| blob | 8eedfcaaf82781b8f6acccbe8f0ba056da02cf47 |
1 2018-02-28 Alexandre Oliva <aoliva@redhat.com>
3 PR rtl-optimization/81611
4 * auto-inc-dec.c (attempt_change): Move dead note from
5 mem_insn if it's the next use of regno
6 (find_address): Take address use of reg holding
7 non-incremented value. Add parm to limit search to the named
8 reg only.
9 (merge_in_block): Attempt to use a mem insn that is the next
10 use of the original regno.
12 2018-02-27 Martin Sebor <msebor@redhat.com>
14 PR c++/83871
15 * gcc/doc/invoke.texi (-Wmissing-attributes): New option.
16 * gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.
18 2018-02-27 Martin Sebor <msebor@redhat.com>
20 PR translation/84207
21 * diagnostic-core.h (warning_n, error_n, inform_n): Change
22 n argument to unsigned HOST_WIDE_INT.
23 * diagnostic.c (warning_n, error_n, inform_n): Ditto.
24 (diagnostic_n_impl): Ditto. Handle arguments in excess of LONG_MAX.
25 * gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
26 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.
28 2018-02-27 Richard Biener <rguenther@suse.de>
30 PR tree-optimization/84512
31 * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
32 Do not use the estimate returned from record_stmt_cost for
33 the scalar iteration cost but sum properly using add_stmt_cost.
35 2018-02-27 Richard Biener <rguenther@suse.de>
37 PR tree-optimization/84466
38 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
39 Adjust last change to less strictly validate use operands.
41 2018-02-27 Martin Liska <mliska@suse.cz>
43 PR gcov-profile/84548
44 * gcov.c (process_file): Allow partial overlap and consider it
45 also as group functions.
46 (output_lines): Properly calculate range of lines for a group.
48 2018-02-27 Martin Liska <mliska@suse.cz>
50 * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
51 'ggc' suffixes. Change first column width.
52 (timer::print): Fix formatting of the column.
54 2018-02-27 Alexandre Oliva <aoliva@redhat.com>
56 * tree-ssa-live.c (remove_unused_scope_block_p): Do not
57 preserve inline entry blocks for the sake of debug inline
58 entry point markers alone.
59 (remove_unused_locals): Suggest in comments a better place to
60 force the preservation of inline entry blocks that are
61 otherwise unused, but do not preserve them.
63 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
65 * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
67 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
69 PR target/84039
70 * config/i386/constraints.md (Bs): Replace
71 ix86_indirect_branch_register with
72 TARGET_INDIRECT_BRANCH_REGISTER.
73 (Bw): Likewise.
74 * config/i386/i386.md (indirect_jump): Likewise.
75 (tablejump): Likewise.
76 (*sibcall_memory): Likewise.
77 (*sibcall_value_memory): Likewise.
78 Peepholes of indirect call and jump via memory: Likewise.
79 (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
80 (*sibcall_value_GOT_32): Likewise.
81 * config/i386/i386.opt: Likewise.
82 * config/i386/predicates.md (indirect_branch_operand): Likewise.
83 (GOT_memory_operand): Likewise.
84 (call_insn_operand): Likewise.
85 (sibcall_insn_operand): Likewise.
86 (GOT32_symbol_operand): Likewise.
87 * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
89 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
91 PR rtl-optimization/83496
92 * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
93 booleans to RTXes. Call fix_reg_dead_note on every non-null element.
94 (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
95 redundant insn, if any.
96 (relax_delay_slots): Likewise.
97 (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
99 2018-02-26 Richard Sandiford <richard.sandiford@linaro.org>
101 PR tree-optimization/83965
102 * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
103 that grouped statements are part of a reduction chain. Return
104 true if the statement is not marked as a reduction itself but
105 is part of a group.
106 (vect_recog_dot_prod_pattern): Don't check whether the statement
107 is part of a group here.
108 (vect_recog_sad_pattern): Likewise.
109 (vect_recog_widen_sum_pattern): Likewise.
111 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
113 PR debug/84545
114 * final.c (rest_of_clean_state): Also look for calls inside sequences.
116 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
118 PR target/84530
119 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
120 the bool argument.
121 (ix86_output_indirect_function_return): New prototype.
122 (ix86_split_simple_return_pop_internal): Likewise.
123 * config/i386/i386.c (indirect_return_via_cx): New.
124 (indirect_return_via_cx_bnd): Likewise.
125 (indirect_thunk_name): Handle return va CX_REG.
