1650 indirect_thunk_name and output_indirect_thunk.
1652 2018-02-02 Julia Koval <julia.koval@intel.com>
1654 * config.gcc: Add -march=icelake.
1655 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1656 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1657 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1658 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1659 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1660 (processor_target_table): Add icelake.
1661 (ix86_option_override_internal): Handle new PTAs.
1662 (get_builtin_code_for_version): Handle icelake.
1663 (M_INTEL_COREI7_ICELAKE): New.
1664 (fold_builtin_cpu): Handle icelake.
1665 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1666 * doc/invoke.texi: Add -march=icelake.
1668 2018-02-02 Julia Koval <julia.koval@intel.com>
1670 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1671 to wide_int_bitmask.
1672 * wide-int-bitmask.h: New.
1674 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1677 * config/i386/i386.md: Replace Pmode with word_mode in
1678 builtin_setjmp_setup and builtin_longjmp to support x32.
1680 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1684 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1686 (rs6000_supported_cpu_names): New static variable.
1687 (linux_cpu_translation_table): Likewise.
1688 (elf_platform) <cpu>: Define new static variable and use it.
1689 Translate kernel AT_PLATFORM name to canonical name if needed.
1690 Error if platform name is unknown.
1692 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1695 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1697 2018-02-01 Jeff Law <law@redhat.com>
1700 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1701 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1702 the scratch if RELEASE_VIA_POP is false.
1703 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1704 If we have to save a temporary register, decrement SIZE appropriately.
1705 Pass new arguments to release_scratch_register_on_entry.
1706 (ix86_adjust_stack_and_probe): Likewise.
1707 (ix86_emit_probe_stack_range): Pass new arguments to
1708 release_scratch_register_on_entry.
1710 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1712 PR rtl-optimization/84157
1713 * combine.c (change_zero_ext): Use REG_P predicate in
1714 front of HARD_REGISTER_P predicate.
1716 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1718 * config/avr/avr.c (avr_option_override): Move disabling of
1719 -fdelete-null-pointer-checks to...
1720 * common/config/avr/avr-common.c (avr_option_optimization_table):
1723 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1725 PR tree-optimization/81635
1726 * tree-data-ref.c (split_constant_offset_1): For types that
1727 wrap on overflow, try to use range info to prove that wrapping
1730 2018-02-01 Renlin Li <renlin.li@arm.com>
1733 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1735 (aarch64_register_move_cost): Likewise.
1736 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1738 (REG_CLASS_NAMES): Likewise.
1739 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1740 TAILCALL_ADDR_REGS. Remove IP registers.
1741 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1743 2018-02-01 Richard Biener <rguenther@suse.de>
1745 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1746 for specifying RPO order and allow NULL for that.
1747 * domwalk.c (dom_walker::dom_walker): Likewise.
1748 (dom_walker::walk): Handle NULL RPO order.
1749 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1751 (rewrite_update_dom_walker): Likewise.
1752 (mark_def_dom_walker): Likewise.
1754 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1756 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1757 (aarch64_maybe_expand_sve_subreg_move): Declare.
1758 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1759 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1761 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1762 that are semantically a reverse operation.
1763 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1764 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1765 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1767 (aarch64_can_change_mode_class): For big-endian, forbid changes
1768 between two SVE modes if they have different element sizes.
1770 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1772 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1773 the TImode handling for big-endian targets.
1775 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1777 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1778 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1780 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1781 Remove BSWAP handing for big-endian targets and use the form of
1782 LD1RQ appropariate for the mode.
1784 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1786 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1787 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1790 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1793 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1794 check for operands that need to go through aarch64_sve_reload_be.
1796 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1798 PR tree-optimization/81661
1799 PR tree-optimization/84117
1800 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1801 * tree-eh.c: Include gimplify.h.
1802 (find_trapping_overflow, replace_trapping_overflow,
1803 rewrite_to_non_trapping_overflow): New functions.
1804 * tree-vect-loop.c: Include tree-eh.h.
1805 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1806 * tree-data-ref.c: Include tree-eh.h.
1807 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1809 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1811 PR rtl-optimization/84123
1812 * combine.c (change_zero_ext): Check if hard register satisfies
1813 can_change_dest_mode before calling gen_lowpart_SUBREG.
1815 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1818 * ira.c (ira_init_register_move_cost): Remove assert.
1820 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1822 PR rtl-optimization/84071
1823 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1824 * doc/tm.texi: Regenerate.
1826 2018-01-31 Richard Biener <rguenther@suse.de>
1828 PR tree-optimization/84132
1829 * tree-data-ref.c (analyze_miv_subscript): Properly
1830 check whether evolution_function_is_affine_multivariate_p
1831 before calling gcd_of_steps_may_divide_p.
1833 2018-01-31 Julia Koval <julia.koval@intel.com>
1836 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1837 * config/i386/i386.md (rdpid_rex64) New.
1838 (rdpid): Make 32bit only.
1840 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1843 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1844 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1846 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1849 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1851 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1853 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1855 PR rtl-optimization/84071
1856 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1857 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1859 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1861 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1862 (arc_attribute_table): Add 'aux' attribute.
1863 (arc_in_small_data_p): Consider aux like variables.
1864 (arc_is_aux_reg_p): New function.
1865 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1866 (arc_get_aux_arg): New function.
1867 (prepare_move_operands): Handle aux-register access.
1868 (arc_handle_aux_attribute): New function.
1869 * doc/extend.texi (ARC Variable attributes): Add subsection.
1871 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1873 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1874 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1875 (arc_attribute_table): Add 'uncached' attribute.
1876 (arc_print_operand): Print '.di' flag for uncached memory
1878 (arc_in_small_data_p): Do not consider for small data the uncached
1880 (arc_is_uncached_mem_p): New function.
1881 * config/arc/predicates.md (compact_store_memory_operand): Check
1882 for uncached memory accesses.
1883 (nonvol_nonimm_operand): Likewise.
1884 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1886 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1889 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1890 falign-loops=): Add Optimization flag.
1892 2018-01-30 Jeff Law <law@redhat.com>
1895 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1896 INT_REGISTERS_SAVED. Check it prior to calling
1897 get_scratch_register_on_entry.
1898 (ix86_adjust_stack_and_probe): Similarly.
1899 (ix86_emit_probe_stack_range): Similarly.
1900 (ix86_expand_prologue): Corresponding changes.
1902 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1905 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1906 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1908 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1911 * lra-constraints.c (curr_insn_transform): Process AND in the
1914 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1916 PR rtl-optimization/83986
1917 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1918 dependence against last_pending_memory_flush in addition to
1921 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1923 PR tree-optimization/81611
1924 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1927 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1930 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1933 2018-01-30 Richard Biener <rguenther@suse.de>
1934 Jakub Jelinek <jakub@redhat.com>
1936 PR tree-optimization/84111
1937 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1938 inner loops added during recursion, as they don't have up-to-date
1941 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1944 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1945 (can_inline_edge_by_limits_p): ... here.
1946 (can_early_inline_edge_p, check_callers,
1947 update_caller_keys, update_callee_keys, recursive_inlining,
1948 add_new_edges_to_heap, speculation_useful_p,
1949 inline_small_functions,
1950 inline_small_functions, flatten_function,
1951 inline_to_all_callers_1): Update.
1953 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1955 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1958 2018-01-30 Richard Biener <rguenther@suse.de>
1960 PR tree-optimization/83008
1961 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1962 invariant and constant vector uses in stmts when they need
1965 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1968 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1969 * configure: Regenerate.
1971 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1973 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1975 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1976 Use gen_rtx_REG rather than gen_lowpart.
1978 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1980 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1981 rather than 0 when creating partial subregs.
1983 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1985 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1988 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1991 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1992 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1993 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1994 flags. This restores the settings used before the 2017-07-24.
1995 Turning off pre increment/decrement/modify allows IVOPTS to
1996 optimize DF/SF loops where the index is an int.
1998 2018-01-29 Richard Biener <rguenther@suse.de>
1999 Kelvin Nilsen <kelvin@gcc.gnu.org>
2002 * tree-vect-stmts.c (vectorizable_call): Don't call
2003 targetm.vectorize_builtin_md_vectorized_function if callee is
2006 2018-01-22 Carl Love <cel@us.ibm.com>
2008 * doc/extend.tex: Fix typo in second arg in
2009 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
2011 2018-01-29 Richard Biener <rguenther@suse.de>
2013 PR tree-optimization/84086
2014 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
2015 (flush_ssaname_freelist): When SSA names were released reset
2016 the SCEV hash table.
2018 2018-01-29 Richard Biener <rguenther@suse.de>
2020 PR tree-optimization/84057
2021 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
2022 removed paths when removing edges.
2024 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
2026 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
2027 -mfunction-return=@var{choice}.
2029 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
2032 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
2034 (layout::print_source_line): Likewise.
2035 (test_get_line_width_without_trailing_whitespace): Add test cases.
2037 2018-01-27 Jakub Jelinek <jakub@redhat.com>
2040 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
2043 2018-01-26 Jim Wilson <jimw@sifive.com>
2045 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
2047 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
2050 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2052 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
2053 and CMP + SUB-immediate -> SUBS.
2055 2018-01-26 Martin Sebor <msebor@redhat.com>
2057 PR tree-optimization/83896
2058 * tree-ssa-strlen.c (get_string_len): Rename...
2059 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
2060 Avoid assuming length is constant.
2061 (handle_char_store): Use HOST_WIDE_INT for string length.
2063 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
2066 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
2067 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
2069 2018-01-26 Richard Biener <rguenther@suse.de>
2071 PR rtl-optimization/84003
2072 * dse.c (record_store): Only record redundant stores when
2073 the earlier store aliases at least all accesses the later one does.
2075 2018-01-26 Jakub Jelinek <jakub@redhat.com>
2077 PR rtl-optimization/83985
2078 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
2079 REG_CFA_RESTORE insns.
2080 (delete_unmarked_insns): Don't ignore separate shrink wrapping
2081 REG_CFA_RESTORE insns here.
2084 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
2085 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
2087 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2089 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
2090 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
2091 (arc_init): Likewise.
2092 (arc_override_options): Likewise.
2093 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
2095 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
2097 * config/arc/arc.h (TARGET_DBNZ): Define.
2098 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
2099 properly set the tune attribute.
2100 (dbnz): Use TARGET_DBNZ guard.
2101 * config/arc/arc.opt (mtune): Add core3 option.
2103 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2105 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
2106 recognize new pic like addresses.
2107 (arc_delegitimize_address): Clean up.
2109 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2111 * config/arc/arc-arches.def: Option mrf16 valid for all
2113 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
2114 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
2115 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
2116 * config/arc/arc-tables.opt: Regenerate.
2117 * config/arc/arc.c (arc_conditional_register_usage): Handle
2118 reduced register file case.
2119 (arc_file_start): Set must have build attributes.
2120 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
2122 * config/arc/arc.opt (mrf16): Add new option.
2123 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
2124 * config/arc/genmultilib.awk: Handle new mrf16 option.
2125 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
2126 * config/arc/t-multilib: Regenerate.
2127 * doc/invoke.texi (ARC Options): Document mrf16 option.
2129 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2131 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
2132 * config/arc/arc.c (arc_handle_secure_attribute): New function.
2133 (arc_attribute_table): Add 'secure_call' attribute.
2134 (arc_print_operand): Print secure call operand.
2135 (arc_function_ok_for_sibcall): Don't optimize tail calls when
2137 (arc_is_secure_call_p): New function. * config/arc/arc.md
2138 (call_i): Add support for sjli instruction.
2139 (call_value_i): Likewise.
2140 * config/arc/constraints.md (Csc): New constraint.
2142 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2143 John Eric Martin <John.Martin@emmicro-us.com>
2145 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
2146 * config/arc/arc.c (_arc_jli_section): New struct.
2147 (arc_jli_section): New type.
2148 (rc_jli_sections): New static variable.
2149 (arc_handle_jli_attribute): New function.
2150 (arc_attribute_table): Add jli_always and jli_fixed attribute.
2151 (arc_file_end): New function.
2152 (TARGET_ASM_FILE_END): Define.
2153 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
2154 (arc_add_jli_section): New function.
2155 (jli_call_scan): Likewise.
2156 (arc_reorg): Call jli_call_scan.
2157 (arc_output_addsi): Remove 'S' from printing asm operand.
2158 (arc_is_jli_call_p): New function.
2159 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
2161 (movhi_insn): Likewise.
2162 (movsi_insn): Likewise.
2163 (movsi_set_cc_insn): Likewise.
2164 (loadqi_update): Likewise.
2165 (load_zeroextendqisi_update): Likewise.
2166 (load_signextendqisi_update): Likewise.
2167 (loadhi_update): Likewise.
2168 (load_zeroextendhisi_update): Likewise.
2169 (load_signextendhisi_update): Likewise.
2170 (loadsi_update): Likewise.
2171 (loadsf_update): Likewise.
2172 (movsicc_insn): Likewise.
2173 (bset_insn): Likewise.
2174 (bxor_insn): Likewise.
2175 (bclr_insn): Likewise.
2176 (bmsk_insn): Likewise.
2177 (bicsi3_insn): Likewise.
2178 (cmpsi_cc_c_insn): Likewise.
2179 (movsi_ne): Likewise.
2180 (movsi_cond_exec): Likewise.
2181 (clrsbsi2): Likewise.
2192 (call_i): Remove 'S' asm letter, add jli instruction.
2193 (call_value_i): Likewise.
2194 * config/arc/arc.op (mjli-always): New option.
2195 * config/arc/constraints.md (Cji): New constraint.
2196 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
2198 (subsf3_fpx): Likewise.
2199 (mulsf3_fpx): Likewise.
2200 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
2202 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
2204 * doc/invoke.texi (ARC): Document mjli-always option.
2206 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
2208 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
2209 avoid addition with 0 and use incw and decw where possible.
2211 2018-01-26 Richard Biener <rguenther@suse.de>
2213 PR tree-optimization/81082
2214 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
2215 association if it requires casting to unsigned.
2216 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
2217 from fold_plusminus_mult_expr to catch important cases late when
2218 range info is available.
2220 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2222 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
2223 * configure.ac (hidden_linkonce): New test.
2224 * configure: Regenerate.
2225 * config.in: Regenerate.
2227 2018-01-26 Julia Koval <julia.koval@intel.com>
2229 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
2230 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
2231 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
2232 _mm_mask_bitshuffle_epi64_mask): Fix type.
2233 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
2234 USI_FTYPE_V4DI_V4DI_USI): Remove.
2235 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
2236 __builtin_ia32_vpshufbitqmb256_mask,
2237 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
2238 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
2239 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
2241 2018-01-26 Alan Modra <amodra@gmail.com>
2244 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
2245 UNSPEC_VBPERMQ. Sort other unspecs.
2247 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
2249 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
2251 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
2254 * predict.c (drop_profile): Do not push/pop cfun; update also
2256 (handle_missing_profiles): Fix logic looking for zero profiles.
2258 2018-01-25 Jakub Jelinek <jakub@redhat.com>
2261 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
2262 on functions with #pragma omp declare simd or functions with simd
2264 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
2265 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
2266 Remove trailing \n from warning_at calls.
2268 2018-01-25 Tom de Vries <tom@codesourcery.com>
2271 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2272 for neutered workers.
2274 2018-01-24 Joseph Myers <joseph@codesourcery.com>
2277 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
2278 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
2280 2018-01-24 Jeff Law <law@redhat.com>
2283 * i386.c (get_probe_interval): Move to earlier point.
2284 (ix86_compute_frame_layout): If -fstack-clash-protection and
2285 the frame is larger than the probe interval, then use pushes
2286 to save registers rather than reg->mem moves.
2287 (ix86_expand_prologue): Remove conditional for int_registers_saved
2290 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
2293 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
2294 min/max for never referenced object.
2296 2018-01-24 Jakub Jelinek <jakub@redhat.com>
2299 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
2301 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
2302 attributes from DECL_ATTRIBUTES (decl) without affecting
2303 DECL_ATTRIBUTES (current_function_decl).
2304 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
2305 functions with non-NULL DECL_ABSTRACT_ORIGIN.
2307 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
2309 PR tree-optimization/83979
2310 * fold-const.c (fold_comparison): Use constant_boolean_node
2311 instead of boolean_{true,false}_node.
2313 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
2315 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
2318 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2320 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2321 Simplify the clause that sets the length attribute.
2322 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2323 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
2324 clause that sets the length attribute.
2325 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2327 2018-01-24 Tom de Vries <tom@codesourcery.com>
2330 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
2331 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
2332 Add strict parameter.
2333 (prevent_branch_around_nothing): Insert dummy insn between branch to
2334 label and label with no ptx insn inbetween.
2335 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
2337 2018-01-24 Tom de Vries <tom@codesourcery.com>
2340 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2341 for neutered threads in warp.
2342 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
2344 2018-01-24 Richard Biener <rguenther@suse.de>
2346 PR tree-optimization/83176
2347 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
2350 2018-01-24 Richard Biener <rguenther@suse.de>
2352 PR tree-optimization/82819
2353 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
2354 code generating pluses that are no-ops in the target precision.
2356 2018-01-24 Richard Biener <rguenther@suse.de>
2359 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
2361 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2363 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
2364 to merge probabilities.
2365 * predict.c (probably_never_executed): Also mark as cold functions
2366 with global 0 profile and guessed local profile.
2367 * profile-count.c (profile_probability::combine_with_count): New
2369 * profile-count.h (profile_probability::operator*,
2370 profile_probability::operator*=, profile_probability::operator/,
2371 profile_probability::operator/=): Reduce precision to adjusted
2372 and set value to guessed on contradictory divisions.
2373 (profile_probability::combine_with_freq): Remove.
2374 (profile_probability::combine_wiht_count): Declare.
2375 (profile_count::force_nonzero):: Set to adjusted.
2376 (profile_count::probability_in):: Set quality to adjusted.
2377 * tree-ssa-tail-merge.c (replace_block_by): Use
2380 2018-01-23 Andrew Waterman <andrew@sifive.com>
2381 Jim Wilson <jimw@sifive.com>
2383 * config/riscv/riscv.c (riscv_stack_boundary): New.
2384 (riscv_option_override): Set riscv_stack_boundary. Handle
2385 riscv_preferred_stack_boundary_arg.
2386 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
2387 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
2388 (STACK_BOUNDARY): Set to riscv_stack_boundary.
2389 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
2390 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
2391 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
2393 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
2396 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
2397 of struct ix86_frame.
2398 (ix86_expand_epilogue): Likewise. Add a local variable for
2399 the reg_save_offset field in struct ix86_frame.
2401 2018-01-23 Bin Cheng <bin.cheng@arm.com>
2403 PR tree-optimization/82604
2404 * tree-loop-distribution.c (enum partition_kind): New enum item
2405 PKIND_PARTIAL_MEMSET.
2406 (partition_builtin_p): Support above new enum item.
2407 (generate_code_for_partition): Ditto.
2408 (compute_access_range): Differentiate cases that equality can be
2409 proven at all loops, the innermost loops or no loops.
2410 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
2411 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
2412 (finalize_partitions, distribute_loop): Don't fuse partition of
2413 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
2414 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
2417 2018-01-23 Martin Liska <mliska@suse.cz>
2419 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
2420 order to ignore the predictor.
2421 (PRED_POLYMORPHIC_CALL): Likewise.
2422 (PRED_RECURSIVE_CALL): Likewise.
2424 2018-01-23 Martin Liska <mliska@suse.cz>
2426 * tree-profile.c (tree_profiling): Print function header to
2427 aware reader which function we are working on.
2428 * value-prof.c (gimple_find_values_to_profile): Do not print
2429 not interesting value histograms.
2431 2018-01-23 Martin Liska <mliska@suse.cz>
2433 * profile-count.h (enum profile_quality): Add
2434 profile_uninitialized as the first value. Do not number values
2435 as they are zero based.
2436 (profile_count::verify): Update sanity check.
2437 (profile_probability::verify): Likewise.
2439 2018-01-23 Nathan Sidwell <nathan@acm.org>
2441 * doc/invoke.texi (ffor-scope): Deprecate.
2443 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2445 PR tree-optimization/83510
2446 * domwalk.c (set_all_edges_as_executable): New function.
2447 (dom_walker::dom_walker): Convert bool param
2448 "skip_unreachable_blocks" to enum reachability. Move setup of
2449 edge flags to set_all_edges_as_executable and only do it when
2450 reachability is REACHABLE_BLOCKS.
2451 * domwalk.h (enum dom_walker::reachability): New enum.
2452 (dom_walker::dom_walker): Convert bool param
2453 "skip_unreachable_blocks" to enum reachability.
2454 (set_all_edges_as_executable): New decl.
2455 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2456 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2458 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2459 but converting true to REACHABLE_BLOCKS.
2460 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2462 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2463 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2464 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2466 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2467 if check_all_array_refs will be called.
2469 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2471 * tree.c (selftest::test_location_wrappers): Add more test
2474 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2476 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2477 (selftest::test_bit_in_range): Likewise.
2479 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2482 * doc/sourcebuild.texi (vect_float): Say that the selector
2483 only describes the situation when -funsafe-math-optimizations is on.
2484 (vect_float_strict): Document.
2486 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2488 PR tree-optimization/83965
2489 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2490 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2491 instead of checking only for a reduction.
2492 (vect_recog_widen_sum_pattern): Likewise.
2494 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2496 * predict.c (probably_never_executed): Only use precise profile info.
2497 (compute_function_frequency): Skip after inlining hack since we now
2498 have quality checking.
2500 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2502 * profile-count.h (profile_probability::very_unlikely,
2503 profile_probability::unlikely, profile_probability::even): Set
2504 precision to guessed.
2506 2018-01-23 Richard Biener <rguenther@suse.de>
2508 PR tree-optimization/83963
2509 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2510 Properly terminate dominator walk when crossing the exit edge not
2511 when visiting its source block.
2513 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2516 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2517 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2519 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2521 PR tree-optimization/83957
2522 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2523 semicolon after for body surrounded by braces.
2525 PR tree-optimization/83081
2526 * profile-count.h (profile_probability::split): New method.
2527 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2528 Use profile_probability::split.
2529 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2530 when splitting a single conditional jump into 2.