126 (output_indirect_thunk_function): Create alias for
127 __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
128 (ix86_output_indirect_jmp): Remove the bool argument.
129 (ix86_output_indirect_function_return): New function.
130 (ix86_split_simple_return_pop_internal): Likewise.
131 * config/i386/i386.md (*indirect_jump): Don't pass false
132 to ix86_output_indirect_jmp.
133 (*tablejump_1): Likewise.
134 (simple_return_pop_internal): Change it to define_insn_and_split.
135 Call ix86_split_simple_return_pop_internal to split it for
136 -mfunction-return=.
137 (simple_return_indirect_internal): Call
138 ix86_output_indirect_function_return instead of
139 ix86_output_indirect_jmp.
141 2018-02-26 Jakub Jelinek <jakub@redhat.com>
143 PR bootstrap/84405
144 * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
145 memset and value initialization afterwards.
147 2018-02-26 Christophe Lyon <christophe.lyon@linaro.org>
149 * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
151 2018-02-26 Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
153 PR target/84521
154 * common/config/aarch64/aarch64-common.c
155 (aarch_option_optimization_table[]): Switch
156 off fomit-frame-pointer
158 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
159 Chung-Ju Wu <jasonwucj@gmail.com>
161 * config/nds32/nds32-multiple.md (load_multiple): Disallow
162 volatile memory.
163 (store_multiple): Ditto.
165 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
167 * config.gcc: Add --with-cpu support for nds32 target.
168 * config/nds32/nds32-opts.h (nds32_cpu_type): New.
169 * config/nds32/nds32.opt: Add -mcpu= option.
171 2018-02-25 Segher Boessenkool <segher@kernel.crashing.org>
173 * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
174 isel=yes): Warn for these deprecated options.
176 2018-02-23 David Edelsohn <dje.gcc@gmail.com>
178 * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
179 ISA_2_5_MASKS_EMBEDDED.
181 2018-02-23 Jakub Jelinek <jakub@redhat.com>
183 * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
184 p->max as pointers rather than using iterative_hash_expr.
186 2018-02-23 Carl Love <cel@us.ibm.com>
188 * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
189 macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
190 BU_P8V_OVERLOAD_2.
191 * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
192 P8V_BUILTIN_VEC_VSIGNED2. Change VSX_BUILTIN_VEC_VUNSIGNED2 to
193 P8V_BUILTIN_VEC_VUNSIGNED2.
195 2018-02-22 Vladimir Makarov <vmakarov@redhat.com>
197 PR target/81572
198 * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
199 * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
200 LRA_UNKNOWN_ALT.
201 * lra-constraints.c (curr_insn_transform): Set up
202 LRA_NON_CLOBBERED_ALT for moves processed on the fast path. Use
203 LRA_UNKNOWN_ALT.
204 (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
205 * lra-eliminations.c (spill_pseudos): Ditto.
206 (process_insn_for_elimination): Ditto.
207 * lra-lives.c (reg_early_clobber_p): Use the new macros.
208 * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
209 LRA_NON_CLOBBERED_ALT.
211 2018-02-22 Martin Sebor <msebor@redhat.com>
213 PR tree-optimization/84480
214 * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
215 to maybe_diag_stxncpy_trunc. Call it.
216 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
217 from gimple_fold_builtin_strcpy. Print inlining stack.
218 (handle_builtin_stxncpy): Print inlining stack.
219 * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
221 2018-02-22 H.J. Lu <hongjiu.lu@intel.com>
223 PR target/84176
224 * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
225 error when -mindirect-branch=thunk-extern, -fcf-protection=branch
226 and -fcheck-pointer-bounds are used together.
227 (indirect_thunk_prefix): New enum.
228 (indirect_thunk_need_prefix): New function.
229 (indirect_thunk_name): Replace need_bnd_p with need_prefix. Use
230 "_nt" instead of "_bnd" for NOTRACK prefix.
231 (output_indirect_thunk): Replace need_bnd_p with need_prefix.
232 (output_indirect_thunk_function): Likewise.
233 (): Likewise.
234 (ix86_code_end): Update output_indirect_thunk_function calls.
235 (ix86_output_indirect_branch_via_reg): Replace
236 ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
237 (ix86_output_indirect_branch_via_push): Likewise.
238 (ix86_output_function_return): Likewise.
239 * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
240 incompatible with -fcf-protection=branch and
241 -fcheck-pointer-bounds.
243 2018-02-22 Steve Ellcey <sellcey@cavium.com>
245 PR target/83335
246 * config/aarch64/aarch64.c (aarch64_print_address_internal):
247 Change gcc_assert call to output_operand_lossage.
249 2018-02-22 Steve Ellcey <sellcey@cavium.com>
251 * doc/extend.texi (__builtin_extend_pointer): Document builtin.
253 2018-02-22 DJ Delorie <dj@redhat.com>
254 Sebastian Perta <sebastian.perta@renesas.com>
255 Oleg Endo <olegendo@gcc.gnu.org>
257 * config/rx/rx.c (rx_rtx_costs): New function.
258 (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
260 2018-02-22 Thomas Preud'homme <thomas.preudhomme@arm.com>
262 * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
264 2018-02-22 Martin Liska <mliska@suse.cz>
266 PR driver/83193
267 * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
268 Add "native" as a possible value.
270 2018-02-22 Martin Liska <mliska@suse.cz>
272 PR driver/83193
273 * config/i386/i386.c (ix86_option_override_internal):
274 Add "native" as a possible value for -march and -mtune.
276 2018-02-22 Jakub Jelinek <jakub@redhat.com>
278 PR target/84502
279 * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
280 to all type variants.
282 PR tree-optimization/84503
283 * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
284 width as info->bitpos + info->bitsize - start.
285 (merged_store_group::merge_overlapping): Simplify width computation.
286 (check_no_overlap): New function.
287 (imm_store_chain_info::try_coalesce_bswap): Compute expected
288 start + width and last_order of the group, fail if check_no_overlap
289 fails.
290 (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
291 to group if check_no_overlap fails.
293 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
295 * config/rs6000/altivec.md: Delete contraint arguments to
296 define_expand, define_split, and define_peephole2, and in
297 define_insn_and_split if always unused.
298 * config/rs6000/darwin.md: Ditto.
299 * config/rs6000/dfp.md: Ditto.
300 * config/rs6000/rs6000.md: Ditto.
301 * config/rs6000/sync.md: Ditto.
302 * config/rs6000/vector.md: Ditto.
303 * config/rs6000/vsx.md: Ditto.
305 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
307 * config/rs6000/altivec.md: Write output control strings as braced
308 blocks instead of double-quoted strings.
309 * config/rs6000/darwin.md: Ditto.
310 * config/rs6000/rs6000.md: Ditto.
311 * config/rs6000/vector.md: Ditto.
312 * config/rs6000/vsx.md: Ditto.
314 2018-02-21 Jason Merrill <jason@redhat.com>
316 PR c++/84314 - ICE with templates and fastcall attribute.
317 * attribs.c (build_type_attribute_qual_variant): Remove assert.
319 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
321 * ipa-cp.c (determine_versionability): Fix comment typos.
323 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
325 PR c/84229
326 * ipa-cp.c (determine_versionability): Do not version functions caling
327 va_arg_pack.
329 2018-02-21 Martin Liska <mliska@suse.cz>
331 PR driver/83193
332 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
333 Add "native" as a possible value.
334 * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT): Define
335 the macro when native cpu detection is available.
337 2018-02-21 Martin Liska <mliska@suse.cz>
339 PR driver/83193
340 * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
341 Add "native" as a possible value.
342 * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
343 when native cpu detection is available.
345 2018-02-21 Jakub Jelinek <jakub@redhat.com>
346 Martin Sebor <msebor@redhat.com>
348 PR tree-optimization/84478
349 * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
350 false.
351 * gimple-fold.c (get_range_strlen): Make minlen const and assume it
352 can't be NULL. Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
353 support which is conservatively correct, for 2 only stay conservative
354 for maxlen. Formatting and comment capitalization fixes. Add STRICT
355 argument to the 2 argument get_range_strlen, adjust 6 arg
356 get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
357 false.
358 (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
359 (gimple_fold_builtin_strlen): Pass true as last argument to
360 get_range_strlen.
362 2018-02-20 Martin Sebor <msebor@redhat.com>
364 PR middle-end/84095
365 * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
366 (builtin_memref::set_base_and_offset): Same. Handle inner references.
367 (builtin_memref::builtin_memref): Factor out parts into
368 set_base_and_offset and call it.
370 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
372 PR middle-end/84406
373 * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
374 is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
375 greater precision. If to_mode is a MODE_PARTIAL_INT, stop the
376 search at the associated MODE_INT.
378 2018-02-20 Jeff Law <law@redhat.com>
380 PR middle-end/82123
381 PR tree-optimization/81592
382 PR middle-end/79257
383 * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
384 for range data rather than using global data.