2532 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2534 PR tree-optimization/69452
2535 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2538 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2540 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2541 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2542 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2544 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2546 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2547 * config/rl78/rl78.md (movdi): New define_expand.
2548 * config/rl78/rl78.c (rl78_split_movdi): New function.
2550 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2553 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2555 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2556 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2557 128-bit to produce an UNSPEC move to get the double word with the
2558 signbit and then a shift directly to do signbit.
2559 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2560 implementation with a new version that just does either a direct
2561 move or a regular move. Move memory interface to separate insns.
2562 Move insns so they are next to the expander.
2563 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2564 with signbit move. Split big and little endian case.
2565 (signbit<mode>2_dm_mem_le): Likewise.
2566 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2567 (signbit<mode>2_dm2): Likewise.
2569 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2571 * config/rl78/rl78.md (anddi3): New define_expand.
2573 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2575 * config/rl78/rl78.md (umindi3): New define_expand.
2577 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2579 * config/rl78/rl78.md (smindi3): New define_expand.
2581 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2583 * config/rl78/rl78.md (smaxdi3): New define_expand.
2585 2018-01-22 Carl Love <cel@us.ibm.com>
2587 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2588 LVX_V1TI): Add macro expansion.
2589 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2590 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2591 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2592 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2593 Change check to determine if the instruction is a byte reversing
2594 entry. Fix typo in comment.
2595 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2596 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2597 Add def_builtin calls for new builtins.
2598 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2599 Add define_insn expansion.
2601 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2603 * config/rl78/rl78.md (umaxdi3): New define_expand.
2605 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2607 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2608 for non-QImode registers.
2610 2018-01-22 Richard Biener <rguenther@suse.de>
2612 PR tree-optimization/83963
2613 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2614 including the loop exit block.
2615 (scop_detection::merge_sese): Likewise.
2616 (scop_detection::add_scop): Do it here instead.
2618 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2620 * doc/sourcebuild.texi (arm_softfloat): Document.
2622 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2625 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2626 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2627 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2629 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2630 David Edelsohn <dje.gcc@gmail.com>
2633 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2634 Change "crset eq" to "crset 2".
2635 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2636 (*call_indirect_aix<mode>_nospec): Likewise.
2637 (*call_value_indirect_aix<mode>_nospec): Likewise.
2638 (*call_indirect_elfv2<mode>_nospec): Likewise.
2639 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2640 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2641 change assembly output from . to $.
2642 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2643 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2644 (*tablejump<mode>_internal1_nospec): Likewise.
2646 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2649 * config/sh/sh_optimize_sett_clrt.cc:
2650 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2652 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2654 PR tree-optimization/83940
2655 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2656 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2657 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2658 use it to pass back the definition type.
2659 (vect_check_store_rhs): Likewise rhs_dt_out.
2660 (vect_build_gather_load_calls): Add a mask_dt argument and use
2661 it instead of a call to vect_is_simple_use.
2662 (vectorizable_store): Update calls to vect_check_load_store_mask
2663 and vect_check_store_rhs. Use the dt returned by the latter instead
2664 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2665 instead of calls to vect_is_simple_use. Pass the scalar rather
2666 than the vector operand to vect_is_simple_use when handling
2667 second and subsequent copies of an rhs value.
2668 (vectorizable_load): Update calls to vect_check_load_store_mask
2669 and vect_build_gather_load_calls. Use the cached mask_dt and
2670 gs_info.offset_dt instead of calls to vect_is_simple_use.
2672 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2675 * tree-emutls.c: Include gimplify.h.
2676 (lower_emutls_2): New function.
2677 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2678 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2679 it before further processing.
2682 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2683 UINTVAL (trueop1) instead of INTVAL (op1).
2685 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2689 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2690 INCOMING_FRAME_SP_OFFSET if not defined.
2691 (scan_trace): Add ENTRY argument. If true and
2692 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2693 emit a note to adjust the CFA offset.
2694 (create_cfi_notes): Adjust scan_trace callers.
2695 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2696 INCOMING_FRAME_SP_OFFSET in the CIE.
2697 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2698 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2700 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2701 * doc/tm.texi: Regenerated.
2703 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2705 PR rtl-optimization/83147
2706 * lra-constraints.c (remove_inheritance_pseudos): Use
2707 lra_substitute_pseudo_within_insn.
2709 2018-01-19 Tom de Vries <tom@codesourcery.com>
2710 Cesar Philippidis <cesar@codesourcery.com>
2713 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2715 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2718 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2719 spaces for function labels.
2721 2018-01-19 Martin Liska <mliska@suse.cz>
2723 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2724 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2725 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2726 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2727 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2728 (PRED_CONST_RETURN): Change from 69 to 65.
2729 (PRED_NULL_RETURN): Change from 91 to 71.
2730 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2731 (PRED_LOOP_GUARD): Change from 66 to 73.
2733 2018-01-19 Martin Liska <mliska@suse.cz>
2735 * predict.c (predict_insn_def): Add new assert.
2736 (struct branch_predictor): Change type to signed integer.
2737 (test_prediction_value_range): Amend test to cover
2739 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2740 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2741 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2742 (PRED_LOOP_IV_COMPARE): Likewise.
2743 * predict.h (PROB_UNINITIALIZED): Define new constant.
2745 2018-01-19 Martin Liska <mliska@suse.cz>
2747 * predict.c (dump_prediction): Add new format for
2748 analyze_brprob.py script which is enabled with -details
2750 * profile-count.h (precise_p): New function.
2752 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2754 PR tree-optimization/83922
2755 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2756 there are no statements that need masking.
2757 (vect_active_double_reduction_p): New function.
2758 (vect_analyze_loop_operations): Use it when handling phis that
2759 are not in the loop header.
2761 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2763 PR tree-optimization/83914
2764 * tree-vect-loop.c (vectorizable_induction): Don't convert
2765 init_expr or apply the peeling adjustment for inductions
2766 that are nested within the vectorized loop.
2768 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2770 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2773 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2777 * function.h (gimplify_parameters): Add gimple_seq * argument.
2778 * function.c: Include gimple.h and options.h.
2779 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2780 for the added local temporaries if needed.
2781 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2782 if there are any parameter cleanups, wrap whole body into a
2783 try/finally with the cleanups.
2785 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2788 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2789 Use GET_MODE_CLASS for scalar floating point.
2791 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2795 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2796 Fix call of call_cgraph_insertion_hooks.
2798 2018-01-18 Martin Sebor <msebor@redhat.com>
2800 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2802 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2805 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2808 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2811 * common.opt: (-ffile-prefix-map): New option.
2812 * opts.c (common_handle_option): Defer it.
2813 * opts-global.c (handle_common_deferred_options): Handle it.
2814 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2815 * file-prefix-map.h: New file.
2816 (remap_debug_filename, add_debug_prefix_map): ...here.
2817 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2818 * final.c (debug_prefix_map, add_debug_prefix_map
2819 remap_debug_filename): Move to...
2820 * file-prefix-map.c: New file.
2821 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2822 generalize, get rid of alloca(), use strrchr() instead of strchr().
2823 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2824 Implement in terms of add_prefix_map().
2825 (remap_macro_filename, remap_debug_filename): Implement in term of
2827 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2828 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2829 * dbxout.c: Include file-prefix-map.h.
2830 * varasm.c: Likewise.
2831 * vmsdbgout.c: Likewise.
2832 * xcoffout.c: Likewise.
2833 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2834 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2835 * doc/invoke.texi (-ffile-prefix-map): Document.
2836 (-fdebug-prefix-map): Update description.
2838 2018-01-18 Martin Liska <mliska@suse.cz>
2840 * config/i386/i386.c (indirect_thunk_name): Document that also
2842 (output_indirect_thunk): Document why both instructions
2843 (pause and lfence) are generated.
2845 2018-01-18 Richard Biener <rguenther@suse.de>
2847 PR tree-optimization/83887
2848 * graphite-scop-detection.c
2849 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2850 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2851 (scop_detection::merge_sese): Re-implement with a flood-fill
2852 algorithm that properly finds a SESE region if it exists.
2854 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2857 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2858 pointer_diff optimizations use view_convert instead of convert.
2860 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2862 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2863 Generate different code for -mno-speculate-indirect-jumps.
2864 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2865 (*call_indirect_aix<mode>): Disable for
2866 -mno-speculate-indirect-jumps.
2867 (*call_indirect_aix<mode>_nospec): New define_insn.
2868 (*call_value_indirect_aix<mode>): Disable for
2869 -mno-speculate-indirect-jumps.
2870 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2871 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2872 -mno-speculate-indirect-jumps.
2873 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2875 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2877 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2878 long double type, set the flags for noting the default long double
2879 type, even if we don't pass or return a long double type.
2881 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2884 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2887 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2889 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2890 support for merge[hl].
2891 (fold_mergehl_helper): New helper function.
2892 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2893 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2894 (altivec_vmrglw_direct): Add xxmrglw insn.
2896 2018-01-17 Andrew Waterman <andrew@sifive.com>
2898 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2899 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2901 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2904 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2905 call the lto_location_cache before reading the
2906 DECL_SOURCE_LOCATION of the types.
2908 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2909 Richard Sandiford <richard.sandiford@linaro.org>
2911 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2912 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2913 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2914 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2915 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2917 * config/aarch64/constraints.md (aarch64_movti_operand):
2919 * config/aarch64/predicates.md (Uti): Add new constraint.
2921 2018-01-17 Carl Love <cel@us.ibm.com>
2923 * config/rs6000/vsx.md (define_expand xl_len_r,
2924 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2925 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2927 (define_expand, define_insn): Move the shift left from the
2928 define_insn to the define_expand for lxvl and stxvl instructions.
2929 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2930 and XL_LEN_R definitions to PURE.
2932 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2934 * config/i386/i386.c (indirect_thunk_name): Declare regno
2935 as unsigned int. Compare regno with INVALID_REGNUM.
2936 (output_indirect_thunk): Ditto.
2937 (output_indirect_thunk_function): Ditto.
2938 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2939 in the call to output_indirect_thunk_function.
2941 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2944 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2945 rather than the size of inner_type to determine the stack slot size
2946 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2948 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2951 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2954 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2956 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2957 endian Linux systems to optionally enable multilibs for selecting
2958 the long double type if the user configured an explicit type.
2959 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2960 have no long double multilibs if not defined.
2961 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2962 warn if the user used -mabi={ieee,ibm}longdouble and we built
2963 multilibs for long double.
2964 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2965 appropriate multilib option.
2966 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2968 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2969 for building long double multilibs.
2970 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2972 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2974 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2977 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2979 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2982 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2985 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2988 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2990 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2991 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2993 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2995 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2996 different rtl trees depending on TARGET_64BIT.
2997 (rs6000_gen_lvx): Likewise.
2999 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3001 * config/visium/visium.md (nop): Tweak comment.
3002 (hazard_nop): Likewise.
3004 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3006 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
3007 -mspeculate-indirect-jumps.
3008 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
3009 for -mno-speculate-indirect-jumps.
3010 (*call_indirect_elfv2<mode>_nospec): New define_insn.
3011 (*call_value_indirect_elfv2<mode>): Disable for
3012 -mno-speculate-indirect-jumps.
3013 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
3014 (indirect_jump): Emit different RTL for
3015 -mno-speculate-indirect-jumps.
3016 (*indirect_jump<mode>): Disable for
3017 -mno-speculate-indirect-jumps.
3018 (*indirect_jump<mode>_nospec): New define_insn.
3019 (tablejump): Emit different RTL for
3020 -mno-speculate-indirect-jumps.
3021 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
3022 (tablejumpsi_nospec): New define_expand.
3023 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
3024 (tablejumpdi_nospec): New define_expand.
3025 (*tablejump<mode>_internal1): Disable for
3026 -mno-speculate-indirect-jumps.
3027 (*tablejump<mode>_internal1_nospec): New define_insn.
3028 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
3031 2018-01-16 Artyom Skrobov tyomitch@gmail.com
3033 * caller-save.c (insert_save): Drop unnecessary parameter. All
3036 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3037 Richard Biener <rguenth@suse.de>
3040 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
3041 return early, inline manually is_gimple_sizepos. Make sure if we
3042 call gimplify_expr we don't end up with a gimple constant.
3043 * tree.c (variably_modified_type_p): Don't return true for
3044 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
3045 * gimplify.h (is_gimple_sizepos): Remove.
3047 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3049 PR tree-optimization/83857
3050 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
3051 vectorizable_live_operation for pure SLP statements.
3052 (vectorizable_live_operation): Handle PHIs.
3054 2018-01-16 Richard Biener <rguenther@suse.de>
3056 PR tree-optimization/83867
3057 * tree-vect-stmts.c (vect_transform_stmt): Precompute
3058 nested_in_vect_loop_p since the scalar stmt may get invalidated.
3060 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3063 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
3064 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
3065 If off is not INTEGER_CST, issue a may not be aligned warning
3066 rather than isn't aligned. Use isn%'t rather than isn't.
3067 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
3069 <case MULT_EXPR>: Improve the case when bottom and one of the
3070 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
3071 operand, in that case check if the other operand is multiple of
3072 bottom divided by the INTEGER_CST operand.
3074 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3077 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
3078 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
3079 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
3080 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
3081 * config/pa/pa.c (pa_function_arg_advance): Likewise.
3082 (pa_function_arg, pa_arg_partial_bytes): Likewise.
3083 (pa_function_arg_size): New function.
3085 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3087 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
3088 in a separate statement.
3090 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3092 PR tree-optimization/83847
3093 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
3094 group gathers and scatters.
3096 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3098 PR rtl-optimization/86620
3099 * params.def (max-sched-ready-insns): Bump minimum value to 1.
3101 PR rtl-optimization/83213
3102 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
3103 to last if both are JUMP_INSNs.
3105 PR tree-optimization/83843
3106 * gimple-ssa-store-merging.c
3107 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
3108 store_immediate_info for bswap/nop orig_stores.
3110 2018-01-15 Andrew Waterman <andrew@sifive.com>
3112 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
3114 <UDIV>: Increase cost if !TARGET_DIV.
3116 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
3118 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
3119 (define_attr "cr_logical_3op"): New.
3120 (cceq_ior_compare): Adjust.
3121 (cceq_ior_compare_complement): Adjust.
3122 (*cceq_rev_compare): Adjust.
3123 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
3124 (is_cracked_insn): Adjust.
3125 (insn_must_be_first_in_group): Adjust.
3126 * config/rs6000/40x.md: Adjust.
3127 * config/rs6000/440.md: Adjust.
3128 * config/rs6000/476.md: Adjust.
3129 * config/rs6000/601.md: Adjust.
3130 * config/rs6000/603.md: Adjust.
3131 * config/rs6000/6xx.md: Adjust.
3132 * config/rs6000/7450.md: Adjust.
3133 * config/rs6000/7xx.md: Adjust.
3134 * config/rs6000/8540.md: Adjust.
3135 * config/rs6000/cell.md: Adjust.
3136 * config/rs6000/e300c2c3.md: Adjust.
3137 * config/rs6000/e500mc.md: Adjust.
3138 * config/rs6000/e500mc64.md: Adjust.
3139 * config/rs6000/e5500.md: Adjust.
3140 * config/rs6000/e6500.md: Adjust.
3141 * config/rs6000/mpc.md: Adjust.
3142 * config/rs6000/power4.md: Adjust.
3143 * config/rs6000/power5.md: Adjust.
3144 * config/rs6000/power6.md: Adjust.
3145 * config/rs6000/power7.md: Adjust.
3146 * config/rs6000/power8.md: Adjust.
3147 * config/rs6000/power9.md: Adjust.
3148 * config/rs6000/rs64.md: Adjust.
3149 * config/rs6000/titan.md: Adjust.
3151 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3153 * config/i386/predicates.md (indirect_branch_operand): Rewrite
3154 ix86_indirect_branch_register logic.
3156 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3158 * config/i386/constraints.md (Bs): Update
3159 ix86_indirect_branch_register check. Don't check
3160 ix86_indirect_branch_register with GOT_memory_operand.
3162 * config/i386/predicates.md (GOT_memory_operand): Don't check
3163 ix86_indirect_branch_register here.
3164 (GOT32_symbol_operand): Likewise.
3166 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3168 * config/i386/predicates.md (constant_call_address_operand):
3169 Rewrite ix86_indirect_branch_register logic.
3170 (sibcall_insn_operand): Likewise.
3172 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3174 * config/i386/constraints.md (Bs): Replace
3175 ix86_indirect_branch_thunk_register with
3176 ix86_indirect_branch_register.
3178 * config/i386/i386.md (indirect_jump): Likewise.
3179 (tablejump): Likewise.
3180 (*sibcall_memory): Likewise.
3181 (*sibcall_value_memory): Likewise.
3182 Peepholes of indirect call and jump via memory: Likewise.
3183 * config/i386/i386.opt: Likewise.
3184 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3185 (GOT_memory_operand): Likewise.
3186 (call_insn_operand): Likewise.
3187 (sibcall_insn_operand): Likewise.
3188 (GOT32_symbol_operand): Likewise.
3190 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3193 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
3194 type rather than type addr's type points to.
3195 (expand_omp_atomic_mutex): Likewise.
3196 (expand_omp_atomic): Likewise.
3198 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3201 * config/i386/i386.c (output_indirect_thunk_function): Use
3202 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
3203 for __x86_return_thunk.
3205 2018-01-15 Richard Biener <rguenther@suse.de>
3208 * expmed.c (extract_bit_field_1): Fix typo.
3210 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3213 * config/arm/iterators.md (VF): New mode iterator.
3214 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
3215 Remove integer-related logic from pattern.
3216 (neon_vabd<mode>_3): Likewise.
3218 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3221 * common.opt (fstrict-overflow): No longer an alias.
3222 (fwrapv-pointer): New option.
3223 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
3224 also for pointer types based on flag_wrapv_pointer.
3225 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
3226 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
3227 opts->x_flag_wrapv got set.
3228 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
3229 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
3230 POINTER_TYPE_OVERFLOW_UNDEFINED.
3231 * match.pd: Likewise in address comparison pattern.
3232 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
3234 2018-01-15 Richard Biener <rguenther@suse.de>
3237 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
3238 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
3239 Reset type names to their identifier if their TYPE_DECL doesn't
3240 have linkage (and thus is used for ODR and devirt).
3241 (save_debug_info_for_decl): Remove.
3242 (save_debug_info_for_type): Likewise.
3243 (add_tree_to_fld_list): Adjust.
3244 * tree-pretty-print.c (dump_generic_node): Make dumping of
3245 type names more robust.
3247 2018-01-15 Richard Biener <rguenther@suse.de>
3249 * BASE-VER: Bump to 8.0.1.
3251 2018-01-14 Martin Sebor <msebor@redhat.com>
3254 * builtins.c (check_access): Avoid warning when the no-warning bit
3257 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
3259 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
3260 * ira-color (allocno_hard_regs_compare): Likewise.
3262 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
3265 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
3266 Use .pushsection/.popsection.
3268 2018-01-14 Martin Sebor <msebor@redhat.com>
3271 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
3273 2018-01-14 Jakub Jelinek <jakub@redhat.com>
3275 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
3276 entry from extra_headers.
3277 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
3278 extra_headers, make the list bitwise identical to the i?86-*-* one.
3280 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3282 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
3283 -mcmodel=large with -mindirect-branch=thunk,
3284 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
3285 -mfunction-return=thunk-extern.
3286 * doc/invoke.texi: Document -mcmodel=large is incompatible with
3287 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
3288 -mfunction-return=thunk and -mfunction-return=thunk-extern.
3290 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3292 * config/i386/i386.c (print_reg): Print the name of the full
3293 integer register without '%'.
3294 (ix86_print_operand): Handle 'V'.
3295 * doc/extend.texi: Document 'V' modifier.
3297 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3299 * config/i386/constraints.md (Bs): Disallow memory operand for
3300 -mindirect-branch-register.
3302 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3303 (GOT_memory_operand): Likewise.
3304 (call_insn_operand): Likewise.
3305 (sibcall_insn_operand): Likewise.
3306 (GOT32_symbol_operand): Likewise.
3307 * config/i386/i386.md (indirect_jump): Call convert_memory_address
3308 for -mindirect-branch-register.
3309 (tablejump): Likewise.
3310 (*sibcall_memory): Likewise.
3311 (*sibcall_value_memory): Likewise.
3312 Disallow peepholes of indirect call and jump via memory for
3313 -mindirect-branch-register.
3314 (*call_pop): Replace m with Bw.
3315 (*call_value_pop): Likewise.
3316 (*sibcall_pop_memory): Replace m with Bs.
3317 * config/i386/i386.opt (mindirect-branch-register): New option.
3318 * doc/invoke.texi: Document -mindirect-branch-register option.
3320 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3322 * config/i386/i386-protos.h (ix86_output_function_return): New.
3323 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
3324 set function_return_type.
3325 (indirect_thunk_name): Add ret_p to indicate thunk for function
3327 (output_indirect_thunk_function): Pass false to
3328 indirect_thunk_name.
3329 (ix86_output_indirect_branch_via_reg): Likewise.
3330 (ix86_output_indirect_branch_via_push): Likewise.
3331 (output_indirect_thunk_function): Create alias for function
3332 return thunk if regno < 0.
3333 (ix86_output_function_return): New function.
3334 (ix86_handle_fndecl_attribute): Handle function_return.
3335 (ix86_attribute_table): Add function_return.
3336 * config/i386/i386.h (machine_function): Add
3337 function_return_type.
3338 * config/i386/i386.md (simple_return_internal): Use
3339 ix86_output_function_return.
3340 (simple_return_internal_long): Likewise.
3341 * config/i386/i386.opt (mfunction-return=): New option.
3342 (indirect_branch): Mention -mfunction-return=.
3343 * doc/extend.texi: Document function_return function attribute.
3344 * doc/invoke.texi: Document -mfunction-return= option.
3346 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3348 * config/i386/i386-opts.h (indirect_branch): New.
3349 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
3350 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
3351 with local indirect jump when converting indirect call and jump.
3352 (ix86_set_indirect_branch_type): New.
3353 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
3354 (indirectlabelno): New.
3355 (indirect_thunk_needed): Likewise.
3356 (indirect_thunk_bnd_needed): Likewise.
3357 (indirect_thunks_used): Likewise.
3358 (indirect_thunks_bnd_used): Likewise.
3359 (INDIRECT_LABEL): Likewise.
3360 (indirect_thunk_name): Likewise.
3361 (output_indirect_thunk): Likewise.
3362 (output_indirect_thunk_function): Likewise.
3363 (ix86_output_indirect_branch_via_reg): Likewise.
3364 (ix86_output_indirect_branch_via_push): Likewise.
3365 (ix86_output_indirect_branch): Likewise.
3366 (ix86_output_indirect_jmp): Likewise.
3367 (ix86_code_end): Call output_indirect_thunk_function if needed.
3368 (ix86_output_call_insn): Call ix86_output_indirect_branch if
3370 (ix86_handle_fndecl_attribute): Handle indirect_branch.
3371 (ix86_attribute_table): Add indirect_branch.
3372 * config/i386/i386.h (machine_function): Add indirect_branch_type
3373 and has_local_indirect_jump.
3374 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
3376 (tablejump): Likewise.
3377 (*indirect_jump): Use ix86_output_indirect_jmp.
3378 (*tablejump_1): Likewise.
3379 (simple_return_indirect_internal): Likewise.
3380 * config/i386/i386.opt (mindirect-branch=): New option.
3381 (indirect_branch): New.
3384 (thunk-inline): Likewise.
3385 (thunk-extern): Likewise.
3386 * doc/extend.texi: Document indirect_branch function attribute.
3387 * doc/invoke.texi: Document -mindirect-branch= option.
3389 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
3392 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
3394 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
3396 * ipa-inline.c (want_inline_small_function_p): Return false if
3397 inlining has already failed with CIF_FINAL_ERROR.
3398 (update_caller_keys): Call want_inline_small_function_p before
3400 (update_callee_keys): Likewise.
3402 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
3404 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
3406 (rs6000_quadword_masked_address_p): Likewise.
3407 (quad_aligned_load_p): Likewise.
3408 (quad_aligned_store_p): Likewise.
3409 (const_load_sequence_p): Add comment to describe the outer-most loop.
3410 (mimic_memory_attributes_and_flags): New function.
3411 (rs6000_gen_stvx): Likewise.
3412 (replace_swapped_aligned_store): Likewise.
3413 (rs6000_gen_lvx): Likewise.
3414 (replace_swapped_aligned_load): Likewise.
3415 (replace_swapped_load_constant): Capitalize argument name in
3416 comment describing this function.
3417 (rs6000_analyze_swaps): Add a third pass to search for vector loads
3418 and stores that access quad-word aligned addresses and replace
3419 with stvx or lvx instructions when appropriate.
3420 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
3421 New function prototype.
3422 (rs6000_quadword_masked_address_p): Likewise.
3423 (rs6000_gen_lvx): Likewise.
3424 (rs6000_gen_stvx): Likewise.
3425 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
3426 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3427 when memory address is aligned.
3428 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3429 this split to select lvx instruction when memory address is aligned.
3430 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3431 instruction when memory address is aligned.
3432 (*vsx_le_perm_load_v16qi): Likewise.
3433 (four unnamed splitters): Modify to select the stvx instruction
3434 when memory is aligned.
3436 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3438 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3439 which appears in the queue multiple times.
3441 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3442 Alan Hayward <alan.hayward@arm.com>
3443 David Sherwood <david.sherwood@arm.com>
3445 * tree-vectorizer.h (vec_lower_bound): New structure.
3446 (_loop_vec_info): Add check_nonzero and lower_bounds.
3447 (LOOP_VINFO_CHECK_NONZERO): New macro.
3448 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3449 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3450 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3451 fields. Make seg_len the distance travelled, not including the
3453 (dr_direction_indicator): Declare.
3454 (dr_zero_step_indicator): Likewise.
3455 (dr_known_forward_stride_p): Likewise.
3456 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3458 (runtime_alias_check_p): Allow runtime alias checks with
3460 (operator ==): Compare access_size and align.
3461 (prune_runtime_alias_test_list): Rework for new distinction between
3462 the access_size and seg_len.
3463 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3465 (get_segment_min_max): New function.
3466 (create_intersect_range_checks): Use it.
3467 (dr_step_indicator): New function.
3468 (dr_direction_indicator): Likewise.
3469 (dr_zero_step_indicator): Likewise.
3470 (dr_known_forward_stride_p): Likewise.
3471 * tree-loop-distribution.c (data_ref_segment_size): Return
3472 DR_STEP * (niters - 1).
3473 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3475 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3476 (vect_preserves_scalar_order_p): New function, split out from...
3477 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3478 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3479 (vect_vfa_access_size): New function.
3480 (vect_vfa_align): Likewise.
3481 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3482 (dump_lower_bound): New function.
3483 (vect_check_lower_bound): Likewise.
3484 (vect_small_gap_p): Likewise.
3485 (vectorizable_with_step_bound_p): Likewise.
3486 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3487 depencies if the vectorization factor is 1. Convert the checks
3488 for nonzero steps into checks on the bounds of DR_STEP. Try using
3489 a bunds check for variable steps if the minimum required step is
3490 relatively small. Update calls to the dr_with_seg_len
3491 constructor and to vect_compile_time_alias.
3492 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3494 (vect_loop_versioning): Call it.
3495 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3497 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3499 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3500 Alan Hayward <alan.hayward@arm.com>
3501 David Sherwood <david.sherwood@arm.com>
3503 * doc/sourcebuild.texi (vect_scatter_store): Document.
3504 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3506 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3508 * genopinit.c (main): Add supports_vec_scatter_store and
3509 supports_vec_scatter_store_cached to target_optabs.
3510 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3511 IFN_MASK_SCATTER_STORE.
3512 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3514 * internal-fn.h (internal_store_fn_p): Declare.
3515 (internal_fn_stored_value_index): Likewise.
3516 * internal-fn.c (scatter_store_direct): New macro.
3517 (expand_scatter_store_optab_fn): New function.
3518 (direct_scatter_store_optab_supported_p): New macro.
3519 (internal_store_fn_p): New function.
3520 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3521 IFN_MASK_SCATTER_STORE.
3522 (internal_fn_mask_index): Likewise.
3523 (internal_fn_stored_value_index): New function.
3524 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3526 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3527 * optabs-query.c (supports_vec_scatter_store_p): New function.
3528 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3529 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3530 true for scatter stores.
3531 (vect_gather_scatter_fn_p): Handle scatter stores too.
3532 (vect_check_gather_scatter): Consider using scatter stores if
3533 supports_vec_scatter_store_p.
3534 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3536 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3537 internal_fn_stored_value_index.
3538 (check_load_store_masking): Handle scatter stores too.
3539 (vect_get_store_rhs): Make public.
3540 (vectorizable_call): Use internal_store_fn_p.
3541 (vectorizable_store): Handle scatter store internal functions.
3542 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3543 when deciding whether the end of the group has been reached.
3544 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3545 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3546 (mask_scatter_store<mode>): New insns.
3548 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3549 Alan Hayward <alan.hayward@arm.com>
3550 David Sherwood <david.sherwood@arm.com>
3552 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3553 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3554 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3556 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3557 Use vect_truncate_gather_scatter_offset if we can't treat the
3558 operation as a normal gather load or scatter store.
3559 (get_group_load_store_type): Take the gather_scatter_info
3560 as argument. Try using a gather load or scatter store for
3561 single-element groups.
3562 (get_load_store_type): Update calls to get_group_load_store_type
3563 and vect_use_strided_gather_scatters_p.
3565 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3566 Alan Hayward <alan.hayward@arm.com>
3567 David Sherwood <david.sherwood@arm.com>
3569 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3570 optional tree argument.
3571 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3573 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3574 but continue to use the current value as a fallback.
3575 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3576 to compare the updates.
3577 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3578 (get_load_store_type): Use it when handling a strided access.
3579 (vect_get_strided_load_store_ops): New function.
3580 (vect_get_data_ptr_increment): Likewise.
3581 (vectorizable_load): Handle strided gather loads. Always pass
3582 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3584 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3585 Alan Hayward <alan.hayward@arm.com>
3586 David Sherwood <david.sherwood@arm.com>
3588 * doc/md.texi (gather_load@var{m}): Document.
3589 (mask_gather_load@var{m}): Likewise.
3590 * genopinit.c (main): Add supports_vec_gather_load and
3591 supports_vec_gather_load_cached to target_optabs.
3592 * optabs-tree.c (init_tree_optimization_optabs): Use
3593 ggc_cleared_alloc to allocate target_optabs.
3594 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3595 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3597 * internal-fn.h (internal_load_fn_p): Declare.
3598 (internal_gather_scatter_fn_p): Likewise.
3599 (internal_fn_mask_index): Likewise.
3600 (internal_gather_scatter_fn_supported_p): Likewise.
3601 * internal-fn.c (gather_load_direct): New macro.
3602 (expand_gather_load_optab_fn): New function.
3603 (direct_gather_load_optab_supported_p): New macro.
3604 (direct_internal_fn_optab): New function.
3605 (internal_load_fn_p): Likewise.
3606 (internal_gather_scatter_fn_p): Likewise.
3607 (internal_fn_mask_index): Likewise.
3608 (internal_gather_scatter_fn_supported_p): Likewise.
3609 * optabs-query.c (supports_at_least_one_mode_p): New function.
3610 (supports_vec_gather_load_p): Likewise.
3611 * optabs-query.h (supports_vec_gather_load_p): Declare.
3612 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3613 and memory_type field.
3614 (NUM_PATTERNS): Bump to 15.
3615 * tree-vect-data-refs.c: Include internal-fn.h.
3616 (vect_gather_scatter_fn_p): New function.
3617 (vect_describe_gather_scatter_call): Likewise.
3618 (vect_check_gather_scatter): Try using internal functions for
3619 gather loads. Recognize existing calls to a gather load function.
3620 (vect_analyze_data_refs): Consider using gather loads if
3621 supports_vec_gather_load_p.
3622 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3623 (vect_get_gather_scatter_offset_type): Likewise.
3624 (vect_convert_mask_for_vectype): Likewise.
3625 (vect_add_conversion_to_patterm): Likewise.
3626 (vect_try_gather_scatter_pattern): Likewise.
3627 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3628 (vect_vect_recog_func_ptrs): Add it.
3629 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3630 internal_fn_mask_index and internal_gather_scatter_fn_p.
3631 (check_load_store_masking): Take the gather_scatter_info as an
3632 argument and handle gather loads.
3633 (vect_get_gather_scatter_ops): New function.
3634 (vectorizable_call): Check internal_load_fn_p.
3635 (vectorizable_load): Likewise. Handle gather load internal
3637 (vectorizable_store): Update call to check_load_store_masking.
3638 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3639 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3640 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3641 (aarch64_gather_scale_operand_d): New predicates.
3642 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3643 (mask_gather_load<mode>): New insns.
3645 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3646 Alan Hayward <alan.hayward@arm.com>
3647 David Sherwood <david.sherwood@arm.com>
3649 * optabs.def (fold_left_plus_optab): New optab.
3650 * doc/md.texi (fold_left_plus_@var{m}): Document.
3651 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3652 * internal-fn.c (fold_left_direct): Define.
3653 (expand_fold_left_optab_fn): Likewise.
3654 (direct_fold_left_optab_supported_p): Likewise.
3655 * fold-const-call.c (fold_const_fold_left): New function.
3656 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3657 * tree-parloops.c (valid_reduction_p): New function.
3658 (gather_scalar_reductions): Use it.
3659 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3660 (vect_finish_replace_stmt): Declare.
3661 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3662 (needs_fold_left_reduction_p): New function, split out from...
3663 (vect_is_simple_reduction): ...here. Accept reductions that
3664 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3665 (vect_force_simple_reduction): Also store the reduction type in
3666 the assignment's STMT_VINFO_REDUC_TYPE.
3667 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3668 (merge_with_identity): New function.
3669 (vect_expand_fold_left): Likewise.
3670 (vectorize_fold_left_reduction): Likewise.
3671 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3672 scalar phi in place for it. Check for target support and reject
3673 cases that would reassociate the operation. Defer the transform
3674 phase to vectorize_fold_left_reduction.
3675 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3676 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3677 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3679 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3681 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3682 call to ifc_temp_var.
3684 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3685 Alan Hayward <alan.hayward@arm.com>
3686 David Sherwood <david.sherwood@arm.com>
3688 * target.def (legitimize_address_displacement): Take the original
3689 offset as a poly_int.
3690 * targhooks.h (default_legitimize_address_displacement): Update
3692 * targhooks.c (default_legitimize_address_displacement): Likewise.
3693 * doc/tm.texi: Regenerate.
3694 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3695 as an argument, moving assert of ad->disp == ad->disp_term to...
3696 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3697 Try calling targetm.legitimize_address_displacement before expanding
3698 the address rather than afterwards, and adjust for the new interface.
3699 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3700 Match the new hook interface. Handle SVE addresses.
3701 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3704 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3706 * Makefile.in (OBJS): Add early-remat.o.
3707 * target.def (select_early_remat_modes): New hook.
3708 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3709 * doc/tm.texi: Regenerate.
3710 * targhooks.h (default_select_early_remat_modes): Declare.
3711 * targhooks.c (default_select_early_remat_modes): New function.
3712 * timevar.def (TV_EARLY_REMAT): New timevar.
3713 * passes.def (pass_early_remat): New pass.
3714 * tree-pass.h (make_pass_early_remat): Declare.
3715 * early-remat.c: New file.
3716 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3718 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3720 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3721 Alan Hayward <alan.hayward@arm.com>
3722 David Sherwood <david.sherwood@arm.com>
3724 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3725 vfm1 with a bound_epilog parameter.
3726 (vect_do_peeling): Update calls accordingly, and move the prologue
3727 call earlier in the function. Treat the base bound_epilog as 0 for
3728 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3729 this base when peeling for gaps.
3730 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3731 with fully-masked loops.
3732 (vect_estimate_min_profitable_iters): Handle the single peeled
3733 iteration in that case.
3735 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3736 Alan Hayward <alan.hayward@arm.com>
3737 David Sherwood <david.sherwood@arm.com>
3739 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3740 single-element interleaving even if the size is not a power of 2.
3741 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3742 accesses for single-element interleaving if the group size is
3745 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3746 Alan Hayward <alan.hayward@arm.com>
3747 David Sherwood <david.sherwood@arm.com>
3749 * doc/md.texi (fold_extract_last_@var{m}): Document.
3750 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3751 * optabs.def (fold_extract_last_optab): New optab.
3752 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3753 * internal-fn.c (fold_extract_direct): New macro.
3754 (expand_fold_extract_optab_fn): Likewise.
3755 (direct_fold_extract_optab_supported_p): Likewise.
3756 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3757 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3758 EXTRACT_LAST_REDUCTION.
3759 (get_initial_def_for_reduction): Do not create an initial vector
3760 for EXTRACT_LAST_REDUCTION reductions.
3761 (vectorizable_reduction): Leave the scalar phi in place for
3762 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3763 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3764 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3765 transform phase to vectorizable_condition.
3766 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3768 (vect_finish_stmt_generation): ...here.
3769 (vect_finish_replace_stmt): New function.
3770 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3771 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3773 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3775 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3776 Alan Hayward <alan.hayward@arm.com>
3777 David Sherwood <david.sherwood@arm.com>
3779 * doc/md.texi (extract_last_@var{m}): Document.
3780 * optabs.def (extract_last_optab): New optab.
3781 * internal-fn.def (EXTRACT_LAST): New internal function.
3782 * internal-fn.c (cond_unary_direct): New macro.
3783 (expand_cond_unary_optab_fn): Likewise.
3784 (direct_cond_unary_optab_supported_p): Likewise.
3785 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3786 loops using EXTRACT_LAST.
3787 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3788 (extract_last_<mode>): ...this optab.
3789 (vec_extract<mode><Vel>): Update accordingly.
3791 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3792 Alan Hayward <alan.hayward@arm.com>
3793 David Sherwood <david.sherwood@arm.com>
3795 * target.def (empty_mask_is_expensive): New hook.
3796 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3797 * doc/tm.texi: Regenerate.
3798 * targhooks.h (default_empty_mask_is_expensive): Declare.
3799 * targhooks.c (default_empty_mask_is_expensive): New function.
3800 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3801 if the target says that empty masks are expensive.
3802 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3804 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3806 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3807 Alan Hayward <alan.hayward@arm.com>
3808 David Sherwood <david.sherwood@arm.com>
3810 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3811 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3812 (vect_use_loop_mask_for_alignment_p): New function.
3813 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3814 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3815 niters_skip argument. Make sure that the first niters_skip elements
3816 of the first iteration are inactive.
3817 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3818 Update call to vect_set_loop_masks_directly.
3819 (get_misalign_in_elems): New function, split out from...
3820 (vect_gen_prolog_loop_niters): ...here.
3821 (vect_update_init_of_dr): Take a code argument that specifies whether
3822 the adjustment should be added or subtracted.
3823 (vect_update_init_of_drs): Likewise.
3824 (vect_prepare_for_masked_peels): New function.
3825 (vect_do_peeling): Skip prologue peeling if we're using a mask
3826 instead. Update call to vect_update_inits_of_drs.
3827 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3829 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3830 alignment. Do not include the number of peeled iterations in
3831 the minimum threshold in that case.
3832 (vectorizable_induction): Adjust the start value down by
3833 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3834 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3835 Take the number of skipped iterations into account when calculating
3837 * tree-vect-stmts.c (vect_gen_while_not): New function.
3839 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3840 Alan Hayward <alan.hayward@arm.com>
3841 David Sherwood <david.sherwood@arm.com>
3843 * doc/sourcebuild.texi (vect_fully_masked): Document.
3844 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3846 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3848 (vect_analyze_loop_2): ...here. Don't check the vectorization
3849 factor against the number of loop iterations if the loop is
3852 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3853 Alan Hayward <alan.hayward@arm.com>
3854 David Sherwood <david.sherwood@arm.com>
3856 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3857 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3858 (dump_groups): Update accordingly.
3859 (iv_use::mem_type): New member variable.
3860 (address_p): New function.
3861 (record_use): Add a mem_type argument and initialize the new
3863 (record_group_use): Add a mem_type argument. Use address_p.
3864 Remove obsolete null checks of base_object. Update call to record_use.
3865 (find_interesting_uses_op): Update call to record_group_use.
3866 (find_interesting_uses_cond): Likewise.
3867 (find_interesting_uses_address): Likewise.
3868 (get_mem_type_for_internal_fn): New function.
3869 (find_address_like_use): Likewise.
3870 (find_interesting_uses_stmt): Try find_address_like_use before
3871 calling find_interesting_uses_op.
3872 (addr_offset_valid_p): Use the iv mem_type field as the type
3873 of the addressed memory.
3874 (add_autoinc_candidates): Likewise.
3875 (get_address_cost): Likewise.
3876 (split_small_address_groups_p): Use address_p.
3877 (split_address_groups): Likewise.
3878 (add_iv_candidate_for_use): Likewise.
3879 (autoinc_possible_for_pair): Likewise.
3880 (rewrite_groups): Likewise.
3881 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3882 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3883 (get_alias_ptr_type_for_ptr_address): New function.
3884 (rewrite_use_address): Rewrite address uses in calls that were
3885 identified by find_address_like_use.
3887 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3888 Alan Hayward <alan.hayward@arm.com>
3889 David Sherwood <david.sherwood@arm.com>
3891 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3893 * gimple-expr.h (is_gimple_addressable: Likewise.
3894 * gimple-expr.c (is_gimple_address): Likewise.
3895 * internal-fn.c (expand_call_mem_ref): New function.
3896 (expand_mask_load_optab_fn): Use it.
3897 (expand_mask_store_optab_fn): Likewise.
3899 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3900 Alan Hayward <alan.hayward@arm.com>
3901 David Sherwood <david.sherwood@arm.com>
3903 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3904 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3905 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3906 (cond_umax@var{mode}): Document.
3907 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3908 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3909 (cond_umin_optab, cond_umax_optab): New optabs.
3910 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3911 (COND_IOR, COND_XOR): New internal functions.
3912 * internal-fn.h (get_conditional_internal_fn): Declare.
3913 * internal-fn.c (cond_binary_direct): New macro.
3914 (expand_cond_binary_optab_fn): Likewise.
3915 (direct_cond_binary_optab_supported_p): Likewise.
3916 (get_conditional_internal_fn): New function.
3917 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3918 Cope with reduction statements that are vectorized as calls rather
3920 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3921 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3922 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3923 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3924 (UNSPEC_COND_EOR): New unspecs.
3925 (optab): Add mappings for them.
3926 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3927 (sve_int_op, sve_fp_op): New int attributes.
3929 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3930 Alan Hayward <alan.hayward@arm.com>
3931 David Sherwood <david.sherwood@arm.com>
3933 * optabs.def (while_ult_optab): New optab.
3934 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3935 * internal-fn.def (WHILE_ULT): New internal function.
3936 * internal-fn.h (direct_internal_fn_supported_p): New override
3937 that takes two types as argument.
3938 * internal-fn.c (while_direct): New macro.
3939 (expand_while_optab_fn): New function.
3940 (convert_optab_supported_p): Likewise.
3941 (direct_while_optab_supported_p): New macro.
3942 * wide-int.h (wi::udiv_ceil): New function.
3943 * tree-vectorizer.h (rgroup_masks): New structure.
3944 (vec_loop_masks): New typedef.
3945 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3947 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3948 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3949 (vect_max_vf): New function.
3950 (slpeel_make_loop_iterate_ntimes): Delete.
3951 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3952 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3953 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3954 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3955 internal-fn.h, stor-layout.h and optabs-query.h.
3956 (vect_set_loop_mask): New function.
3957 (add_preheader_seq): Likewise.
3958 (add_header_seq): Likewise.
3959 (interleave_supported_p): Likewise.
3960 (vect_maybe_permute_loop_masks): Likewise.
3961 (vect_set_loop_masks_directly): Likewise.
3962 (vect_set_loop_condition_masked): Likewise.
3963 (vect_set_loop_condition_unmasked): New function, split out from
3964 slpeel_make_loop_iterate_ntimes.
3965 (slpeel_make_loop_iterate_ntimes): Rename to..
3966 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3967 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3968 (vect_do_peeling): Update call accordingly.