385 * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
386 range data rather than using global data.
387 * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
388 pass it to children as needed.
389 (struct directive::fmtresult): Similarly.
390 (struct directive::set_width): Similarly.
391 (struct directive::set_precision): Similarly.
392 (format_integer, format_directive, parse_directive): Similarly.
393 (format_none): Accept unnamed vr_values parameter.
394 (format_percent, format_floating, format_character): Similarly.
395 (format_string, format_plain): Similarly.
396 * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
397 the EVRP range analyzer for range data rather than using global data.
398 * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
399 gimple-ssa-evrp-analyze.h
400 (class sprintf_dom_walker): Add after_dom_children member function.
401 Add evrp_range_analyzer member.
402 (sprintf_dom_walker::before_dom_children): Call into the EVRP
403 range analyzer as needed.
404 (sprintf_dom_walker::after_dom_children): New member function.
405 * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
406 if not optimizing.
407 (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
408 (evrp_range_analyzer::pop_to_marker): Likewise.
410 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
412 PR tree-optimization/84419
413 * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
414 with the required type if its current type is compatible but
415 different.
417 2018-02-20 Jakub Jelinek <jakub@redhat.com>
419 PR middle-end/82004
420 * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
421 after vectorization.
423 2018-02-20 Martin Liska <mliska@suse.cz>
425 PR driver/83193
426 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
427 possible values if we don't have a hint.
429 2018-02-20 Martin Liska <mliska@suse.cz>
431 PR c/84310
432 PR target/79747
433 * final.c (shorten_branches): Build align_tab array with one
434 more element.
435 * opts.c (finish_options): Add alignment option limit check.
436 (MAX_CODE_ALIGN): Likewise.
437 (MAX_CODE_ALIGN_VALUE): Likewise.
438 * doc/invoke.texi: Document maximum allowed option value for
439 all -falign-* options.
441 2018-02-19 Jakub Jelinek <jakub@redhat.com>
443 PR target/84146
444 * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
445 * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
446 * var-tracking.c (emit_note_insn_var_location): Remove all references
447 to NOTE_INSN_CALL_ARG_LOCATION.
448 (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
449 the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
450 Use copy_rtx_if_shared.
451 * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
452 NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
453 (dwarf2out_var_location): Remove handling of
454 NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
455 on call_insn.
456 * final.c (final_scan_insn): Remove all references to
457 NOTE_INSN_CALL_ARG_LOCATION.
458 (rest_of_clean_state): Likewise. Remove REG_CALL_ARG_LOCATION notes
459 before dumping final insns.
460 * except.c (emit_note_eh_region_end): Remove all references to
461 NOTE_INSN_CALL_ARG_LOCATION.
462 * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
463 * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
464 * config/arc/arc.c (hwloop_optimize): Likewise.
465 * config/arm/arm.c (create_fix_barrier): Likewise.
466 * config/s390/s390.c (s390_chunkify_start): Likewise.
467 * config/sh/sh.c (find_barrier): Likewise.
468 * config/i386/i386.c (rest_of_insert_endbranch,
469 ix86_seh_fixup_eh_fallthru): Likewise.
470 * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
471 * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
472 * config/frv/frv.c (frv_function_prologue): Likewise.
473 * emit-rtl.c (try_split): Likewise. Copy over REG_CALL_ARG_LOCATION
474 reg note.
475 (note_outside_basic_block_p): Remove all references to
476 NOTE_INSN_CALL_ARG_LOCATION.
477 * gengtype.c (adjust_field_rtx_def): Likewise.
478 * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
479 Likewise.
480 * jump.c (cleanup_barriers, delete_related_insns): Likewise.
481 * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
483 PR c++/84444
484 * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
485 is ADDR_EXPR.
487 PR tree-optimization/84452
488 * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
489 expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
490 is NULL.
492 2018-02-19 Martin Liska <mliska@suse.cz>
494 PR sanitizer/82183
495 * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
497 2018-02-19 Martin Liska <mliska@suse.cz>
498 Richard Sandiford <richard.sandiford@linaro.org>
500 PR tree-optimization/82491
501 * gimple-fold.c (get_base_constructor): Make earlier bail out
502 to prevent ubsan.
504 2018-02-19 Carl Love <cel@us.ibm.com>
506 * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
507 BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
508 BU_P8V_OVERLOAD_1.
509 * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
510 P8V_BUILTIN_VEC_NEG.
512 2018-02-19 Sebastian Perta <sebastian.perta@renesas.com>
514 * config/rl78/rl78.md (movdf): New define expand.