3969 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3971 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3972 mask_compare_type, can_fully_mask_p and fully_masked_p.
3973 (release_vec_loop_masks): New function.
3974 (_loop_vec_info): Use it to free the loop masks.
3975 (can_produce_all_loop_masks_p): New function.
3976 (vect_get_max_nscalars_per_iter): Likewise.
3977 (vect_verify_full_masking): Likewise.
3978 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3979 retries, and free the mask rgroups before retrying. Check loop-wide
3980 reasons for disallowing fully-masked loops. Make the final decision
3981 about whether use a fully-masked loop or not.
3982 (vect_estimate_min_profitable_iters): Do not assume that peeling
3983 for the number of iterations will be needed for fully-masked loops.
3984 (vectorizable_reduction): Disable fully-masked loops.
3985 (vectorizable_live_operation): Likewise.
3986 (vect_halve_mask_nunits): New function.
3987 (vect_double_mask_nunits): Likewise.
3988 (vect_record_loop_mask): Likewise.
3989 (vect_get_loop_mask): Likewise.
3990 (vect_transform_loop): Handle the case in which the final loop
3991 iteration might handle a partial vector. Call vect_set_loop_condition
3992 instead of slpeel_make_loop_iterate_ntimes.
3993 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3994 (check_load_store_masking): New function.
3995 (prepare_load_store_mask): Likewise.
3996 (vectorizable_store): Handle fully-masked loops.
3997 (vectorizable_load): Likewise.
3998 (supportable_widening_operation): Use vect_halve_mask_nunits for
4000 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
4001 (vect_gen_while): New function.
4002 * config/aarch64/aarch64.md (umax<mode>3): New expander.
4003 (aarch64_uqdec<mode>): New insn.
4005 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4006 Alan Hayward <alan.hayward@arm.com>
4007 David Sherwood <david.sherwood@arm.com>
4009 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
4010 (reduc_xor_scal_optab): New optabs.
4011 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
4012 (reduc_xor_scal_@var{m}): Document.
4013 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
4014 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
4016 * fold-const-call.c (fold_const_call): Handle them.
4017 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
4018 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
4019 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
4020 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
4021 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
4022 (UNSPEC_XORV): New unspecs.
4023 (optab): Add entries for them.
4024 (BITWISEV): New int iterator.
4025 (bit_reduc_op): New int attributes.
4027 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4028 Alan Hayward <alan.hayward@arm.com>
4029 David Sherwood <david.sherwood@arm.com>
4031 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
4032 * internal-fn.def (VEC_SHL_INSERT): New internal function.
4033 * optabs.def (vec_shl_insert_optab): New optab.
4034 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
4035 (duplicate_and_interleave): Likewise.
4036 * tree-vect-loop.c: Include internal-fn.h.
4037 (neutral_op_for_slp_reduction): New function, split out from
4038 get_initial_defs_for_reduction.
4039 (get_initial_def_for_reduction): Handle option 2 for variable-length
4040 vectors by loading the neutral value into a vector and then shifting
4041 the initial value into element 0.
4042 (get_initial_defs_for_reduction): Replace the code argument with
4043 the neutral value calculated by neutral_op_for_slp_reduction.
4044 Use gimple_build_vector for constant-length vectors.
4045 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
4046 but the first group_size elements have a neutral value.
4047 Use duplicate_and_interleave otherwise.
4048 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
4049 Update call to get_initial_defs_for_reduction. Handle SLP
4050 reductions for variable-length vectors by creating one vector
4051 result for each scalar result, with the elements associated
4052 with other scalar results stubbed out with the neutral value.
4053 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
4054 Require IFN_VEC_SHL_INSERT for double reductions on
4055 variable-length vectors, or SLP reductions that have
4056 a neutral value. Require can_duplicate_and_interleave_p
4057 support for variable-length unchained SLP reductions if there
4058 is no neutral value, such as for MIN/MAX reductions. Also require
4059 the number of vector elements to be a multiple of the number of
4060 SLP statements when doing variable-length unchained SLP reductions.
4061 Update call to vect_create_epilog_for_reduction.
4062 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
4063 and remove initial values.
4064 (duplicate_and_interleave): Make public.
4065 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
4066 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
4068 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4069 Alan Hayward <alan.hayward@arm.com>
4070 David Sherwood <david.sherwood@arm.com>
4072 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
4073 (can_duplicate_and_interleave_p): New function.
4074 (vect_get_and_check_slp_defs): Take the vector of statements
4075 rather than just the current one. Remove excess parentheses.
4076 Restriction rejectinon of vect_constant_def and vect_external_def
4077 for variable-length vectors to boolean types, or types for which
4078 can_duplicate_and_interleave_p is false.
4079 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
4080 (duplicate_and_interleave): New function.
4081 (vect_get_constant_vectors): Use gimple_build_vector for
4082 constant-length vectors and suitable variable-length constant
4083 vectors. Use duplicate_and_interleave for other variable-length
4084 vectors. Don't defer the update when inserting new statements.
4086 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4087 Alan Hayward <alan.hayward@arm.com>
4088 David Sherwood <david.sherwood@arm.com>
4090 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
4091 min_profitable_iters doesn't go negative.
4093 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4094 Alan Hayward <alan.hayward@arm.com>
4095 David Sherwood <david.sherwood@arm.com>
4097 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
4098 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
4099 * optabs.def (vec_mask_load_lanes_optab): New optab.
4100 (vec_mask_store_lanes_optab): Likewise.
4101 * internal-fn.def (MASK_LOAD_LANES): New internal function.
4102 (MASK_STORE_LANES): Likewise.
4103 * internal-fn.c (mask_load_lanes_direct): New macro.
4104 (mask_store_lanes_direct): Likewise.
4105 (expand_mask_load_optab_fn): Handle masked operations.
4106 (expand_mask_load_lanes_optab_fn): New macro.
4107 (expand_mask_store_optab_fn): Handle masked operations.
4108 (expand_mask_store_lanes_optab_fn): New macro.
4109 (direct_mask_load_lanes_optab_supported_p): Likewise.
4110 (direct_mask_store_lanes_optab_supported_p): Likewise.
4111 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
4113 (vect_load_lanes_supported): Likewise.
4114 * tree-vect-data-refs.c (strip_conversion): New function.
4115 (can_group_stmts_p): Likewise.
4116 (vect_analyze_data_ref_accesses): Use it instead of checking
4117 for a pair of assignments.
4118 (vect_store_lanes_supported): Take a masked_p parameter.
4119 (vect_load_lanes_supported): Likewise.
4120 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
4121 vect_store_lanes_supported and vect_load_lanes_supported.
4122 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
4123 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
4124 parameter. Don't allow gaps for masked accesses.
4125 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
4126 and vect_load_lanes_supported.
4127 (get_load_store_type): Take a masked_p parameter and update
4128 call to get_group_load_store_type.
4129 (vectorizable_store): Update call to get_load_store_type.
4130 Handle IFN_MASK_STORE_LANES.
4131 (vectorizable_load): Update call to get_load_store_type.
4132 Handle IFN_MASK_LOAD_LANES.
4134 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4135 Alan Hayward <alan.hayward@arm.com>
4136 David Sherwood <david.sherwood@arm.com>
4138 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
4140 * config/aarch64/aarch64-protos.h
4141 (aarch64_sve_struct_memory_operand_p): Declare.
4142 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
4143 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
4144 (VPRED, vpred): Handle SVE structure modes.
4145 * config/aarch64/constraints.md (Utx): New constraint.
4146 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
4147 (aarch64_sve_struct_nonimmediate_operand): New predicates.
4148 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
4149 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
4150 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
4151 structure modes. Split into pieces after RA.
4152 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
4153 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
4155 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
4156 SVE structure modes.
4157 (aarch64_classify_address): Likewise.
4158 (sizetochar): Move earlier in file.
4159 (aarch64_print_operand): Handle SVE register lists.
4160 (aarch64_array_mode): New function.
4161 (aarch64_sve_struct_memory_operand_p): Likewise.
4162 (TARGET_ARRAY_MODE): Redefine.
4164 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4165 Alan Hayward <alan.hayward@arm.com>
4166 David Sherwood <david.sherwood@arm.com>
4168 * target.def (array_mode): New target hook.
4169 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
4170 * doc/tm.texi: Regenerate.
4171 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
4172 * hooks.c (hook_optmode_mode_uhwi_none): New function.
4173 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
4175 * stor-layout.c (mode_for_array): Likewise. Support polynomial
4178 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4179 Alan Hayward <alan.hayward@arm.com>
4180 David Sherwood <david.sherwood@arm.com>
4182 * fold-const.c (fold_binary_loc): Check the argument types
4183 rather than the result type when testing for a vector operation.
4185 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4187 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
4188 * doc/tm.texi: Regenerate.
4190 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4191 Alan Hayward <alan.hayward@arm.com>
4192 David Sherwood <david.sherwood@arm.com>
4194 * doc/invoke.texi (-msve-vector-bits=): Document new option.
4195 (sve): Document new AArch64 extension.
4196 * doc/md.texi (w): Extend the description of the AArch64
4197 constraint to include SVE vectors.
4198 (Upl, Upa): Document new AArch64 predicate constraints.
4199 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
4201 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
4202 (msve-vector-bits=): New option.
4203 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
4204 SVE when these are disabled.
4205 (sve): New extension.
4206 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
4207 modes. Adjust their number of units based on aarch64_sve_vg.
4208 (MAX_BITSIZE_MODE_ANY_MODE): Define.
4209 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
4210 aarch64_addr_query_type.
4211 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
4212 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
4213 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
4214 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
4215 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
4216 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
4217 (aarch64_simd_imm_zero_p): Delete.
4218 (aarch64_check_zero_based_sve_index_immediate): Declare.
4219 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4220 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4221 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4222 (aarch64_sve_float_mul_immediate_p): Likewise.
4223 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4225 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
4226 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
4227 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
4228 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
4229 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
4230 (aarch64_regmode_natural_size): Likewise.
4231 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
4232 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
4234 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
4235 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
4236 for VG and the SVE predicate registers.
4237 (V_ALIASES): Add a "z"-prefixed alias.
4238 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
4239 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
4240 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
4241 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
4242 (REG_CLASS_NAMES): Add entries for them.
4243 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
4244 and the predicate registers.
4245 (aarch64_sve_vg): Declare.
4246 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
4247 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
4248 (REGMODE_NATURAL_SIZE): Define.
4249 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
4251 * config/aarch64/aarch64.c: Include cfgrtl.h.
4252 (simd_immediate_info): Add a constructor for series vectors,
4253 and an associated step field.
4254 (aarch64_sve_vg): New variable.
4255 (aarch64_dbx_register_number): Handle VG and the predicate registers.
4256 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
4257 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
4258 (VEC_ANY_DATA, VEC_STRUCT): New constants.
4259 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
4260 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
4261 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
4262 (aarch64_get_mask_mode): New functions.
4263 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
4264 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
4265 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
4266 predicate modes and predicate registers. Explicitly restrict
4267 GPRs to modes of 16 bytes or smaller. Only allow FP registers
4268 to store a vector mode if it is recognized by
4269 aarch64_classify_vector_mode.
4270 (aarch64_regmode_natural_size): New function.
4271 (aarch64_hard_regno_caller_save_mode): Return the original mode
4273 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
4274 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
4275 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
4276 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
4278 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
4279 does not overlap dest if the function is frame-related. Handle
4281 (aarch64_split_add_offset): New function.
4282 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
4283 them aarch64_add_offset.
4284 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
4285 and update call to aarch64_sub_sp.
4286 (aarch64_add_cfa_expression): New function.
4287 (aarch64_expand_prologue): Pass extra temporary registers to the
4288 functions above. Handle the case in which we need to emit new
4289 DW_CFA_expressions for registers that were originally saved
4290 relative to the stack pointer, but now have to be expressed
4291 relative to the frame pointer.
4292 (aarch64_output_mi_thunk): Pass extra temporary registers to the
4294 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
4295 IP0 and IP1 values for SVE frames.
4296 (aarch64_expand_vec_series): New function.
4297 (aarch64_expand_sve_widened_duplicate): Likewise.
4298 (aarch64_expand_sve_const_vector): Likewise.
4299 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
4300 Handle SVE constants. Use emit_move_insn to move a force_const_mem
4301 into the register, rather than emitting a SET directly.
4302 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
4303 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
4304 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
4305 (offset_9bit_signed_scaled_p): New functions.
4306 (aarch64_replicate_bitmask_imm): New function.
4307 (aarch64_bitmask_imm): Use it.
4308 (aarch64_cannot_force_const_mem): Reject expressions involving
4309 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
4310 (aarch64_classify_index): Handle SVE indices, by requiring
4311 a plain register index with a scale that matches the element size.
4312 (aarch64_classify_address): Handle SVE addresses. Assert that
4313 the mode of the address is VOIDmode or an integer mode.
4314 Update call to aarch64_classify_symbol.
4315 (aarch64_classify_symbolic_expression): Update call to
4316 aarch64_classify_symbol.
4317 (aarch64_const_vec_all_in_range_p): New function.
4318 (aarch64_print_vector_float_operand): Likewise.
4319 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
4320 "vN" for FP registers with SVE modes. Handle (const ...) vectors
4321 and the FP immediates 1.0 and 0.5.
4322 (aarch64_print_address_internal): Handle SVE addresses.
4323 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
4324 (aarch64_regno_regclass): Handle predicate registers.
4325 (aarch64_secondary_reload): Handle big-endian reloads of SVE
4327 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
4328 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
4329 (aarch64_convert_sve_vector_bits): New function.
4330 (aarch64_override_options): Use it to handle -msve-vector-bits=.
4331 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4333 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
4334 Handle SVE vector and predicate modes. Accept VL-based constants
4335 that need only one temporary register, and VL offsets that require
4336 no temporary registers.
4337 (aarch64_conditional_register_usage): Mark the predicate registers
4338 as fixed if SVE isn't available.
4339 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
4340 Return true for SVE vector and predicate modes.
4341 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
4342 rather than an unsigned int. Handle SVE modes.
4343 (aarch64_preferred_simd_mode): Update call accordingly. Handle
4345 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
4347 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4348 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4349 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4350 (aarch64_sve_float_mul_immediate_p): New functions.
4351 (aarch64_sve_valid_immediate): New function.
4352 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
4353 Explicitly reject structure modes. Check for INDEX constants.
4354 Handle PTRUE and PFALSE constants.
4355 (aarch64_check_zero_based_sve_index_immediate): New function.
4356 (aarch64_simd_imm_zero_p): Delete.
4357 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
4358 vector modes. Accept constants in the range of CNT[BHWD].
4359 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
4360 ask for an Advanced SIMD mode.
4361 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
4362 (aarch64_simd_vector_alignment): Handle SVE predicates.
4363 (aarch64_vectorize_preferred_vector_alignment): New function.
4364 (aarch64_simd_vector_alignment_reachable): Use it instead of
4366 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
4367 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
4369 (MAX_VECT_LEN): Delete.
4370 (expand_vec_perm_d): Add a vec_flags field.
4371 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
4372 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
4373 (aarch64_evpc_ext): Don't apply a big-endian lane correction
4375 (aarch64_evpc_rev): Rename to...
4376 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
4377 (aarch64_evpc_rev_global): New function.
4378 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
4379 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
4381 (aarch64_evpc_sve_tbl): New function.
4382 (aarch64_expand_vec_perm_const_1): Update after rename of
4383 aarch64_evpc_rev. Handle SVE permutes too, trying
4384 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
4385 than aarch64_evpc_tbl.
4386 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
4387 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
4388 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
4389 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
4390 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
4391 (aarch64_expand_sve_vcond): New functions.
4392 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
4393 of aarch64_vector_mode_p.
4394 (aarch64_dwarf_poly_indeterminate_value): New function.
4395 (aarch64_compute_pressure_classes): Likewise.
4396 (aarch64_can_change_mode_class): Likewise.
4397 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
4398 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
4399 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
4400 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
4401 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
4402 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
4403 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
4404 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
4406 (Dn, Dl, Dr): Accept const as well as const_vector.
4407 (Dz): Likewise. Compare against CONST0_RTX.
4408 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
4409 of "vector" where appropriate.
4410 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
4411 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
4412 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
4413 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
4414 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
4415 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
4416 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
4417 (v_int_equiv): Extend to SVE modes.
4418 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
4420 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
4421 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
4422 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
4423 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
4424 (SVE_COND_FP_CMP): New int iterators.
4425 (perm_hilo): Handle the new unpack unspecs.
4426 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4428 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4429 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4430 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4431 (aarch64_equality_operator, aarch64_constant_vector_operand)
4432 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4433 (aarch64_sve_nonimmediate_operand): Likewise.
4434 (aarch64_sve_general_operand): Likewise.
4435 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4436 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4437 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4438 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4439 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4440 (aarch64_sve_float_arith_immediate): Likewise.
4441 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4442 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4443 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4444 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4445 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4446 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4447 (aarch64_sve_float_arith_operand): Likewise.
4448 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4449 (aarch64_sve_float_mul_operand): Likewise.
4450 (aarch64_sve_vec_perm_operand): Likewise.
4451 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4452 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4453 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4454 as well as const_vector.
4455 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4456 in file. Use CONST0_RTX and CONSTM1_RTX.
4457 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4458 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4459 Use aarch64_simd_imm_zero.
4460 * config/aarch64/aarch64-sve.md: New file.
4461 * config/aarch64/aarch64.md: Include it.
4462 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4463 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4464 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4465 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4466 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4467 (sve): New attribute.
4468 (enabled): Disable instructions with the sve attribute unless
4470 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4471 aarch64_expand_mov_immediate.
4472 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4473 CNT[BHSD] immediates.
4474 (movti): Split CONST_POLY_INT moves into two halves.
4475 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4476 Split additions that need a temporary here if the destination
4477 is the stack pointer.
4478 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4479 (*add<mode>3_poly_1): New instruction.
4480 (set_clobber_cc): New expander.
4482 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4484 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4485 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4486 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4487 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4488 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4489 Change innermode from fixed_mode_size to machine_mode.
4490 (simplify_subreg): Update call accordingly. Handle a constant-sized
4491 subreg of a variable-length CONST_VECTOR.
4493 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4494 Alan Hayward <alan.hayward@arm.com>
4495 David Sherwood <david.sherwood@arm.com>
4497 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4498 (add_offset_to_base): New function, split out from...
4499 (create_mem_ref): ...here. When handling a scale other than 1,
4500 check first whether the address is valid without the offset.
4501 Add it into the base if so, leaving the index and scale as-is.
4503 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4506 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4507 fold_for_warn before checking if arg2 is INTEGER_CST.
4509 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4511 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4512 (store_multiple_operation): Delete.
4513 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4514 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4515 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4516 guarded by TARGET_STRING.
4517 (rs6000_output_load_multiple): Delete.
4518 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4519 OPTION_MASK_STRING / TARGET_STRING handling.
4520 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4521 (const rs6000_opt_masks) <"string">: Change mask to 0.
4522 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4523 (MASK_STRING): Delete.
4524 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4526 (load_multiple): Delete.
4533 (store_multiple): Delete.
4540 (movmemsi_8reg): Delete.
4541 (corresponding unnamed define_insn): Delete.
4542 (movmemsi_6reg): Delete.
4543 (corresponding unnamed define_insn): Delete.
4544 (movmemsi_4reg): Delete.
4545 (corresponding unnamed define_insn): Delete.
4546 (movmemsi_2reg): Delete.
4547 (corresponding unnamed define_insn): Delete.
4548 (movmemsi_1reg): Delete.
4549 (corresponding unnamed define_insn): Delete.
4550 * config/rs6000/rs6000.opt (mno-string): New.
4551 (mstring): Replace by deprecation warning stub.
4552 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4554 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4556 * regrename.c (regrename_do_replace): If replacing the same
4557 reg multiple times, try to reuse last created gen_raw_REG.
4560 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4561 main to workaround a bug in GDB.
4563 2018-01-12 Tom de Vries <tom@codesourcery.com>
4566 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4568 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4570 PR rtl-optimization/80481
4571 * ira-color.c (get_cap_member): New function.
4572 (allocnos_conflict_by_live_ranges_p): Use it.
4573 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4574 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4576 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4579 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4580 (*saddl_se_1): Ditto.
4582 (*ssubl_se_1): Ditto.
4584 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4586 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4587 rather than wi::to_widest for DR_INITs.
4588 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4589 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4590 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4592 (vect_analyze_group_access_1): Note that here.
4594 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4596 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4597 polynomial type sizes.
4599 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4601 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4602 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4603 (gimple_add_tmp_var): Likewise.
4605 2018-01-12 Martin Liska <mliska@suse.cz>
4607 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4608 (gimple_alloc_sizes): Likewise.
4609 (dump_gimple_statistics): Use PRIu64 in printf format.
4610 * gimple.h: Change uint64_t to int.
4612 2018-01-12 Martin Liska <mliska@suse.cz>
4614 * tree-core.h: Use uint64_t instead of int.
4615 * tree.c (tree_node_counts): Likewise.
4616 (tree_node_sizes): Likewise.
4617 (dump_tree_statistics): Use PRIu64 in printf format.
4619 2018-01-12 Martin Liska <mliska@suse.cz>
4621 * Makefile.in: As qsort_chk is implemented in vec.c, add
4622 vec.o to linkage of gencfn-macros.
4623 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4624 passing the info to record_node_allocation_statistics.
4625 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4627 * ggc-common.c (struct ggc_usage): Add operator== and use
4628 it in operator< and compare function.
4629 * mem-stats.h (struct mem_usage): Likewise.
4630 * vec.c (struct vec_usage): Remove operator< and compare
4631 function. Can be simply inherited.
4633 2018-01-12 Martin Jambor <mjambor@suse.cz>
4636 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4637 * tree-ssa-math-opts.c: Include domwalk.h.
4638 (convert_mult_to_fma_1): New function.
4639 (fma_transformation_info): New type.
4640 (fma_deferring_state): Likewise.
4641 (cancel_fma_deferring): New function.
4642 (result_of_phi): Likewise.
4643 (last_fma_candidate_feeds_initial_phi): Likewise.
4644 (convert_mult_to_fma): Added deferring logic, split actual
4645 transformation to convert_mult_to_fma_1.