516 2018-02-19 Martin Liska <mliska@suse.cz>
518 PR other/80589
519 * doc/invoke.texi: Fix typo.
520 * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
522 2018-02-18 Segher Boessenkool <segher@kernel.crashing.org>
524 * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
525 handle rs6000_single_float and rs6000_double_float specially for
526 e500 family CPUs.
528 2018-02-16 Jeff Law <law@redhat.com>
530 * config/rx/rx.c (add_pop_cfi_notes): New function.;
531 (pop_regs): Use it.
533 2018-02-16 Jakub Jelinek <jakub@redhat.com>
535 PR ipa/84425
536 * ipa-inline.c (inline_small_functions): Fix a typo.
538 2018-02-16 Nathan Sidwell <nathan@acm.org>
540 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
542 2018-02-16 Carl Love <cel@us.ibm.com>
544 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
545 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
546 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
547 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
548 expansion to P8V_BUILTIN_VEC_FLOAT2.
550 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
552 PR rtl-optimization/70023
553 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
554 src_regno into account.
556 2018-02-16 Carl Love <cel@us.ibm.com>
558 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
559 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
560 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
561 * config/rs6000/rs6000.c: Remove case statements for
562 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
563 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
564 and P9V_BUILTIN_VEC_VINSERT4B.
565 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
566 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
567 * config/rs6000/vsx.md:
568 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
569 vec_insert4b.
571 2018-02-16 Carl Love <cel@us.ibm.com>
573 * config/rs6000/altivec.h: Add builtin names vec_extract4b
574 vec_insert4b.
575 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
576 definitions.
577 * config/rs6000/rs6000-c.c: Add the definitions for
578 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
579 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
580 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
581 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
582 definition for insert4b and define insn *insert3b_internal.
583 * doc/extend.texi: Add documentation for vec_extract4b.
585 2018-02-16 Nathan Sidwell <nathan@acm.org>
587 * doc/extend.texi (Backwards Compatibility): Mention friend
588 injection. Note for-scope is deprecated.
589 * doc/invoke.texi (-ffriend-injection): Deprecate.
591 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
593 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
594 that moved to I2, also allow destinations that are a paradoxical
595 subreg (instead of a normal reg).
597 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
599 PR target/83831
600 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
601 to QImode.
603 2018-02-16 Richard Biener <rguenther@suse.de>
605 PR tree-optimization/84037
606 PR tree-optimization/84016
607 PR target/82862
608 * config/i386/i386.c (ix86_builtin_vectorization_cost):
609 Adjust vec_construct for the fact we need additional higher latency
610 128bit inserts for AVX256 and AVX512 vector builds.
611 (ix86_add_stmt_cost): Scale vector construction cost for
612 elementwise loads.
614 2018-02-16 Richard Biener <rguenther@suse.de>
616 PR tree-optimization/84417
617 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
618 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
619 (non_rewritable_lvalue_p): Likewise, use poly-ints.
621 2018-02-16 Martin Liska <mliska@suse.cz>
623 PR sanitizer/84307
624 * internal-fn.def (ASAN_CHECK): Set proper flags.
625 (ASAN_MARK): Likewise.
627 2018-02-16 Julia Koval <julia.koval@intel.com>
629 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
630 from PTA_CANNONLAKE.
632 2018-02-16 Jakub Jelinek <jakub@redhat.com>
634 PR target/84272
635 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
636 Use ++iter rather than iter++ for std::list iterators.
637 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
638 defer deleting them until all nodes in the forest are processed. Do
639 free even leaf nodes. Change to_process into auto_vec.
641 PR bootstrap/84405
642 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
643 * vec.h (vec_default_construct): Use memset instead of placement new
644 if BROKEN_VALUE_INITIALIZATION is defined.
645 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
646 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
647 is defined.
649 PR rtl-optimization/83723
650 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
651 * lra.c (lra_substitute_pseudo): Likewise. If true, use
652 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
653 recursive calls.
654 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
655 callers.
656 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
658 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
660 PR rtl-optimization/81443
661 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
662 from inner REGs to paradoxical SUBREGs.
664 2018-02-16 Richard Biener <rguenther@suse.de>
666 PR tree-optimization/84399
667 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
668 For operands we can analyze at their definition make sure we can
669 analyze them at each use as well.
671 2018-02-16 Richard Biener <rguenther@suse.de>
673 PR tree-optimization/84190
674 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
675 volatile accesses if the decl isn't volatile.
677 2018-02-15 Jason Merrill <jason@redhat.com>
679 PR c++/84314 - ICE with templates and fastcall attribute.
680 * attribs.c (build_type_attribute_qual_variant): Don't clobber
681 TYPE_CANONICAL on an existing type.
683 2018-02-15 Jakub Jelinek <jakub@redhat.com>
685 PR tree-optimization/84383
686 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
687 dstoff nor call operand_equal_p if dstbase is NULL.
689 PR tree-optimization/84334
690 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
691 also a CONSTANT_CLASS_P, punt.
693 2018-02-14 Jim Wilson <jimw@sifive.com>
695 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
696 first SMALL_OPERAND check. New local min_second_step. Move assert
697 to where locals are set. Add TARGET_RVC support.
698 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
700 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
702 * doc/invoke.texi: Correct -Wformat-overflow code sample.
704 2018-02-14 Martin Sebor <msebor@redhat.com>
706 PR tree-optimization/83698
707 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
708 arrays constrain the offset range to their bounds.
709 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
710 (builtin_access::overlap): Avoid setting the size of overlap if it's
711 already been set.
712 (maybe_diag_overlap): Also consider arrays when deciding what values
713 of offsets to include in diagnostics.
715 2018-02-14 Martin Sebor <msebor@redhat.com>
717 PR c/84108
718 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
719 that correspond to the kind of a declaration.
721 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
723 PR target/83984
724 * config/pa/pa.md: Load address of PIC label using the linkage table
725 if the label is nonlocal.
727 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
729 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
730 warning message if user requests -maltivec=be.
731 * doc/invoke.texi: Document deprecation of -maltivec=be.
733 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
735 PR target/84220
736 * config/rs6000/rs6000-c.c: Update definitions for
737 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
738 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
740 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
742 PR target/84239
743 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
744 add _get_ssp intrinsics. Remove argument from
745 __builtin_ia32_rdssp[d|q].
746 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
747 * config/i386/i386-builtin.def: Remove argument from
748 __builtin_ia32_rdssp[d|q].
749 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
750 ix86_expand_special_args_builtin for _rdssp[d|q].
751 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
752 Clear register before usage.
753 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
754 Add documentation for new _get_ssp and _inc_ssp intrinsics.
756 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
758 PR tree-optimization/84357
759 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
760 operand 1 of an ARRAY_REF too.
762 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
764 PR target/83831
765 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
766 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
767 declarations.
768 (set_of_reg): New struct.
769 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
770 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
771 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
772 functions.
773 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
774 Split into bitclr, bitset, bitinvert patterns if appropriate.
775 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
776 use rx_fuse_in_memory_bitop.
777 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
778 to named insn, correct maximum insn length.
780 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
782 PR target/79242
783 * machmode.def: Define a complex mode for PARTIAL_INT.
784 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
785 MODE_PARTIAL_INT.
786 * doc/rtl.texi: Document CSPImode.
787 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
788 handling.
789 (msp430_hard_regno_nregs_with_padding): Likewise.
791 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
793 PR target/84279
794 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
796 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
798 PR rtl-optimization/84169
799 * combine.c (try_combine): New variable split_i2i3. Set it to true if
800 we generated a parallel as new i3 and we split that to new i2 and i3
801 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
802 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
803 those to i2, not i1. Partially rewrite this scan code.
805 2018-02-13 Jakub Jelinek <jakub@redhat.com>
807 PR c/82210
808 * stor-layout.c (place_field): For variable length fields, adjust
809 offset_align afterwards not just based on the field's alignment,
810 but also on the size.
812 PR middle-end/84309
813 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
814 of exps and logs in the use_exp2 case.
816 2018-02-13 Jeff Law <law@redhat.com>
818 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
819 entry for "vector".
821 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
822 ARGS as unused.
824 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
826 PR debug/84342
827 PR debug/84319
828 * common.opt (gas-loc-support, gas-locview-support): New.
829 (ginline-points, ginternal-reset-location-views): New.
830 * doc/invoke.texi: Document them. Use @itemx where intended.
831 (gvariable-location-views): Adjust.
832 * target.def (reset_location_view): New.
833 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
834 (TARGET_RESET_LOCATION_VIEW): New.
835 * doc/tm.texi: Rebuilt.
836 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
837 (dwarf2out_default_as_locview_support): New.
838 (output_asm_line_debug_info): Use option variables.
839 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
840 (output_loc_list): Likewise.
841 (add_high_low_attributes): Check option variables.