4646 (math_opts_dom_walker): New type.
4647 (math_opts_dom_walker::after_dom_children): New method, body moved
4648 here from pass_optimize_widening_mul::execute, added deferring logic
4650 (pass_optimize_widening_mul::execute): Moved most of code to
4651 math_opts_dom_walker::after_dom_children.
4652 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4653 * config/i386/i386.c (ix86_option_override_internal): Added
4654 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4656 2018-01-12 Richard Biener <rguenther@suse.de>
4659 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4660 inline instance vars.
4662 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4665 * config/rx/rx.c (rx_is_restricted_memory_address):
4668 2018-01-12 Richard Biener <rguenther@suse.de>
4670 PR tree-optimization/80846
4671 * target.def (split_reduction): New target hook.
4672 * targhooks.c (default_split_reduction): New function.
4673 * targhooks.h (default_split_reduction): Declare.
4674 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4675 target requests first reduce vectors by combining low and high
4677 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4678 (get_vectype_for_scalar_type_and_size): Export.
4679 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4680 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4681 * doc/tm.texi: Regenerate.
4682 * config/i386/i386.c (ix86_split_reduction): Implement
4683 TARGET_VECTORIZE_SPLIT_REDUCTION.
4685 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4688 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4689 in PIC mode except for TARGET_VXWORKS_RTP.
4690 * config/sparc/sparc.c: Include cfgrtl.h.
4691 (TARGET_INIT_PIC_REG): Define.
4692 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4693 (sparc_pic_register_p): New predicate.
4694 (sparc_legitimate_address_p): Use it.
4695 (sparc_legitimize_pic_address): Likewise.
4696 (sparc_delegitimize_address): Likewise.
4697 (sparc_mode_dependent_address_p): Likewise.
4698 (gen_load_pcrel_sym): Remove 4th parameter.
4699 (load_got_register): Adjust call to above. Remove obsolete stuff.
4700 (sparc_expand_prologue): Do not call load_got_register here.
4701 (sparc_flat_expand_prologue): Likewise.
4702 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4703 (sparc_use_pseudo_pic_reg): New function.
4704 (sparc_init_pic_reg): Likewise.
4705 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4706 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4708 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4710 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4711 Add item for branch_cost.
4713 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4715 PR rtl-optimization/83565
4716 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4717 not extend the result to a larger mode for rotate operations.
4718 (num_sign_bit_copies1): Likewise.
4720 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4723 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4725 Use values-Xc.o for -pedantic.
4726 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4728 2018-01-12 Martin Liska <mliska@suse.cz>
4731 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4733 (possible_polymorphic_call_targets): Use it.
4734 (ipa_devirt): Likewise.
4736 2018-01-12 Martin Liska <mliska@suse.cz>
4738 * profile-count.h (enum profile_quality): Use 0 as invalid
4739 enum value of profile_quality.
4741 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4743 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4744 -mext-string options.
4746 2018-01-12 Richard Biener <rguenther@suse.de>
4748 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4749 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4750 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4752 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4754 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4756 * configure.ac (--with-long-double-format): Add support for the
4757 configuration option to change the default long double format on
4759 * config.gcc (powerpc*-linux*-*): Likewise.
4760 * configure: Regenerate.
4761 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4762 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4763 used without modification.
4765 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4767 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4768 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4769 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4770 MISC_BUILTIN_SPEC_BARRIER.
4771 (rs6000_init_builtins): Likewise.
4772 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4774 (speculation_barrier): New define_insn.
4775 * doc/extend.texi: Document __builtin_speculation_barrier.
4777 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4780 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4781 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4782 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4784 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4785 integral modes instead of "ss" and "sd".
4786 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4787 vectors with 32-bit and 64-bit elements.
4788 (vecdupssescalarmodesuffix): New mode attribute.
4789 (vec_dup<mode>): Use it.
4791 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4794 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4795 frame if argument is passed on stack.
4797 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4800 * ree.c (combine_reaching_defs): Optimize also
4801 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4802 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4804 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4807 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4809 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4812 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4813 after they are computed.
4815 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4817 PR tree-optimization/83695
4818 * gimple-loop-linterchange.cc
4819 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4820 reset cached scev information after interchange.
4821 (pass_linterchange::execute): Remove call to scev_reset_htab.
4823 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4825 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4826 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4827 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4828 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4829 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4830 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4831 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4832 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4833 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4834 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4835 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4836 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4837 (V_lane_reg): Likewise.
4838 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4840 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4841 (vfmal_lane_low<mode>_intrinsic,
4842 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4843 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4844 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4845 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4846 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4847 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4849 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4851 * config/arm/arm-cpus.in (fp16fml): New feature.
4852 (ALL_SIMD): Add fp16fml.
4853 (armv8.2-a): Add fp16fml as an option.
4854 (armv8.3-a): Likewise.
4855 (armv8.4-a): Add fp16fml as part of fp16.
4856 * config/arm/arm.h (TARGET_FP16FML): Define.
4857 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4859 * config/arm/arm-modes.def (V2HF): Define.
4860 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4861 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4862 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4863 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4864 vfmsl_low, vfmsl_high): New set of builtins.
4865 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4866 (vfml_op): New code attribute.
4867 (VFMLHALVES): New int iterator.
4868 (VFML, VFMLSEL): New mode attributes.
4869 (V_reg): Define mapping for V2HF.
4870 (V_hi, V_lo): New mode attributes.
4871 (VF_constraint): Likewise.
4872 (vfml_half, vfml_half_selector): New int attributes.
4873 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4875 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4876 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4878 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4879 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4880 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4881 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4883 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4884 Document new effective target and option set.
4886 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4888 * config/arm/arm-cpus.in (armv8_4): New feature.
4889 (ARMv8_4a): New fgroup.
4890 (armv8.4-a): New arch.
4891 * config/arm/arm-tables.opt: Regenerate.
4892 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4893 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4894 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4895 Add matching rules for -march=armv8.4-a and extensions.
4896 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4898 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4901 * config/rx/rx.md (BW): New mode attribute.
4902 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4904 2018-01-11 Richard Biener <rguenther@suse.de>
4906 PR tree-optimization/83435
4907 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4908 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4909 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4911 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4912 Alan Hayward <alan.hayward@arm.com>
4913 David Sherwood <david.sherwood@arm.com>
4915 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4917 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4918 (aarch64_print_address_internal): Use it to check for a zero offset.
4920 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4921 Alan Hayward <alan.hayward@arm.com>
4922 David Sherwood <david.sherwood@arm.com>
4924 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4925 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4926 Return a poly_int64 rather than a HOST_WIDE_INT.
4927 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4928 rather than a HOST_WIDE_INT.
4929 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4930 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4931 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4932 final_offset from HOST_WIDE_INT to poly_int64.
4933 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4934 to_constant when getting the number of units in an Advanced SIMD
4936 (aarch64_builtin_vectorized_function): Check for a constant number
4938 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4940 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4941 attribute instead of GET_MODE_NUNITS.
4942 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4943 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4944 GET_MODE_SIZE for fixed-size registers.
4945 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4946 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4947 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4948 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4949 (aarch64_print_operand, aarch64_print_address_internal)
4950 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4951 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4952 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4953 Handle polynomial GET_MODE_SIZE.
4954 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4955 wider than SImode without modification.
4956 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4957 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4958 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4959 passing and returning SVE modes.
4960 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4961 rather than GEN_INT.
4962 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4963 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4964 (aarch64_allocate_and_probe_stack_space): Likewise.
4965 (aarch64_layout_frame): Cope with polynomial offsets.
4966 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4967 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4969 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4970 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4971 poly_int64 rather than a HOST_WIDE_INT.
4972 (aarch64_get_separate_components, aarch64_process_components)
4973 (aarch64_expand_prologue, aarch64_expand_epilogue)
4974 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4975 (aarch64_anchor_offset): New function, split out from...
4976 (aarch64_legitimize_address): ...here.
4977 (aarch64_builtin_vectorization_cost): Handle polynomial
4978 TYPE_VECTOR_SUBPARTS.
4979 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4981 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4982 number of elements from the PARALLEL rather than the mode.
4983 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4984 rather than GET_MODE_BITSIZE.
4985 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4986 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4987 (aarch64_expand_vec_perm_const_1): Handle polynomial
4988 d->perm.length () and d->perm elements.
4989 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4990 Apply to_constant to d->perm elements.
4991 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4992 polynomial CONST_VECTOR_NUNITS.
4993 (aarch64_move_pointer): Take amount as a poly_int64 rather
4995 (aarch64_progress_pointer): Avoid temporary variable.
4996 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4997 the mode attribute instead of GET_MODE.
4999 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5000 Alan Hayward <alan.hayward@arm.com>
5001 David Sherwood <david.sherwood@arm.com>
5003 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
5004 x exists before using it.
5005 (aarch64_add_constant_internal): Rename to...
5006 (aarch64_add_offset_1): ...this. Replace regnum with separate
5007 src and dest rtxes. Handle the case in which they're different,
5008 including when the offset is zero. Replace scratchreg with an rtx.
5009 Use 2 additions if there is no spare register into which we can
5010 move a 16-bit constant.
5011 (aarch64_add_constant): Delete.
5012 (aarch64_add_offset): Replace reg with separate src and dest
5013 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
5014 Use aarch64_add_offset_1.
5015 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
5016 an rtx rather than an int. Take the delta as a poly_int64
5017 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
5018 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
5019 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
5020 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
5021 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
5023 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
5024 aarch64_add_constant.
5026 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5028 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
5029 Use scalar_float_mode.
5031 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5033 * config/aarch64/aarch64-simd.md
5034 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
5035 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
5036 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
5037 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
5038 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
5039 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
5040 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
5041 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
5042 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
5043 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
5045 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5048 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
5049 targ_options->x_arm_arch_string is non NULL.
5051 2018-01-11 Tamar Christina <tamar.christina@arm.com>
5053 * config/aarch64/aarch64.h
5054 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
5056 2018-01-11 Sudakshina Das <sudi.das@arm.com>
5059 * expmed.c (emit_store_flag_force): Swap if const op0
5060 and change VOIDmode to mode of op0.
5062 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5064 PR rtl-optimization/83761
5065 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
5066 than bytes to mode_for_size.
5068 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5071 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
5072 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
5075 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5078 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
5079 when in layout mode.
5080 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
5081 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
5084 2018-01-10 Michael Collison <michael.collison@arm.com>
5086 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
5087 * config/aarch64/aarch64-option-extension.def: Add
5088 AARCH64_OPT_EXTENSION of 'fp16fml'.
5089 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5090 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
5091 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
5092 * config/aarch64/constraints.md (Ui7): New constraint.
5093 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
5094 (VFMLA_SEL_W): Ditto.
5097 (VFMLA16_LOW): New int iterator.
5098 (VFMLA16_HIGH): Ditto.
5099 (UNSPEC_FMLAL): New unspec.
5100 (UNSPEC_FMLSL): Ditto.
5101 (UNSPEC_FMLAL2): Ditto.
5102 (UNSPEC_FMLSL2): Ditto.
5103 (f16mac): New code attribute.
5104 * config/aarch64/aarch64-simd-builtins.def
5105 (aarch64_fmlal_lowv2sf): Ditto.
5106 (aarch64_fmlsl_lowv2sf): Ditto.
5107 (aarch64_fmlalq_lowv4sf): Ditto.
5108 (aarch64_fmlslq_lowv4sf): Ditto.
5109 (aarch64_fmlal_highv2sf): Ditto.
5110 (aarch64_fmlsl_highv2sf): Ditto.
5111 (aarch64_fmlalq_highv4sf): Ditto.
5112 (aarch64_fmlslq_highv4sf): Ditto.
5113 (aarch64_fmlal_lane_lowv2sf): Ditto.
5114 (aarch64_fmlsl_lane_lowv2sf): Ditto.
5115 (aarch64_fmlal_laneq_lowv2sf): Ditto.
5116 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
5117 (aarch64_fmlalq_lane_lowv4sf): Ditto.
5118 (aarch64_fmlsl_lane_lowv4sf): Ditto.
5119 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
5120 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
5121 (aarch64_fmlal_lane_highv2sf): Ditto.
5122 (aarch64_fmlsl_lane_highv2sf): Ditto.
5123 (aarch64_fmlal_laneq_highv2sf): Ditto.
5124 (aarch64_fmlsl_laneq_highv2sf): Ditto.
5125 (aarch64_fmlalq_lane_highv4sf): Ditto.
5126 (aarch64_fmlsl_lane_highv4sf): Ditto.
5127 (aarch64_fmlalq_laneq_highv4sf): Ditto.
5128 (aarch64_fmlsl_laneq_highv4sf): Ditto.
5129 * config/aarch64/aarch64-simd.md:
5130 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
5131 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5132 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
5133 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5134 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
5135 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
5136 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
5137 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
5138 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
5139 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
5140 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
5141 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
5142 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
5143 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
5144 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
5145 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
5146 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
5147 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
5148 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
5149 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
5150 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
5151 (vfmlsl_low_u32): Ditto.
5152 (vfmlalq_low_u32): Ditto.
5153 (vfmlslq_low_u32): Ditto.
5154 (vfmlal_high_u32): Ditto.
5155 (vfmlsl_high_u32): Ditto.
5156 (vfmlalq_high_u32): Ditto.
5157 (vfmlslq_high_u32): Ditto.
5158 (vfmlal_lane_low_u32): Ditto.
5159 (vfmlsl_lane_low_u32): Ditto.
5160 (vfmlal_laneq_low_u32): Ditto.
5161 (vfmlsl_laneq_low_u32): Ditto.
5162 (vfmlalq_lane_low_u32): Ditto.
5163 (vfmlslq_lane_low_u32): Ditto.
5164 (vfmlalq_laneq_low_u32): Ditto.
5165 (vfmlslq_laneq_low_u32): Ditto.
5166 (vfmlal_lane_high_u32): Ditto.
5167 (vfmlsl_lane_high_u32): Ditto.
5168 (vfmlal_laneq_high_u32): Ditto.
5169 (vfmlsl_laneq_high_u32): Ditto.
5170 (vfmlalq_lane_high_u32): Ditto.
5171 (vfmlslq_lane_high_u32): Ditto.
5172 (vfmlalq_laneq_high_u32): Ditto.
5173 (vfmlslq_laneq_high_u32): Ditto.
5174 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
5175 (AARCH64_FL_FOR_ARCH8_4): New.
5176 (AARCH64_ISA_F16FML): New ISA flag.
5177 (TARGET_F16FML): New feature flag for fp16fml.
5178 (doc/invoke.texi): Document new fp16fml option.
5180 2018-01-10 Michael Collison <michael.collison@arm.com>
5182 * config/aarch64/aarch64-builtins.c:
5183 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
5184 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5185 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
5186 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
5187 (AARCH64_ISA_SHA3): New ISA flag.
5188 (TARGET_SHA3): New feature flag for sha3.
5189 * config/aarch64/iterators.md (sha512_op): New int attribute.
5190 (CRYPTO_SHA512): New int iterator.
5191 (UNSPEC_SHA512H): New unspec.
5192 (UNSPEC_SHA512H2): Ditto.
5193 (UNSPEC_SHA512SU0): Ditto.
5194 (UNSPEC_SHA512SU1): Ditto.
5195 * config/aarch64/aarch64-simd-builtins.def
5196 (aarch64_crypto_sha512hqv2di): New builtin.
5197 (aarch64_crypto_sha512h2qv2di): Ditto.
5198 (aarch64_crypto_sha512su0qv2di): Ditto.
5199 (aarch64_crypto_sha512su1qv2di): Ditto.
5200 (aarch64_eor3qv8hi): Ditto.
5201 (aarch64_rax1qv2di): Ditto.
5202 (aarch64_xarqv2di): Ditto.
5203 (aarch64_bcaxqv8hi): Ditto.
5204 * config/aarch64/aarch64-simd.md:
5205 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
5206 (aarch64_crypto_sha512su0qv2di): Ditto.
5207 (aarch64_crypto_sha512su1qv2di): Ditto.
5208 (aarch64_eor3qv8hi): Ditto.
5209 (aarch64_rax1qv2di): Ditto.
5210 (aarch64_xarqv2di): Ditto.
5211 (aarch64_bcaxqv8hi): Ditto.
5212 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
5213 (vsha512h2q_u64): Ditto.
5214 (vsha512su0q_u64): Ditto.
5215 (vsha512su1q_u64): Ditto.
5216 (veor3q_u16): Ditto.
5217 (vrax1q_u64): Ditto.
5219 (vbcaxq_u16): Ditto.
5220 * config/arm/types.md (crypto_sha512): New type attribute.
5221 (crypto_sha3): Ditto.
5222 (doc/invoke.texi): Document new sha3 option.
5224 2018-01-10 Michael Collison <michael.collison@arm.com>
5226 * config/aarch64/aarch64-builtins.c:
5227 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
5228 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5229 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
5230 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
5231 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
5232 (AARCH64_ISA_SM4): New ISA flag.
5233 (TARGET_SM4): New feature flag for sm4.
5234 * config/aarch64/aarch64-simd-builtins.def
5235 (aarch64_sm3ss1qv4si): Ditto.
5236 (aarch64_sm3tt1aq4si): Ditto.
5237 (aarch64_sm3tt1bq4si): Ditto.
5238 (aarch64_sm3tt2aq4si): Ditto.
5239 (aarch64_sm3tt2bq4si): Ditto.
5240 (aarch64_sm3partw1qv4si): Ditto.
5241 (aarch64_sm3partw2qv4si): Ditto.
5242 (aarch64_sm4eqv4si): Ditto.
5243 (aarch64_sm4ekeyqv4si): Ditto.
5244 * config/aarch64/aarch64-simd.md:
5245 (aarch64_sm3ss1qv4si): Ditto.
5246 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
5247 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
5248 (aarch64_sm4eqv4si): Ditto.
5249 (aarch64_sm4ekeyqv4si): Ditto.
5250 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
5251 (sm3part_op): Ditto.
5252 (CRYPTO_SM3TT): Ditto.
5253 (CRYPTO_SM3PART): Ditto.
5254 (UNSPEC_SM3SS1): New unspec.
5255 (UNSPEC_SM3TT1A): Ditto.
5256 (UNSPEC_SM3TT1B): Ditto.
5257 (UNSPEC_SM3TT2A): Ditto.
5258 (UNSPEC_SM3TT2B): Ditto.
5259 (UNSPEC_SM3PARTW1): Ditto.
5260 (UNSPEC_SM3PARTW2): Ditto.
5261 (UNSPEC_SM4E): Ditto.
5262 (UNSPEC_SM4EKEY): Ditto.
5263 * config/aarch64/constraints.md (Ui2): New constraint.
5264 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
5265 * config/arm/types.md (crypto_sm3): New type attribute.
5266 (crypto_sm4): Ditto.
5267 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
5268 (vsm3tt1aq_u32): Ditto.
5269 (vsm3tt1bq_u32): Ditto.
5270 (vsm3tt2aq_u32): Ditto.
5271 (vsm3tt2bq_u32): Ditto.
5272 (vsm3partw1q_u32): Ditto.
5273 (vsm3partw2q_u32): Ditto.
5274 (vsm4eq_u32): Ditto.
5275 (vsm4ekeyq_u32): Ditto.
5276 (doc/invoke.texi): Document new sm4 option.
5278 2018-01-10 Michael Collison <michael.collison@arm.com>
5280 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
5281 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
5282 (AARCH64_FL_FOR_ARCH8_4): New.
5283 (AARCH64_FL_V8_4): New flag.
5284 (doc/invoke.texi): Document new armv8.4-a option.
5286 2018-01-10 Michael Collison <michael.collison@arm.com>
5288 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5289 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
5290 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
5291 * config/aarch64/aarch64-option-extension.def: Add
5292 AARCH64_OPT_EXTENSION of 'sha2'.
5293 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
5294 (crypto): Disable sha2 and aes if crypto disabled.
5295 (crypto): Enable aes and sha2 if enabled.
5296 (simd): Disable sha2 and aes if simd disabled.
5297 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
5299 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
5300 (TARGET_SHA2): New feature flag for sha2.
5301 (TARGET_AES): New feature flag for aes.
5302 * config/aarch64/aarch64-simd.md:
5303 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
5304 conditional on TARGET_AES.
5305 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
5306 (aarch64_crypto_sha1hsi): Make pattern conditional
5308 (aarch64_crypto_sha1hv4si): Ditto.
5309 (aarch64_be_crypto_sha1hv4si): Ditto.
5310 (aarch64_crypto_sha1su1v4si): Ditto.
5311 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
5312 (aarch64_crypto_sha1su0v4si): Ditto.
5313 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
5314 (aarch64_crypto_sha256su0v4si): Ditto.
5315 (aarch64_crypto_sha256su1v4si): Ditto.
5316 (doc/invoke.texi): Document new aes and sha2 options.
5318 2018-01-10 Martin Sebor <msebor@redhat.com>
5320 PR tree-optimization/83781
5321 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
5324 2018-01-11 Martin Sebor <msebor@gmail.com>
5325 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5327 PR tree-optimization/83501
5328 PR tree-optimization/81703
5330 * tree-ssa-strlen.c (get_string_cst): Rename...
5331 (get_string_len): ...to this. Handle global constants.
5332 (handle_char_store): Adjust.
5334 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
5335 Jim Wilson <jimw@sifive.com>
5337 * config/riscv/riscv-protos.h (riscv_output_return): New.
5338 * config/riscv/riscv.c (struct machine_function): New naked_p field.
5339 (riscv_attribute_table, riscv_output_return),
5340 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
5341 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
5342 (riscv_compute_frame_info): Only compute frame->mask if not a naked
5344 (riscv_expand_prologue): Add early return for naked function.
5345 (riscv_expand_epilogue): Likewise.
5346 (riscv_function_ok_for_sibcall): Return false for naked function.
5347 (riscv_set_current_function): New.
5348 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
5349 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
5350 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
5351 * doc/extend.texi (RISC-V Function Attributes): New.
5353 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
5355 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
5356 check for 128-bit long double before checking TCmode.
5357 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
5358 128-bit long doubles before checking TFmode or TCmode.
5359 (FLOAT128_IBM_P): Likewise.