842 Don't output entry view attribute in strict mode.
843 (gen_inlined_subroutine_die): Check option variables.
844 (dwarf2out_inline_entry): Likewise.
845 (init_sections_and_labels): Likewise.
846 (dwarf2out_early_finish): Likewise.
847 (maybe_reset_location_view): New, from...
848 (dwarf2out_var_location): ... here. Call it.
849 * debug.h (dwarf2out_default_as_loc_support): Declare.
850 (dwarf2out_default_as_locview_support): Declare.
851 * hooks.c (hook_int_rtx_insn_0): New.
852 * hooks.h (hook_int_rtx_insn_0): Declare.
853 * toplev.c (process_options): Take -gas-loc-support and
854 -gas-locview-support from dwarf2out. Enable
855 -gvariable-location-views by default only with locview
856 assembler support. Enable -ginternal-reset-location-views by
857 default only if the target defines the corresponding hook.
858 Enable -ginline-points by default if location views are
859 enabled; force it disabled if statement frontiers are
860 disabled.
861 * tree-inline.c (expand_call_inline): Check option variables.
862 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
864 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
866 PR tree-optimization/84321
867 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
868 handling. Also check whether the anti-range contains any values
869 that satisfy the mask; switch to a VR_RANGE if not.
871 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
873 PR sanitizer/84340
874 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
876 2018-02-13 Martin Jambor <mjambor@suse.cz>
878 PR c++/83990
879 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
880 of call statements, also set location of a load to a temporary.
882 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
884 * config/rl78/rl78.c (add_vector_labels): New function.
885 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
886 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
887 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
888 which checks that no arguments are passed.
889 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
890 * doc/extend.texi: Documentation for the new attribute.
892 2018-02-13 Andreas Schwab <schwab@suse.de>
894 * config/riscv/linux.h (CPP_SPEC): Define.
896 2018-02-13 Jakub Jelinek <jakub@redhat.com>
898 PR target/84335
899 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
900 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
901 OPTION_MASK_ISA_AES as first argument to def_builtin_const
902 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
903 instead of OPTION_MASK_ISA_PCLMUL as first argument to
904 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
905 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
906 temporarily for AES and PCLMUL builtins.
908 PR tree-optimization/84339
909 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
910 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
911 Formatting fixes.
913 PR middle-end/84309
914 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
915 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
916 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
917 inline function.
918 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
919 inline function.
920 * omp-simd-clone.h: New file.
921 * omp-simd-clone.c: Include omp-simd-clone.h.
922 (expand_simd_clones): No longer static.
923 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
924 cgraph.h and omp-simd-clone.h.
925 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
926 (vect_recog_widen_shift_pattern): Formatting fix.
927 (vect_pattern_recog_1): Don't check optab for calls.
929 PR target/84336
930 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
931 operands[2] into a REG before using gen_lowpart on it.
933 2018-02-12 Jeff Law <law@redhat.com>
935 PR target/83760
936 * config/sh/sh.c (find_barrier): Consider a sibling call
937 a barrier as well.
939 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
940 successfully back substituting a reg.
942 2018-02-12 Richard Biener <rguenther@suse.de>
944 PR tree-optimization/84037
945 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
946 parameter, move visited init to caller.
947 (vect_slp_analyze_operations): Separate cost from validity
948 check, initialize visited once for all instances.
949 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
950 for all instances.
951 * tree-vect-stmts.c (vect_model_simple_cost): Make early
952 out an assert.
953 (vect_model_promotion_demotion_cost): Likewise.
954 (vectorizable_bswap): Guard cost modeling with !slp_node
955 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
956 SLP stmts.
957 (vectorizable_call): Likewise.
958 (vectorizable_conversion): Likewise.
959 (vectorizable_assignment): Likewise.
960 (vectorizable_shift): Likewise.
961 (vectorizable_operation): Likewise.
962 (vectorizable_store): Likewise.
963 (vectorizable_load): Likewise.
964 (vectorizable_condition): Likewise.
965 (vectorizable_comparison): Likewise.
967 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
969 PR sanitizer/84307
970 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
971 (ASAN_MARK): Fix fnspec to account for return value, change pointer
972 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
974 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
976 PR middle-end/83665
977 * params.def (inline-min-speedup): Increase from 8 to 15.
978 (max-inline-insns-auto): Decrease from 40 to 30.
979 * ipa-split.c (consider_split): Add some buffer for function to
980 be considered inlining candidate.
981 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
982 default values.