5361 2018-01-10 Martin Sebor <msebor@redhat.com>
5363 PR tree-optimization/83671
5364 * builtins.c (c_strlen): Unconditionally return zero for the empty
5366 Use -Warray-bounds for warnings.
5367 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
5368 for non-constant array indices with COMPONENT_REF, arrays of
5369 arrays, and pointers to arrays.
5370 (gimple_fold_builtin_strlen): Determine and set length range for
5371 non-constant character arrays.
5373 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
5376 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
5379 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
5381 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
5383 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5386 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
5387 VECTOR_MEM_ALTIVEC_OR_VSX_P.
5388 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
5389 indexed_or_indirect_operand predicate.
5390 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
5391 (*vsx_le_perm_load_v8hi): Likewise.
5392 (*vsx_le_perm_load_v16qi): Likewise.
5393 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
5394 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
5395 (*vsx_le_perm_store_v8hi): Likewise.
5396 (*vsx_le_perm_store_v16qi): Likewise.
5397 (eight unnamed splitters): Likewise.
5399 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5401 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
5402 * config/rs6000/emmintrin.h: Likewise.
5403 * config/rs6000/mmintrin.h: Likewise.
5404 * config/rs6000/xmmintrin.h: Likewise.
5406 2018-01-10 David Malcolm <dmalcolm@redhat.com>
5409 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
5411 * tree.c (tree_nop_conversion): Return true for location wrapper
5413 (maybe_wrap_with_location): New function.
5414 (selftest::check_strip_nops): New function.
5415 (selftest::test_location_wrappers): New function.
5416 (selftest::tree_c_tests): Call it.
5417 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
5418 (maybe_wrap_with_location): New decl.
5419 (EXPR_LOCATION_WRAPPER_P): New macro.
5420 (location_wrapper_p): New inline function.
5421 (tree_strip_any_location_wrapper): New inline function.
5423 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
5426 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5427 stack_realign_offset for the largest alignment of stack slot
5429 (ix86_find_max_used_stack_alignment): New function.
5430 (ix86_finalize_stack_frame_flags): Use it. Set
5431 max_used_stack_alignment if we don't realign stack.
5432 * config/i386/i386.h (machine_function): Add
5433 max_used_stack_alignment.
5435 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5437 * config/arm/arm.opt (-mbranch-cost): New option.
5438 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5441 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5444 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5445 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5447 2018-01-10 Richard Biener <rguenther@suse.de>
5450 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5451 early out so it also covers the case where we have a non-NULL
5454 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5456 PR tree-optimization/83753
5457 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5458 for non-strided grouped accesses if the number of elements is 1.
5460 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5463 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5464 * i386.h (TARGET_USE_GATHER): Define.
5465 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5467 2018-01-10 Martin Liska <mliska@suse.cz>
5470 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5471 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5473 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5474 CLEANUP_NO_PARTITIONING is not set.
5476 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5478 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5479 for vectors, as a partial revert of r254296.
5480 * rtl.h (const_vec_p): Delete.
5481 (const_vec_duplicate_p): Don't test for vector CONSTs.
5482 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5483 * expmed.c (make_tree): Likewise.
5486 * common.md (E, F): Use CONSTANT_P instead of checking for
5488 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5489 checking for CONST_VECTOR.
5491 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5494 * predict.c (force_edge_cold): Handle in more sane way edges
5497 2018-01-09 Carl Love <cel@us.ibm.com>
5499 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5501 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5502 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5503 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5504 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5505 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5506 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5507 * config/rs6000/rs6000-protos.h: Add extern defition for
5508 rs6000_generate_float2_double_code.
5509 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5511 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5512 (float2_v2df): Add define_expand.
5514 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5517 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5518 op_mode in the force_to_mode call.
5520 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5522 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5523 instead of checking each element individually.
5524 (aarch64_evpc_uzp): Likewise.
5525 (aarch64_evpc_zip): Likewise.
5526 (aarch64_evpc_ext): Likewise.
5527 (aarch64_evpc_rev): Likewise.
5528 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5529 instead of checking each element individually. Return true without
5531 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5532 whether all selected elements come from the same input, instead of
5533 checking each element individually. Remove calls to gen_rtx_REG,
5534 start_sequence and end_sequence and instead assert that no rtl is
5537 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5539 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5540 order of HIGH and CONST checks.
5542 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5544 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5545 if the destination isn't an SSA_NAME.
5547 2018-01-09 Richard Biener <rguenther@suse.de>
5549 PR tree-optimization/83668
5550 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5552 (canonicalize_loop_form): ... here, renamed from ...
5553 (canonicalize_loop_closed_ssa_form): ... this and amended to
5554 swap successor edges for loop exit blocks to make us use
5555 the RPO order we need for initial schedule generation.
5557 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5559 PR tree-optimization/64811
5560 * match.pd: When optimizing comparisons with Inf, avoid
5561 introducing or losing exceptions from comparisons with NaN.
5563 2018-01-09 Martin Liska <mliska@suse.cz>
5566 * asan.c (shadow_mem_size): Add gcc_assert.
5568 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5570 Don't save registers in main().
5573 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5574 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5575 * config/avr/avr.c (avr_set_current_function): Don't error if
5576 naked, OS_task or OS_main are specified at the same time.
5577 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5579 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5581 * common/config/avr/avr-common.c (avr_option_optimization_table):
5582 Switch on -mmain-is-OS_task for optimizing compilations.
5584 2018-01-09 Richard Biener <rguenther@suse.de>
5586 PR tree-optimization/83572
5587 * graphite.c: Include cfganal.h.
5588 (graphite_transform_loops): Connect infinite loops to exit
5589 and remove fake edges at the end.
5591 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5593 * ipa-inline.c (edge_badness): Revert accidental checkin.
5595 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5598 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5599 symbols; not inline clones.
5601 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5604 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5605 hard registers. Formatting fixes.
5607 PR preprocessor/83722
5608 * gcc.c (try_generate_repro): Pass
5609 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5610 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5613 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5614 Kito Cheng <kito.cheng@gmail.com>
5616 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5617 (riscv_leaf_function_p): Delete.
5618 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5620 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5622 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5624 (do_ifelse): New function.
5625 (do_isel): New function.
5626 (do_sub3): New function.
5627 (do_add3): New function.
5628 (do_load_mask_compare): New function.
5629 (do_overlap_load_compare): New function.
5630 (expand_compare_loop): New function.
5631 (expand_block_compare): Call expand_compare_loop() when appropriate.
5632 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5634 (-mblock-compare-inline-loop-limit): New option.
5636 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5639 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5640 Reverse order of second and third operands in first alternative.
5641 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5642 of first and second elements in UNSPEC_VPERMR vector.
5643 (altivec_expand_vec_perm_le): Likewise.
5645 2018-01-08 Jeff Law <law@redhat.com>
5647 PR rtl-optimizatin/81308
5648 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5649 (process_switch): If group_case_labels makes a change, then set
5651 (pass_convert_switch::execute): If a switch is converted, then
5652 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5654 PR rtl-optimization/81308
5655 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5658 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5660 PR target/83663 - Revert r255946
5661 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5662 generation for cases where splatting a value is not useful.
5663 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5664 across a vec_duplicate and a paradoxical subreg forming a vector
5665 mode to a vec_concat.
5667 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5669 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5670 -march=armv8.3-a variants.
5671 * config/arm/t-multilib: Likewise.
5672 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5674 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5676 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5678 (cceq_ior_compare_complement): Give it a name so I can use it, and
5679 change boolean_or_operator predicate to boolean_operator so it can
5680 be used to generate a crand.
5681 (eqne): New code iterator.
5682 (bd/bd_neg): New code_attrs.
5683 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5684 a single define_insn.
5685 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5686 decrement (bdnzt/bdnzf/bdzt/bdzf).
5687 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5688 with the new names of the branch decrement patterns, and added the
5689 names of the branch decrement conditional patterns.
5691 2018-01-08 Richard Biener <rguenther@suse.de>
5693 PR tree-optimization/83563
5694 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5697 2018-01-08 Richard Biener <rguenther@suse.de>
5700 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5702 2018-01-08 Richard Biener <rguenther@suse.de>
5704 PR tree-optimization/83685
5705 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5706 references to abnormals.
5708 2018-01-08 Richard Biener <rguenther@suse.de>
5711 * dwarf2out.c (output_indirect_strings): Handle empty
5712 skeleton_debug_str_hash.
5713 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5715 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5717 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5718 (emit_store_direct): Likewise.
5719 (arc_trampoline_adjust_address): Likewise.
5720 (arc_asm_trampoline_template): New function.
5721 (arc_initialize_trampoline): Use asm_trampoline_template.
5722 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5723 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5724 * config/arc/arc.md (flush_icache): Delete pattern.
5726 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5728 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5729 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5732 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5735 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5736 by not USED_FOR_TARGET.
5737 (make_pass_resolve_sw_modes): Likewise.
5739 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5741 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5744 2018-01-08 Richard Biener <rguenther@suse.de>
5747 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5749 2018-01-08 Richard Biener <rguenther@suse.de>
5752 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5754 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5757 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5758 basic blocks with a small number of successors.
5759 (convert_control_dep_chain_into_preds): Improve handling of
5761 (dump_predicates): Split apart into...
5762 (dump_pred_chain): ...here...
5763 (dump_pred_info): ...and here.
5764 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5765 (can_chain_union_be_invalidated_p): Improve check for invalidation
5767 (uninit_uses_cannot_happen): Avoid unnecessary if
5768 convert_control_dep_chain_into_preds yielded nothing.
5770 2018-01-06 Martin Sebor <msebor@redhat.com>
5772 PR tree-optimization/83640
5773 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5774 subtracting negative offset from size.
5775 (builtin_access::overlap): Adjust offset bounds of the access to fall
5776 within the size of the object if possible.
5778 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5780 PR rtl-optimization/83699
5781 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5782 extract_bit_field_as_subreg to cases in which the extracted
5783 value is also a vector.
5785 * lra-constraints.c (process_alt_operands): Test for the equivalence
5786 substitutions when detecting a possible reload cycle.
5788 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5791 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5792 by default if flag_selective_schedling{,2}. Formatting fixes.
5794 PR rtl-optimization/83682
5795 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5796 if it has non-VECTOR_MODE element mode.
5797 (vec_duplicate_p): Likewise.
5800 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5801 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5803 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5806 * config/i386/i386-builtin.def
5807 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5808 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5809 Require also OPTION_MASK_ISA_AVX512F in addition to
5810 OPTION_MASK_ISA_GFNI.
5811 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5812 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5813 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5814 to OPTION_MASK_ISA_GFNI.
5815 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5816 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5817 OPTION_MASK_ISA_AVX512BW.
5818 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5819 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5820 addition to OPTION_MASK_ISA_GFNI.
5821 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5822 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5823 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5824 to OPTION_MASK_ISA_GFNI.
5825 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5826 a requirement for all ISAs rather than any of them with a few
5828 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5830 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5831 bitmasks to be enabled with 3 exceptions, instead of requiring any
5832 enabled ISA with lots of exceptions.
5833 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5834 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5835 Change avx512bw in isa attribute to avx512f.
5836 * config/i386/sgxintrin.h: Add license boilerplate.
5837 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5838 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5839 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5840 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5842 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5843 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5844 temporarily sse2 rather than sse if not enabled already.
5847 * config/i386/sse.md (VI248_VLBW): Rename to ...
5848 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5849 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5850 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5851 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5852 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5853 mode iterator instead of VI248_VLBW.
5855 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5857 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5858 (record_modified): Skip clobbers; add debug output.
5859 (param_change_prob): Use sreal frequencies.
5861 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5863 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5864 punt for user-aligned variables.
5866 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5868 * tree-chrec.c (chrec_contains_symbols): Return true for
5871 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5874 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5875 of (x|y) == x for BICS pattern.
5877 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5879 PR tree-optimization/83605
5880 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5881 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5884 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5886 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5887 * config/epiphany/rtems.h: New file.
5889 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5890 Uros Bizjak <ubizjak@gmail.com>
5893 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5894 QIreg_operand instead of register_operand predicate.
5895 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5896 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5897 comments instead of -fmitigate[-_]rop.
5899 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5902 * cgraphunit.c (symbol_table::compile): Switch to text_section
5903 before calling assembly_start debug hook.
5904 * run-rtl-passes.c (run_rtl_passes): Likewise.
5907 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5909 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5910 range_int_cst_p rather than !symbolic_range_p before calling
5911 extract_range_from_multiplicative_op_1.
5913 2018-01-04 Jeff Law <law@redhat.com>
5915 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5916 redundant test in assertion.
5918 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5920 * doc/rtl.texi: Document machine_mode wrapper classes.
5922 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5924 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5927 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5929 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5930 the VEC_PERM_EXPR fold to fail.
5932 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5935 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5936 to switched_sections.
5938 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5941 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5944 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5947 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5948 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5950 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5953 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5954 is BLKmode and bitpos not zero or mode change is needed.
5956 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5959 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5962 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5965 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5966 instead of MULT rtx. Update all corresponding splitters.
5968 (*ssub<modesuffix>): Ditto.
5970 (*cmp_sadd_di): Update split patterns.
5971 (*cmp_sadd_si): Ditto.
5972 (*cmp_sadd_sidi): Ditto.
5973 (*cmp_ssub_di): Ditto.
5974 (*cmp_ssub_si): Ditto.
5975 (*cmp_ssub_sidi): Ditto.
5976 * config/alpha/predicates.md (const23_operand): New predicate.
5977 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5978 Look for ASHIFT, not MULT inner operand.
5979 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5981 2018-01-04 Martin Liska <mliska@suse.cz>
5983 PR gcov-profile/83669
5984 * gcov.c (output_intermediate_file): Add version to intermediate
5986 * doc/gcov.texi: Document new field 'version' in intermediate
5987 file format. Fix location of '-k' option of gcov command.
5989 2018-01-04 Martin Liska <mliska@suse.cz>
5992 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5994 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5996 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5998 2018-01-03 Martin Sebor <msebor@redhat.com>
6000 PR tree-optimization/83655
6001 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
6002 checking calls with invalid arguments.
6004 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6006 * tree-vect-stmts.c (vect_get_store_rhs): New function.
6007 (vectorizable_mask_load_store): Delete.
6008 (vectorizable_call): Return false for masked loads and stores.
6009 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
6010 instead of gimple_assign_rhs1.
6011 (vectorizable_load): Handle IFN_MASK_LOAD.
6012 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
6014 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6016 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
6018 (vectorizable_mask_load_store): ...here.
6019 (vectorizable_load): ...and here.
6021 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6023 * tree-vect-stmts.c (vect_build_all_ones_mask)
6024 (vect_build_zero_merge_argument): New functions, split out from...
6025 (vectorizable_load): ...here.
6027 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6029 * tree-vect-stmts.c (vect_check_store_rhs): New function,
6031 (vectorizable_mask_load_store): ...here.
6032 (vectorizable_store): ...and here.
6034 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6036 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
6038 (vectorizable_mask_load_store): ...here.
6040 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6042 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
6043 (vect_model_store_cost): Take a vec_load_store_type instead of a
6045 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
6046 (vect_model_store_cost): Take a vec_load_store_type instead of a
6048 (vectorizable_mask_load_store): Update accordingly.
6049 (vectorizable_store): Likewise.
6050 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
6052 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6054 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
6055 IFN_MASK_LOAD calls here rather than...
6056 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
6058 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6059 Alan Hayward <alan.hayward@arm.com>
6060 David Sherwood <david.sherwood@arm.com>
6062 * expmed.c (extract_bit_field_1): For vector extracts,
6063 fall back to extract_bit_field_as_subreg if vec_extract
6066 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6067 Alan Hayward <alan.hayward@arm.com>
6068 David Sherwood <david.sherwood@arm.com>
6070 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
6071 they are variable or constant sized.
6072 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
6073 slots for constant-sized data.
6075 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6076 Alan Hayward <alan.hayward@arm.com>
6077 David Sherwood <david.sherwood@arm.com>
6079 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
6080 handling COND_EXPRs with boolean comparisons, try to find a better
6081 basis for the mask type than the boolean itself.
6083 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6085 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
6086 is calculated and how it can be overridden.
6087 * genmodes.c (max_bitsize_mode_any_mode): New variable.
6088 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
6090 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
6093 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6094 Alan Hayward <alan.hayward@arm.com>
6095 David Sherwood <david.sherwood@arm.com>
6097 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
6098 Remove the mode argument.
6099 (aarch64_simd_valid_immediate): Remove the mode and inverse
6101 * config/aarch64/iterators.md (bitsize): New iterator.
6102 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
6103 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
6104 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
6105 aarch64_simd_valid_immediate.
6106 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
6107 (aarch64_reg_or_bic_imm): Likewise.
6108 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
6109 with an insn_type enum and msl with a modifier_type enum.
6110 Replace element_width with a scalar_mode. Change the shift
6111 to unsigned int. Add constructors for scalar_float_mode and
6112 scalar_int_mode elements.
6113 (aarch64_vect_float_const_representable_p): Delete.
6114 (aarch64_can_const_movi_rtx_p)
6115 (aarch64_simd_scalar_immediate_valid_for_move)
6116 (aarch64_simd_make_constant): Update call to
6117 aarch64_simd_valid_immediate.
6118 (aarch64_advsimd_valid_immediate_hs): New function.
6119 (aarch64_advsimd_valid_immediate): Likewise.
6120 (aarch64_simd_valid_immediate): Remove mode and inverse
6121 arguments. Rewrite to use the above. Use const_vec_duplicate_p
6122 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
6123 and aarch64_float_const_representable_p on the result.
6124 (aarch64_output_simd_mov_immediate): Remove mode argument.
6125 Update call to aarch64_simd_valid_immediate and use of
6126 simd_immediate_info.
6127 (aarch64_output_scalar_simd_mov_immediate): Update call
6130 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6131 Alan Hayward <alan.hayward@arm.com>
6132 David Sherwood <david.sherwood@arm.com>
6134 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
6135 (mode_nunits): Likewise CONST_MODE_NUNITS.
6136 * machmode.def (ADJUST_NUNITS): Document.
6137 * genmodes.c (mode_data::need_nunits_adj): New field.
6138 (blank_mode): Update accordingly.
6139 (adj_nunits): New variable.
6140 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
6142 (emit_mode_size_inline): Set need_bytesize_adj for all modes
6143 listed in adj_nunits.
6144 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
6145 listed in adj_nunits. Don't emit case statements for such modes.
6146 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
6147 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
6148 nothing if adj_nunits is nonnull.
6149 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
6150 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
6151 (emit_mode_fbit): Update use of print_maybe_const_decl.
6152 (emit_move_size): Likewise. Treat the array as non-const
6154 (emit_mode_adjustments): Handle adj_nunits.
6156 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6158 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
6159 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
6160 (VECTOR_MODES): Use it.
6161 (make_vector_modes): Take the prefix as an argument.
6163 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6164 Alan Hayward <alan.hayward@arm.com>
6165 David Sherwood <david.sherwood@arm.com>
6167 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
6168 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
6169 for MODE_VECTOR_BOOL.
6170 * machmode.def (VECTOR_BOOL_MODE): Document.
6171 * genmodes.c (VECTOR_BOOL_MODE): New macro.
6172 (make_vector_bool_mode): New function.
6173 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
6175 * lto-streamer-in.c (lto_input_mode_table): Likewise.
6176 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
6178 * stor-layout.c (int_mode_for_mode): Likewise.
6179 * tree.c (build_vector_type_for_mode): Likewise.
6180 * varasm.c (output_constant_pool_2): Likewise.
6181 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
6182 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
6183 for MODE_VECTOR_BOOL.
6184 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
6185 of mode class checks.
6186 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
6187 instead of a list of mode class checks.
6188 (expand_vector_scalar_condition): Likewise.
6189 (type_for_widest_vector_mode): Handle BImode as an inner mode.
6191 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6192 Alan Hayward <alan.hayward@arm.com>
6193 David Sherwood <david.sherwood@arm.com>
6195 * machmode.h (mode_size): Change from unsigned short to
6197 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
6198 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6199 or if measurement_type is not polynomial.
6200 (fixed_size_mode::includes_p): Check for constant-sized modes.
6201 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
6202 return a poly_uint16 rather than an unsigned short.
6203 (emit_mode_size): Change the type of mode_size from unsigned short
6204 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
6205 (emit_mode_adjustments): Cope with polynomial vector sizes.
6206 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6208 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6210 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
6211 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
6212 * caller-save.c (setup_save_areas): Likewise.
6213 (replace_reg_with_saved_mem): Likewise.
6214 * calls.c (emit_library_call_value_1): Likewise.
6215 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
6216 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
6217 (gen_lowpart_for_combine): Likewise.
6218 * convert.c (convert_to_integer_1): Likewise.
6219 * cse.c (equiv_constant, cse_insn): Likewise.
6220 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
6221 (cselib_subst_to_values): Likewise.
6222 * dce.c (word_dce_process_block): Likewise.
6223 * df-problems.c (df_word_lr_mark_ref): Likewise.
6224 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
6225 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
6226 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
6227 (rtl_for_decl_location): Likewise.
6228 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
6229 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
6230 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
6231 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
6232 (expand_expr_real_1): Likewise.
6233 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
6234 (pad_below): Likewise.
6235 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6236 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
6237 * ira.c (get_subreg_tracking_sizes): Likewise.
6238 * ira-build.c (ira_create_allocno_objects): Likewise.
6239 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
6240 (ira_sort_regnos_for_alter_reg): Likewise.
6241 * ira-costs.c (record_operand_costs): Likewise.
6242 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
6243 (resolve_simple_move): Likewise.
6244 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
6245 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
6246 (lra_constraints): Likewise.
6247 (CONST_POOL_OK_P): Reject variable-sized modes.
6248 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
6249 (add_pseudo_to_slot, lra_spill): Likewise.
6250 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6251 * optabs-query.c (get_best_extraction_insn): Likewise.
6252 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6253 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
6254 (expand_mult_highpart, valid_multiword_target_p): Likewise.
6255 * recog.c (offsettable_address_addr_space_p): Likewise.
6256 * regcprop.c (maybe_mode_change): Likewise.
6257 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
6258 * regrename.c (build_def_use): Likewise.
6259 * regstat.c (dump_reg_info): Likewise.
6260 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
6261 (find_reloads, find_reloads_subreg_address): Likewise.
6262 * reload1.c (eliminate_regs_1): Likewise.
6263 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
6264 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
6265 (simplify_binary_operation_1, simplify_subreg): Likewise.
6266 * targhooks.c (default_function_arg_padding): Likewise.
6267 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
6268 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
6269 (verify_gimple_assign_ternary): Likewise.
6270 * tree-inline.c (estimate_move_cost): Likewise.
6271 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6272 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
6273 (get_address_cost_ainc): Likewise.
6274 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
6275 (vect_supportable_dr_alignment): Likewise.
6276 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6277 (vectorizable_reduction): Likewise.
6278 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
6279 (vectorizable_operation, vectorizable_load): Likewise.
6280 * tree.c (build_same_sized_truth_vector_type): Likewise.
6281 * valtrack.c (cleanup_auto_inc_dec): Likewise.
6282 * var-tracking.c (emit_note_insn_var_location): Likewise.
6283 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
6284 (ADDR_VEC_ALIGN): Likewise.
6286 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6287 Alan Hayward <alan.hayward@arm.com>
6288 David Sherwood <david.sherwood@arm.com>
6290 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
6292 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6293 or if measurement_type is polynomial.
6294 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
6295 * combine.c (make_extraction): Likewise.
6296 * dse.c (find_shift_sequence): Likewise.
6297 * dwarf2out.c (mem_loc_descriptor): Likewise.
6298 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
6299 (extract_bit_field, extract_low_bits): Likewise.
6300 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
6301 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
6302 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
6303 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
6304 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6305 * reload.c (find_reloads): Likewise.
6306 * reload1.c (alter_reg): Likewise.
6307 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
6308 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
6309 * tree-if-conv.c (predicate_mem_writes): Likewise.
6310 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
6311 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
6312 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
6313 * valtrack.c (dead_debug_insert_temp): Likewise.
6314 * varasm.c (mergeable_constant_section): Likewise.
6315 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
6317 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6318 Alan Hayward <alan.hayward@arm.com>
6319 David Sherwood <david.sherwood@arm.com>
6321 * expr.c (expand_assignment): Cope with polynomial mode sizes
6322 when assigning to a CONCAT.
6324 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6325 Alan Hayward <alan.hayward@arm.com>
6326 David Sherwood <david.sherwood@arm.com>
6328 * machmode.h (mode_precision): Change from unsigned short to
6330 (mode_to_precision): Return a poly_uint16 rather than an unsigned
6332 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
6333 or if measurement_type is not polynomial.
6334 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
6335 in which the mode is already known to be a scalar_int_mode.
6336 * genmodes.c (emit_mode_precision): Change the type of mode_precision
6337 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
6339 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6340 for GET_MODE_PRECISION.
6341 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6342 for GET_MODE_PRECISION.
6343 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
6345 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
6346 (expand_field_assignment, make_extraction): Likewise.
6347 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
6348 (get_last_value): Likewise.
6349 * convert.c (convert_to_integer_1): Likewise.
6350 * cse.c (cse_insn): Likewise.
6351 * expr.c (expand_expr_real_1): Likewise.
6352 * lra-constraints.c (simplify_operand_subreg): Likewise.
6353 * optabs-query.c (can_atomic_load_p): Likewise.
6354 * optabs.c (expand_atomic_load): Likewise.
6355 (expand_atomic_store): Likewise.
6356 * ree.c (combine_reaching_defs): Likewise.
6357 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
6358 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
6359 * tree.h (type_has_mode_precision_p): Likewise.
6360 * ubsan.c (instrument_si_overflow): Likewise.
6362 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6363 Alan Hayward <alan.hayward@arm.com>
6364 David Sherwood <david.sherwood@arm.com>
6366 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
6367 polynomial numbers of units.
6368 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
6369 (valid_vector_subparts_p): New function.
6370 (build_vector_type): Remove temporary shim and take the number
6371 of units as a poly_uint64 rather than an int.
6372 (build_opaque_vector_type): Take the number of units as a
6373 poly_uint64 rather than an int.
6374 * tree.c (build_vector_from_ctor): Handle polynomial
6375 TYPE_VECTOR_SUBPARTS.
6376 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
6377 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
6378 (build_vector_from_val): If the number of units is variable,
6379 use build_vec_duplicate_cst for constant operands and
6380 VEC_DUPLICATE_EXPR otherwise.
6381 (make_vector_type): Remove temporary is_constant ().
6382 (build_vector_type, build_opaque_vector_type): Take the number of
6383 units as a poly_uint64 rather than an int.
6384 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
6386 * cfgexpand.c (expand_debug_expr): Likewise.
6387 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
6388 (store_constructor, expand_expr_real_1): Likewise.
6389 (const_scalar_mask_from_tree): Likewise.
6390 * fold-const-call.c (fold_const_reduction): Likewise.
6391 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
6392 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
6393 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
6394 (fold_relational_const): Likewise.
6395 (native_interpret_vector): Likewise. Change the size from an
6396 int to an unsigned int.
6397 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
6398 TYPE_VECTOR_SUBPARTS.
6399 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
6400 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
6401 duplicating a non-constant operand into a variable-length vector.
6402 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
6403 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
6404 * ipa-icf.c (sem_variable::equals): Likewise.
6405 * match.pd: Likewise.
6406 * omp-simd-clone.c (simd_clone_subparts): Likewise.
6407 * print-tree.c (print_node): Likewise.
6408 * stor-layout.c (layout_type): Likewise.
6409 * targhooks.c (default_builtin_vectorization_cost): Likewise.
6410 * tree-cfg.c (verify_gimple_comparison): Likewise.
6411 (verify_gimple_assign_binary): Likewise.
6412 (verify_gimple_assign_ternary): Likewise.
6413 (verify_gimple_assign_single): Likewise.
6414 * tree-pretty-print.c (dump_generic_node): Likewise.
6415 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6416 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
6417 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
6418 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
6419 (vect_shift_permute_load_chain): Likewise.
6420 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
6421 (expand_vector_condition, optimize_vector_constructor): Likewise.
6422 (lower_vec_perm, get_compute_type): Likewise.
6423 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6424 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
6425 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
6426 (vect_recog_mask_conversion_pattern): Likewise.
6427 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6428 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6429 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6430 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6431 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6432 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6433 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6434 (supportable_widening_operation): Likewise.
6435 (supportable_narrowing_operation): Likewise.
6436 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6438 * varasm.c (output_constant): Likewise.
6440 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6441 Alan Hayward <alan.hayward@arm.com>
6442 David Sherwood <david.sherwood@arm.com>
6444 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6445 so that both the length == 3 and length != 3 cases set up their
6446 own permute vectors. Add comments explaining why we know the
6447 number of elements is constant.
6448 (vect_permute_load_chain): Likewise.
6450 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6451 Alan Hayward <alan.hayward@arm.com>
6452 David Sherwood <david.sherwood@arm.com>
6454 * machmode.h (mode_nunits): Change from unsigned char to
6456 (ONLY_FIXED_SIZE_MODES): New macro.
6457 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6458 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6459 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6461 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6462 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6463 or if measurement_type is not polynomial.
6464 * genmodes.c (ZERO_COEFFS): New macro.
6465 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6467 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6468 Use ZERO_COEFFS when emitting initializers.
6469 * data-streamer.h (bp_pack_poly_value): New function.
6470 (bp_unpack_poly_value): Likewise.
6471 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6472 for GET_MODE_NUNITS.
6473 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6474 for GET_MODE_NUNITS.
6475 * tree.c (make_vector_type): Remove temporary shim and make
6476 the real function take the number of units as a poly_uint64
6478 (build_vector_type_for_mode): Handle polynomial nunits.
6479 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6480 * emit-rtl.c (const_vec_series_p_1): Likewise.
6481 (gen_rtx_CONST_VECTOR): Likewise.
6482 * fold-const.c (test_vec_duplicate_folding): Likewise.
6483 * genrecog.c (validate_pattern): Likewise.
6484 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6485 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6486 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6487 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6488 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6489 * rtlanal.c (subreg_get_info): Likewise.
6490 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6491 (vect_grouped_load_supported): Likewise.
6492 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6493 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6494 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6495 (simplify_const_unary_operation, simplify_binary_operation_1)
6496 (simplify_const_binary_operation, simplify_ternary_operation)
6497 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6498 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6499 instead of CONST_VECTOR_NUNITS.
6500 * varasm.c (output_constant_pool_2): Likewise.
6501 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6502 explicit-encoded elements in the XVEC for variable-length vectors.
6504 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6506 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6508 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6509 Alan Hayward <alan.hayward@arm.com>
6510 David Sherwood <david.sherwood@arm.com>
6512 * coretypes.h (fixed_size_mode): Declare.
6513 (fixed_size_mode_pod): New typedef.
6514 * builtins.h (target_builtins::x_apply_args_mode)
6515 (target_builtins::x_apply_result_mode): Change type to
6516 fixed_size_mode_pod.
6517 * builtins.c (apply_args_size, apply_result_size, result_vector)
6518 (expand_builtin_apply_args_1, expand_builtin_apply)
6519 (expand_builtin_return): Update accordingly.
6521 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6523 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6524 * cselib.c (cselib_hash_rtx): Likewise.
6525 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6526 CONST_VECTOR encoding.
6528 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6529 Jeff Law <law@redhat.com>
6532 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6533 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6534 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6535 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6538 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6539 explicitly probe *sp in a noreturn function if there were any callee
6540 register saves or frame pointer is needed.
6542 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6545 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6546 BLKmode for ternary, binary or unary expressions.
6549 * var-tracking.c (delete_vta_debug_insn): New inline function.
6550 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6551 insns from get_insns () to NULL instead of each bb separately.
6552 Use delete_vta_debug_insn. No longer static.
6553 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6554 delete_vta_debug_insns callers.
6555 * rtl.h (delete_vta_debug_insns): Declare.
6556 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6557 instead of variable_tracking_main.
6559 2018-01-03 Martin Sebor <msebor@redhat.com>
6561 PR tree-optimization/83603
6562 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6563 arguments past the endof the argument list in functions declared
6564 without a prototype.
6565 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6566 Avoid checking when arguments are null.
6568 2018-01-03 Martin Sebor <msebor@redhat.com>
6571 * doc/extend.texi (attribute const): Fix a typo.
6572 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6573 issuing -Wsuggest-attribute for void functions.
6575 2018-01-03 Martin Sebor <msebor@redhat.com>
6577 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6578 offset_int::from instead of wide_int::to_shwi.
6579 (maybe_diag_overlap): Remove assertion.
6580 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6581 * gimple-ssa-sprintf.c (format_directive): Same.
6582 (parse_directive): Same.
6583 (sprintf_dom_walker::compute_format_length): Same.
6584 (try_substitute_return_value): Same.
6586 2018-01-03 Jeff Law <law@redhat.com>
6589 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6590 non-constant residual for zero at runtime and avoid probing in
6591 that case. Reorganize code for trailing problem to mirror handling
6594 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6596 PR tree-optimization/83501
6597 * tree-ssa-strlen.c (get_string_cst): New.
6598 (handle_char_store): Call get_string_cst.
6600 2018-01-03 Martin Liska <mliska@suse.cz>
6602 PR tree-optimization/83593
6603 * tree-ssa-strlen.c: Include tree-cfg.h.
6604 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6605 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6606 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6608 (strlen_dom_walker::before_dom_children): Call
6609 gimple_purge_dead_eh_edges. Dump tranformation with details
6611 (strlen_dom_walker::before_dom_children): Update call by adding
6612 new argument cleanup_eh.
6613 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6615 2018-01-03 Martin Liska <mliska@suse.cz>
6618 * cif-code.def (VARIADIC_THUNK): New enum value.
6619 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6622 2018-01-03 Jan Beulich <jbeulich@suse.com>
6624 * sse.md (mov<mode>_internal): Tighten condition for when to use
6625 vmovdqu<ssescalarsize> for TI and OI modes.
6627 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6629 Update copyright years.
6631 2018-01-03 Martin Liska <mliska@suse.cz>
6634 * ipa-visibility.c (function_and_variable_visibility): Skip
6635 functions with noipa attribure.
6637 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6639 * gcc.c (process_command): Update copyright notice dates.
6640 * gcov-dump.c (print_version): Ditto.
6641 * gcov.c (print_version): Ditto.
6642 * gcov-tool.c (print_version): Ditto.
6643 * gengtype.c (create_file): Ditto.
6644 * doc/cpp.texi: Bump @copying's copyright year.
6645 * doc/cppinternals.texi: Ditto.
6646 * doc/gcc.texi: Ditto.
6647 * doc/gccint.texi: Ditto.
6648 * doc/gcov.texi: Ditto.
6649 * doc/install.texi: Ditto.
6650 * doc/invoke.texi: Ditto.
6652 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6654 * vector-builder.h (vector_builder::m_full_nelts): Change from
6655 unsigned int to poly_uint64.
6656 (vector_builder::full_nelts): Update prototype accordingly.
6657 (vector_builder::new_vector): Likewise.
6658 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6659 (vector_builder::operator ==): Likewise.
6660 (vector_builder::finalize): Likewise.
6661 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6662 Take the number of elements as a poly_uint64 rather than an
6664 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6665 from unsigned int to poly_uint64.
6666 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6667 (vec_perm_indices::new_vector): Likewise.
6668 (vec_perm_indices::length): Likewise.
6669 (vec_perm_indices::nelts_per_input): Likewise.
6670 (vec_perm_indices::input_nelts): Likewise.
6671 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6672 number of elements per input as a poly_uint64 rather than an
6673 unsigned int. Use the original encoding for variable-length
6674 vectors, rather than clamping each individual element.
6675 For the second and subsequent elements in each pattern,
6676 clamp the step and base before clamping their sum.
6677 (vec_perm_indices::series_p): Handle polynomial element counts.
6678 (vec_perm_indices::all_in_range_p): Likewise.
6679 (vec_perm_indices_to_tree): Likewise.
6680 (vec_perm_indices_to_rtx): Likewise.
6681 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6682 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6683 (tree_vector_builder::new_binary_operation): Handle polynomial
6684 element counts. Return false if we need to know the number
6685 of elements at compile time.
6686 * fold-const.c (fold_vec_perm): Punt if the number of elements
6687 isn't known at compile time.
6689 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6691 * vec-perm-indices.h (vec_perm_builder): Change element type
6692 from HOST_WIDE_INT to poly_int64.
6693 (vec_perm_indices::element_type): Update accordingly.
6694 (vec_perm_indices::clamp): Handle polynomial element_types.
6695 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6696 (vec_perm_indices::all_in_range_p): Likewise.
6697 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6699 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6700 polynomial vec_perm_indices element types.
6701 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6702 * fold-const.c (fold_vec_perm): Likewise.
6703 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6704 * tree-vect-generic.c (lower_vec_perm): Likewise.
6705 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6706 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6707 element type to HOST_WIDE_INT.
6709 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6710 Alan Hayward <alan.hayward@arm.com>
6711 David Sherwood <david.sherwood@arm.com>
6713 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6714 rather than an int. Use plus_constant.
6715 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6716 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6718 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6719 Alan Hayward <alan.hayward@arm.com>
6720 David Sherwood <david.sherwood@arm.com>
6722 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6723 a HOST_WIDE_INT to a poly_int64.
6725 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6726 Alan Hayward <alan.hayward@arm.com>
6727 David Sherwood <david.sherwood@arm.com>
6729 * calls.c (load_register_parameters): Cope with polynomial
6730 mode sizes. Require a constant size for BLKmode parameters
6731 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6732 forces a parameter to be padded at the lsb end in order to
6733 fill a complete number of words, require the parameter size
6734 to be ordered wrt UNITS_PER_WORD.
6736 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6737 Alan Hayward <alan.hayward@arm.com>
6738 David Sherwood <david.sherwood@arm.com>
6740 * reload1.c (spill_stack_slot_width): Change element type
6741 from unsigned int to poly_uint64_pod.
6742 (alter_reg): Treat mode sizes as polynomial.
6744 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6745 Alan Hayward <alan.hayward@arm.com>
6746 David Sherwood <david.sherwood@arm.com>
6748 * reload.c (complex_word_subreg_p): New function.
6749 (reload_inner_reg_of_subreg, push_reload): Use it.
6751 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6752 Alan Hayward <alan.hayward@arm.com>
6753 David Sherwood <david.sherwood@arm.com>
6755 * lra-constraints.c (process_alt_operands): Reject matched
6756 operands whose sizes aren't ordered.
6757 (match_reload): Refer to this check here.
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 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6764 that the mode size is in the set {1, 2, 4, 8, 16}.
6766 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6767 Alan Hayward <alan.hayward@arm.com>
6768 David Sherwood <david.sherwood@arm.com>
6770 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6771 Use plus_constant instead of gen_rtx_PLUS.
6773 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6774 Alan Hayward <alan.hayward@arm.com>
6775 David Sherwood <david.sherwood@arm.com>
6777 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6778 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6779 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6780 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6781 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6782 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6783 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6784 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6785 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6786 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6788 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6789 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6790 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6791 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6792 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6793 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6794 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6795 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6796 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6797 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6799 * expr.c (emit_move_resolve_push): Treat the input and result
6800 of PUSH_ROUNDING as a poly_int64.
6801 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6802 (emit_push_insn): Likewise.
6803 * lra-eliminations.c (mark_not_eliminable): Likewise.
6804 * recog.c (push_operand): Likewise.
6805 * reload1.c (elimination_effects): Likewise.
6806 * rtlanal.c (nonzero_bits1): Likewise.
6807 * calls.c (store_one_arg): Likewise. Require the padding to be
6808 known at compile time.
6810 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6811 Alan Hayward <alan.hayward@arm.com>
6812 David Sherwood <david.sherwood@arm.com>
6814 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6815 Use plus_constant instead of gen_rtx_PLUS.
6817 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6818 Alan Hayward <alan.hayward@arm.com>
6819 David Sherwood <david.sherwood@arm.com>
6821 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
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 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6829 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6830 via stack temporaries. Treat the mode size as polynomial too.
6832 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6833 Alan Hayward <alan.hayward@arm.com>
6834 David Sherwood <david.sherwood@arm.com>
6836 * expr.c (expand_expr_real_2): When handling conversions involving
6837 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6838 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6839 as a poly_uint64 too.
6841 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6842 Alan Hayward <alan.hayward@arm.com>
6843 David Sherwood <david.sherwood@arm.com>
6845 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
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 * combine.c (can_change_dest_mode): Handle polynomial
6852 REGMODE_NATURAL_SIZE.
6853 * expmed.c (store_bit_field_1): Likewise.
6854 * expr.c (store_constructor): Likewise.
6855 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6856 and polynomial REGMODE_NATURAL_SIZE.
6857 (gen_lowpart_common): Likewise.
6858 * reginfo.c (record_subregs_of_mode): Likewise.
6859 * rtlanal.c (read_modify_subreg_p): Likewise.
6861 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6862 Alan Hayward <alan.hayward@arm.com>
6863 David Sherwood <david.sherwood@arm.com>
6865 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6866 numbers of elements.
6868 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6869 Alan Hayward <alan.hayward@arm.com>
6870 David Sherwood <david.sherwood@arm.com>
6872 * match.pd: Cope with polynomial numbers of vector elements.
6874 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6875 Alan Hayward <alan.hayward@arm.com>
6876 David Sherwood <david.sherwood@arm.com>
6878 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6879 in a POINTER_PLUS_EXPR.
6881 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6882 Alan Hayward <alan.hayward@arm.com>
6883 David Sherwood <david.sherwood@arm.com>
6885 * omp-simd-clone.c (simd_clone_subparts): New function.
6886 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6887 (ipa_simd_modify_function_body): Likewise.
6889 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6890 Alan Hayward <alan.hayward@arm.com>
6891 David Sherwood <david.sherwood@arm.com>
6893 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6894 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6895 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6896 (expand_vector_condition, vector_element): Likewise.
6897 (subparts_gt): New function.
6898 (get_compute_type): Use subparts_gt.
6899 (count_type_subparts): Delete.
6900 (expand_vector_operations_1): Use subparts_gt instead of
6901 count_type_subparts.
6903 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6904 Alan Hayward <alan.hayward@arm.com>
6905 David Sherwood <david.sherwood@arm.com>
6907 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6908 (vect_compile_time_alias): ...this new function. Do the calculation
6909 on poly_ints rather than trees.
6910 (vect_prune_runtime_alias_test_list): Update call accordingly.
6912 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6913 Alan Hayward <alan.hayward@arm.com>
6914 David Sherwood <david.sherwood@arm.com>
6916 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6918 (vect_schedule_slp_instance): Likewise.
6920 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6921 Alan Hayward <alan.hayward@arm.com>
6922 David Sherwood <david.sherwood@arm.com>
6924 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6925 constant and extern definitions for variable-length vectors.
6926 (vect_get_constant_vectors): Note that the number of units
6927 is known to be constant.
6929 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6930 Alan Hayward <alan.hayward@arm.com>
6931 David Sherwood <david.sherwood@arm.com>
6933 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6934 of units as polynomial. Choose between WIDE and NARROW based
6937 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6938 Alan Hayward <alan.hayward@arm.com>
6939 David Sherwood <david.sherwood@arm.com>
6941 * tree-vect-stmts.c (simd_clone_subparts): New function.
6942 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6944 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6945 Alan Hayward <alan.hayward@arm.com>
6946 David Sherwood <david.sherwood@arm.com>
6948 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6949 vectors as polynomial. Use build_index_vector for
6952 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6953 Alan Hayward <alan.hayward@arm.com>
6954 David Sherwood <david.sherwood@arm.com>
6956 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6957 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6958 for variable-length vectors.
6959 (vectorizable_mask_load_store): Treat the number of units as
6960 polynomial, asserting that it is constant if the condition has
6961 already been enforced.
6962 (vectorizable_store, vectorizable_load): Likewise.
6964 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6965 Alan Hayward <alan.hayward@arm.com>
6966 David Sherwood <david.sherwood@arm.com>
6968 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6969 of units as polynomial. Punt if we can't tell at compile time
6970 which vector contains the final result.
6972 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6973 Alan Hayward <alan.hayward@arm.com>
6974 David Sherwood <david.sherwood@arm.com>
6976 * tree-vect-loop.c (vectorizable_induction): Treat the number
6977 of units as polynomial. Punt on SLP inductions. Use an integer
6978 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6979 cast of such a series for variable-length floating-point
6982 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6983 Alan Hayward <alan.hayward@arm.com>
6984 David Sherwood <david.sherwood@arm.com>
6986 * tree.h (build_index_vector): Declare.
6987 * tree.c (build_index_vector): New function.
6988 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6989 of units as polynomial, forcibly converting it to a constant if
6990 vectorizable_reduction has already enforced the condition.
6991 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6992 to create a {1,2,3,...} vector.
6993 (vectorizable_reduction): Treat the number of units as polynomial.
6994 Choose vectype_in based on the largest scalar element size rather
6995 than the smallest number of units. Enforce the restrictions
6998 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6999 Alan Hayward <alan.hayward@arm.com>
7000 David Sherwood <david.sherwood@arm.com>
7002 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
7003 number of units as polynomial.
7005 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7006 Alan Hayward <alan.hayward@arm.com>
7007 David Sherwood <david.sherwood@arm.com>
7009 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
7010 * target.def (autovectorize_vector_sizes): Return the vector sizes
7011 by pointer, using vector_sizes rather than a bitmask.
7012 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
7013 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
7014 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
7016 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
7017 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
7018 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
7019 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
7020 * omp-general.c (omp_max_vf): Likewise.
7021 * omp-low.c (omp_clause_aligned_alignment): Likewise.
7022 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
7023 * tree-vect-loop.c (vect_analyze_loop): Likewise.
7024 * tree-vect-slp.c (vect_slp_bb): Likewise.
7025 * doc/tm.texi: Regenerate.
7026 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
7028 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
7029 the vector size as a poly_uint64 rather than an unsigned int.
7030 (current_vector_size): Change from an unsigned int to a poly_uint64.
7031 (get_vectype_for_scalar_type): Update accordingly.
7032 * tree.h (build_truth_vector_type): Take the size and number of
7033 units as a poly_uint64 rather than an unsigned int.
7034 (build_vector_type): Add a temporary overload that takes
7035 the number of units as a poly_uint64 rather than an unsigned int.
7036 * tree.c (make_vector_type): Likewise.
7037 (build_truth_vector_type): Take the number of units as a poly_uint64
7038 rather than an unsigned int.
7040 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7041 Alan Hayward <alan.hayward@arm.com>
7042 David Sherwood <david.sherwood@arm.com>
7044 * target.def (get_mask_mode): Take the number of units and length
7045 as poly_uint64s rather than unsigned ints.
7046 * targhooks.h (default_get_mask_mode): Update accordingly.
7047 * targhooks.c (default_get_mask_mode): Likewise.
7048 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
7049 * doc/tm.texi: Regenerate.
7051 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7052 Alan Hayward <alan.hayward@arm.com>
7053 David Sherwood <david.sherwood@arm.com>
7055 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
7056 * omp-general.c (omp_max_vf): Likewise.
7057 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
7058 (expand_omp_simd): Handle polynomial safelen.
7059 * omp-low.c (omplow_simd_context): Add a default constructor.
7060 (omplow_simd_context::max_vf): Change from int to poly_uint64.
7061 (lower_rec_simd_input_clauses): Update accordingly.
7062 (lower_rec_input_clauses): Likewise.
7064 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7065 Alan Hayward <alan.hayward@arm.com>
7066 David Sherwood <david.sherwood@arm.com>
7068 * tree-vectorizer.h (vect_nunits_for_cost): New function.
7069 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
7070 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
7071 (vect_analyze_slp_cost): Likewise.
7072 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
7073 (vect_model_load_cost): Likewise.
7075 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7076 Alan Hayward <alan.hayward@arm.com>
7077 David Sherwood <david.sherwood@arm.com>
7079 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
7080 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
7081 from an unsigned int * to a poly_uint64_pod *.
7082 (calculate_unrolling_factor): New function.
7083 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
7085 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7086 Alan Hayward <alan.hayward@arm.com>
7087 David Sherwood <david.sherwood@arm.com>
7089 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
7090 from an unsigned int to a poly_uint64.
7091 (_loop_vec_info::slp_unrolling_factor): Likewise.
7092 (_loop_vec_info::vectorization_factor): Change from an int
7094 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
7095 (vect_get_num_vectors): New function.
7096 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
7097 (vect_get_num_copies): Use vect_get_num_vectors.
7098 (vect_analyze_data_ref_dependences): Change max_vf from an int *
7099 to an unsigned int *.
7100 (vect_analyze_data_refs): Change min_vf from an int * to a
7102 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7103 than an unsigned HOST_WIDE_INT.
7104 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
7105 (vect_analyze_data_ref_dependence): Change max_vf from an int *
7106 to an unsigned int *.
7107 (vect_analyze_data_ref_dependences): Likewise.
7108 (vect_compute_data_ref_alignment): Handle polynomial vf.
7109 (vect_enhance_data_refs_alignment): Likewise.
7110 (vect_prune_runtime_alias_test_list): Likewise.
7111 (vect_shift_permute_load_chain): Likewise.
7112 (vect_supportable_dr_alignment): Likewise.
7113 (dependence_distance_ge_vf): Take the vectorization factor as a
7114 poly_uint64 rather than an unsigned HOST_WIDE_INT.
7115 (vect_analyze_data_refs): Change min_vf from an int * to a
7117 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
7118 vfm1 as a poly_uint64 rather than an int. Make the same change
7119 for the returned bound_scalar.
7120 (vect_gen_vector_loop_niters): Handle polynomial vf.
7121 (vect_do_peeling): Likewise. Update call to
7122 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
7123 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
7125 * tree-vect-loop.c (vect_determine_vectorization_factor)
7126 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
7127 (vect_get_known_peeling_cost): Likewise.
7128 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
7129 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
7130 (vect_transform_loop): Likewise. Use the lowest possible VF when
7131 updating the upper bounds of the loop.
7132 (vect_min_worthwhile_factor): Make static. Return an unsigned int
7134 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
7135 polynomial unroll factors.
7136 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
7137 (vect_make_slp_decision): Likewise.
7138 (vect_supported_load_permutation_p): Likewise, and polynomial
7140 (vect_analyze_slp_cost): Handle polynomial vf.
7141 (vect_slp_analyze_node_operations): Likewise.
7142 (vect_slp_analyze_bb_1): Likewise.
7143 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7144 than an unsigned HOST_WIDE_INT.
7145 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
7146 (vectorizable_load): Handle polynomial vf.
7147 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
7149 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
7151 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7152 Alan Hayward <alan.hayward@arm.com>
7153 David Sherwood <david.sherwood@arm.com>
7155 * match.pd: Handle bit operations involving three constants
7156 and try to fold one pair.
7158 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7160 * tree-vect-loop-manip.c: Include gimple-fold.h.
7161 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
7162 niters_maybe_zero parameters. Handle other cases besides a step of 1.
7163 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
7164 Add a path that uses a step of VF instead of 1, but disable it
7166 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
7167 and niters_no_overflow parameters. Update calls to
7168 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
7169 Create a new SSA name if the latter choses to use a ste other
7170 than zero, and return it via niters_vector_mult_vf_var.
7171 * tree-vect-loop.c (vect_transform_loop): Update calls to
7172 vect_do_peeling, vect_gen_vector_loop_niters and
7173 slpeel_make_loop_iterate_ntimes.
7174 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
7175 (vect_gen_vector_loop_niters): Update declarations after above changes.
7177 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
7179 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
7180 128-bit round to integer instructions.
7181 (ceil<mode>2): Likewise.
7182 (btrunc<mode>2): Likewise.
7183 (round<mode>2): Likewise.
7185 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7187 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
7188 unaligned VSX load/store on P8/P9.
7189 (expand_block_clear): Allow the use of unaligned VSX
7190 load/store on P8/P9.
7192 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
7194 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
7196 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
7197 swap associated with both a load and a store.
7199 2018-01-02 Andrew Waterman <andrew@sifive.com>
7201 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
7202 * config/riscv/riscv.md (clear_cache): Use it.
7204 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
7206 * web.c: Remove out-of-date comment.
7208 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7210 * expr.c (fixup_args_size_notes): Check that any existing
7211 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
7212 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
7213 (emit_single_push_insn): ...here.
7215 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7217 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
7218 (const_vector_encoded_nelts): New function.
7219 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
7220 (const_vector_int_elt, const_vector_elt): Declare.
7221 * emit-rtl.c (const_vector_int_elt_1): New function.
7222 (const_vector_elt): Likewise.
7223 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
7224 of CONST_VECTOR_ELT.
7226 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7228 * expr.c: Include rtx-vector-builder.h.
7229 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
7230 directly on the tree encoding.
7231 (const_vector_from_tree): Likewise.
7232 * optabs.c: Include rtx-vector-builder.h.
7233 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
7234 sequence of "u" values.
7235 * vec-perm-indices.c: Include rtx-vector-builder.h.
7236 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
7237 directly on the vec_perm_indices encoding.
7239 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7241 * doc/rtl.texi (const_vector): Describe new encoding scheme.
7242 * Makefile.in (OBJS): Add rtx-vector-builder.o.
7243 * rtx-vector-builder.h: New file.
7244 * rtx-vector-builder.c: Likewise.
7245 * rtl.h (rtx_def::u2): Add a const_vector field.
7246 (CONST_VECTOR_NPATTERNS): New macro.
7247 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
7248 (CONST_VECTOR_DUPLICATE_P): Likewise.
7249 (CONST_VECTOR_STEPPED_P): Likewise.
7250 (CONST_VECTOR_ENCODED_ELT): Likewise.
7251 (const_vec_duplicate_p): Check for a duplicated vector encoding.
7252 (unwrap_const_vec_duplicate): Likewise.
7253 (const_vec_series_p): Check for a non-duplicated vector encoding.
7254 Say that the function only returns true for integer vectors.
7255 * emit-rtl.c: Include rtx-vector-builder.h.
7256 (gen_const_vec_duplicate_1): Delete.
7257 (gen_const_vector): Call gen_const_vec_duplicate instead of
7258 gen_const_vec_duplicate_1.
7259 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
7260 (gen_const_vec_duplicate): Use rtx_vector_builder.
7261 (gen_const_vec_series): Likewise.
7262 (gen_rtx_CONST_VECTOR): Likewise.
7263 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
7264 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7265 Build a new vector rather than modifying a CONST_VECTOR in-place.
7266 (handle_special_swappables): Update call accordingly.
7267 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
7268 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7269 Build a new vector rather than modifying a CONST_VECTOR in-place.
7270 (handle_special_swappables): Update call accordingly.
7272 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7274 * simplify-rtx.c (simplify_const_binary_operation): Use
7275 CONST_VECTOR_ELT instead of XVECEXP.
7277 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7279 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
7280 the selector elements to be different from the data elements
7281 if the selector is a VECTOR_CST.
7282 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
7283 ssizetype for the selector.
7285 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7287 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
7288 before testing each element individually.
7289 * tree-vect-generic.c (lower_vec_perm): Likewise.
7291 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7293 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
7294 * selftest-run-tests.c (selftest::run_tests): Call it.
7295 * vector-builder.h (vector_builder::operator ==): New function.
7296 (vector_builder::operator !=): Likewise.
7297 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
7298 (vec_perm_indices::all_from_input_p): New function.
7299 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7300 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
7301 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
7302 instead of reading the VECTOR_CST directly. Detect whether both
7303 vector inputs are the same before constructing the vec_perm_indices,
7304 and update the number of inputs argument accordingly. Use the
7305 utility functions added above. Only construct sel2 if we need to.
7307 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7309 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
7310 the broadcast of the low byte.
7311 (expand_mult_highpart): Use an explicit encoding for the permutes.
7312 * optabs-query.c (can_mult_highpart_p): Likewise.
7313 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
7314 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7315 (vectorizable_bswap): Likewise.
7316 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
7317 explicit encoding for the power-of-2 permutes.
7318 (vect_permute_store_chain): Likewise.
7319 (vect_grouped_load_supported): Likewise.
7320 (vect_permute_load_chain): Likewise.
7322 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7324 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
7325 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
7326 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
7327 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7328 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
7329 (vect_gen_perm_mask_any): Likewise.
7331 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7333 * int-vector-builder.h: New file.
7334 * vec-perm-indices.h: Include int-vector-builder.h.
7335 (vec_perm_indices): Redefine as an int_vector_builder.
7336 (auto_vec_perm_indices): Delete.
7337 (vec_perm_builder): Redefine as a stand-alone class.
7338 (vec_perm_indices::vec_perm_indices): New function.
7339 (vec_perm_indices::clamp): Likewise.
7340 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
7341 (vec_perm_indices::new_vector): New function.
7342 (vec_perm_indices::new_expanded_vector): Update for new
7343 vec_perm_indices class.
7344 (vec_perm_indices::rotate_inputs): New function.
7345 (vec_perm_indices::all_in_range_p): Operate directly on the
7346 encoded form, without computing elided elements.
7347 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
7348 encoding. Update for new vec_perm_indices class.
7349 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
7350 the given vec_perm_builder.
7351 (expand_vec_perm_var): Update vec_perm_builder constructor.
7352 (expand_mult_highpart): Use vec_perm_builder instead of
7353 auto_vec_perm_indices.
7354 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
7355 vec_perm_indices instead of auto_vec_perm_indices. Use a single
7356 or double series encoding as appropriate.
7357 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
7358 vec_perm_indices instead of auto_vec_perm_indices.
7359 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7360 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7361 (vect_permute_store_chain): Likewise.
7362 (vect_grouped_load_supported): Likewise.
7363 (vect_permute_load_chain): Likewise.
7364 (vect_shift_permute_load_chain): Likewise.
7365 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7366 (vect_transform_slp_perm_load): Likewise.
7367 (vect_schedule_slp_instance): Likewise.
7368 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7369 (vectorizable_mask_load_store): Likewise.
7370 (vectorizable_bswap): Likewise.
7371 (vectorizable_store): Likewise.
7372 (vectorizable_load): Likewise.
7373 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
7374 vec_perm_indices instead of auto_vec_perm_indices. Use
7375 tree_to_vec_perm_builder to read the vector from a tree.
7376 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
7377 vec_perm_builder instead of a vec_perm_indices.
7378 (have_whole_vector_shift): Use vec_perm_builder and
7379 vec_perm_indices instead of auto_vec_perm_indices. Leave the
7380 truncation to calc_vec_perm_mask_for_shift.
7381 (vect_create_epilog_for_reduction): Likewise.
7382 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
7383 from auto_vec_perm_indices to vec_perm_indices.
7384 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7385 instead of changing individual elements.
7386 (aarch64_vectorize_vec_perm_const): Use new_vector to install
7387 the vector in d.perm.
7388 * config/arm/arm.c (expand_vec_perm_d::perm): Change
7389 from auto_vec_perm_indices to vec_perm_indices.
7390 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7391 instead of changing individual elements.
7392 (arm_vectorize_vec_perm_const): Use new_vector to install
7393 the vector in d.perm.
7394 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
7395 Update vec_perm_builder constructor.
7396 (rs6000_expand_interleave): Likewise.
7397 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
7398 (rs6000_expand_interleave): Likewise.
7400 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7402 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
7403 to qimode could truncate the indices.
7404 * optabs.c (expand_vec_perm_var): Likewise.
7406 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7408 * Makefile.in (OBJS): Add vec-perm-indices.o.
7409 * vec-perm-indices.h: New file.
7410 * vec-perm-indices.c: Likewise.
7411 * target.h (vec_perm_indices): Replace with a forward class
7413 (auto_vec_perm_indices): Move to vec-perm-indices.h.
7414 * optabs.h: Include vec-perm-indices.h.
7415 (expand_vec_perm): Delete.
7416 (selector_fits_mode_p, expand_vec_perm_var): Declare.
7417 (expand_vec_perm_const): Declare.
7418 * target.def (vec_perm_const_ok): Replace with...
7419 (vec_perm_const): ...this new hook.
7420 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
7421 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
7422 * doc/tm.texi: Regenerate.
7423 * optabs.def (vec_perm_const): Delete.
7424 * doc/md.texi (vec_perm_const): Likewise.
7425 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
7426 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7427 expand_vec_perm for constant permutation vectors. Assert that
7428 the mode of variable permutation vectors is the integer equivalent
7429 of the mode that is being permuted.
7430 * optabs-query.h (selector_fits_mode_p): Declare.
7431 * optabs-query.c: Include vec-perm-indices.h.
7432 (selector_fits_mode_p): New function.
7433 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7434 is defined, instead of checking whether the vec_perm_const_optab
7435 exists. Use targetm.vectorize.vec_perm_const instead of
7436 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7437 fit in the vector mode before using a variable permute.
7438 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7439 vec_perm_indices instead of an rtx.
7440 (expand_vec_perm): Replace with...
7441 (expand_vec_perm_const): ...this new function. Take the selector
7442 as a vec_perm_indices rather than an rtx. Also take the mode of
7443 the selector. Update call to shift_amt_for_vec_perm_mask.
7444 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7445 Use vec_perm_indices::new_expanded_vector to expand the original
7446 selector into bytes. Check whether the indices fit in the vector
7447 mode before using a variable permute.
7448 (expand_vec_perm_var): Make global.
7449 (expand_mult_highpart): Use expand_vec_perm_const.
7450 * fold-const.c: Includes vec-perm-indices.h.
7451 * tree-ssa-forwprop.c: Likewise.
7452 * tree-vect-data-refs.c: Likewise.
7453 * tree-vect-generic.c: Likewise.
7454 * tree-vect-loop.c: Likewise.
7455 * tree-vect-slp.c: Likewise.
7456 * tree-vect-stmts.c: Likewise.
7457 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7459 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7460 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7461 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7462 (aarch64_vectorize_vec_perm_const): ...this new function.
7463 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7464 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7465 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7466 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7467 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7468 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7469 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7471 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7472 check for NEON modes.
7473 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7474 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7475 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7476 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7478 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7479 the old VEC_PERM_CONST conditions.
7480 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7481 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7482 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7483 (ia64_vectorize_vec_perm_const_ok): Merge into...
7484 (ia64_vectorize_vec_perm_const): ...this new function.
7485 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7486 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7487 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7488 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7489 * config/mips/mips.c (mips_expand_vec_perm_const)
7490 (mips_vectorize_vec_perm_const_ok): Merge into...
7491 (mips_vectorize_vec_perm_const): ...this new function.
7492 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7493 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7494 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7495 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7496 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7497 (rs6000_expand_vec_perm_const): Delete.
7498 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7500 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7501 (altivec_expand_vec_perm_const_le): Take each operand individually.
7502 Operate on constant selectors rather than rtxes.
7503 (altivec_expand_vec_perm_const): Likewise. Update call to
7504 altivec_expand_vec_perm_const_le.
7505 (rs6000_expand_vec_perm_const): Delete.
7506 (rs6000_vectorize_vec_perm_const_ok): Delete.
7507 (rs6000_vectorize_vec_perm_const): New function.
7508 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7509 an element count and rtx array.
7510 (rs6000_expand_extract_even): Update call accordingly.
7511 (rs6000_expand_interleave): Likewise.
7512 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7513 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7514 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7515 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7516 (rs6000_expand_vec_perm_const): Delete.
7517 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7518 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7519 (altivec_expand_vec_perm_const_le): Take each operand individually.
7520 Operate on constant selectors rather than rtxes.
7521 (altivec_expand_vec_perm_const): Likewise. Update call to
7522 altivec_expand_vec_perm_const_le.
7523 (rs6000_expand_vec_perm_const): Delete.
7524 (rs6000_vectorize_vec_perm_const_ok): Delete.
7525 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7526 reference to the SPE evmerge intructions.
7527 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7528 an element count and rtx array.
7529 (rs6000_expand_extract_even): Update call accordingly.
7530 (rs6000_expand_interleave): Likewise.
7531 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7532 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7534 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7536 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7538 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7539 vector mode and that that mode matches the mode of the data
7541 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7542 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7543 directly using expand_vec_perm_1 when forcing selectors into
7545 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7547 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7549 * optabs-query.h (can_vec_perm_p): Delete.
7550 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7551 * optabs-query.c (can_vec_perm_p): Split into...
7552 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7553 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7554 particular selector is valid.
7555 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7556 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7557 (vect_grouped_load_supported): Likewise.
7558 (vect_shift_permute_load_chain): Likewise.
7559 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7560 (vect_transform_slp_perm_load): Likewise.
7561 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7562 (vectorizable_bswap): Likewise.
7563 (vect_gen_perm_mask_checked): Likewise.
7564 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7565 implementations of variable permutation vectors into account
7566 when deciding which selector to use.
7567 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7568 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7569 with a false third argument.
7570 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7571 to test whether the constant selector is valid and can_vec_perm_var_p
7572 to test whether a variable selector is valid.
7574 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7576 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7577 * optabs-query.c (can_vec_perm_p): Likewise.
7578 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7579 instead of vec_perm_indices.
7580 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7581 (vect_gen_perm_mask_checked): Likewise,
7582 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7583 (vect_gen_perm_mask_checked): Likewise,
7585 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7587 * optabs-query.h (qimode_for_vec_perm): Declare.
7588 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7589 (qimode_for_vec_perm): ...this new function.
7590 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7592 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7594 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7595 does not have a conditional at the top.
7597 2018-01-02 Richard Biener <rguenther@suse.de>
7599 * ipa-inline.c (big_speedup_p): Fix expression.
7601 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7604 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7607 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7611 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7612 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7613 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7614 cond_taken_branch_cost 3->4.
7616 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7618 PR tree-optimization/83581
7619 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7620 TODO_cleanup_cfg if any changes have been made.
7623 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7624 convert_modes if target mode has the right side, but different mode
7628 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7629 last argument when extracting from CONCAT. If either from_real or
7630 from_imag is NULL, use expansion through memory. If result is not
7631 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7632 the parts directly to inner mode, if even that fails, use expansion
7636 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7637 check for bswap in mode rather than HImode and use that in expand_unop
7640 Copyright (C) 2018 Free Software Foundation, Inc.
7642 Copying and distribution of this file, with or without modification,
7643 are permitted in any medium without royalty provided the copyright
7644 notice and this notice are preserved.