984 2018-02-12 Richard Biener <rguenther@suse.de>
986 PR tree-optimization/84037
987 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
988 matched stmts if we cannot swap the non-matched ones.
990 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
992 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
993 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
994 _mm_maskz_scalef_round_ss): New intrinsics.
995 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
996 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
997 __builtin_ia32_scalefss_round): Remove.
998 (__builtin_ia32_scalefsd_mask_round,
999 __builtin_ia32_scalefss_mask_round): New intrinsics.
1000 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
1001 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
1002 ((match_operand:VF_128 2 "<round_nimm_predicate>"
1003 "<round_constraint>")): Changed to ...
1004 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
1005 "<round_scalar_constraint>")): ... this.
1006 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
1007 %0, %1, %2<round_op3>}"): Changed to ...
1008 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
1009 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
1010 %2<round_scalar_mask_op3>}"): ... this.
1011 * config/i386/subst.md (round_scalar_nimm_predicate): New.
1013 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
1015 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
1016 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
1017 (_mm_maskz_sqrt_round_ss): New intrinsics.
1018 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
1019 (__builtin_ia32_sqrtsd_mask_round)
1020 (__builtin_ia32_sqrtss_mask_round): New builtins.
1021 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
1022 (__builtin_ia32_sqrtss_round): Remove.
1023 (__builtin_ia32_sqrtsd_mask_round)
1024 (__builtin_ia32_sqrtss_mask_round): New builtins.
1025 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
1026 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
1027 ((match_operand:VF_128 1 "vector_operand"
1028 "xBm,<round_constraint>")): Changed to ...
1029 ((match_operand:VF_128 1 "vector_operand"
1030 "xBm,<round_scalar_constraint>")): ... this.
1031 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
1032 %0, %2, %<iptr>1<round_op3>}): Changed to ...
1033 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
1034 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
1035 %<iptr>1<round_scalar_mask_op3>}): ... this.
1036 ((set_attr "prefix" "<round_prefix>")): Changed to ...
1037 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
1039 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
1041 PR target/84266
1042 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
1043 Cast vec_cmpeq result to correct type.
1044 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
1045 Cast vec_cmpgt result to correct type.
1047 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
1049 * final.c (final_scan_insn_1): Renamed from...
1050 (final_scan_insn): ... this. New wrapper, to recover
1051 seen from the outermost call in recursive ones.
1052 * config/sparc/sparc.c (output_return): Drop seen from call.
1053 (output_sibcall): Likewise.
1054 * config/visium/visium.c (output_branch): Likewise.
1056 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
1058 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
1059 function label.
1061 2018-02-10 Alan Modra <amodra@gmail.com>
1063 PR target/84300
1064 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
1065 Specify LR as an input.
1067 2018-02-10 Jakub Jelinek <jakub@redhat.com>
1069 PR sanitizer/83987
1070 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
1071 remove_member_access_dummy_vars): New functions.
1072 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
1073 lower_omp_1, execute_lower_omp): Use them.
1075 PR rtl-optimization/84308
1076 * shrink-wrap.c (spread_components): Release todo vector.
1078 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
1080 PR rtl-optimization/57193
1081 * ira-color.c (struct allocno_color_data): Add member
1082 conflict_allocno_hard_prefs.
1083 (update_conflict_allocno_hard_prefs): New.
1084 (bucket_allocno_compare_func): Add a preference based on
1085 conflict_allocno_hard_prefs.
1086 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
1087 (color_allocnos): Remove a dead code. Initiate
1088 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
1090 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1092 PR target/84226
1093 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
1094 constraint from =wa to wa. Avoid a subreg on the output operand,
1095 instead use a pseudo and subreg it in a move.
1096 (p9_xxbrd_<mode>): Changed to ...
1097 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
1098 (p9_xxbrd_v2df): New expander.
1099 (p9_xxbrw_<mode>): Changed to ...
1100 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
1101 (p9_xxbrw_v4sf): New expander.
1103 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1105 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
1107 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
1109 PR target/83926
1110 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
1111 multiply in 32-bit mode.
1112 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
1113 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
1114 mode.
1116 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1118 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
1119 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
1120 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
1121 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
1123 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
1125 PR lto/84213
1126 * dwarf2out.c (is_trivial_indirect_ref): New function.
1127 (dwarf2out_late_global_decl): Do not generate a location
1128 attribute for variables that have a non-trivial DECL_VALUE_EXPR
1129 and that are not defined in the current unit.
1131 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1133 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
1134 instead of a libcall for UNORDERED.
1136 2018-02-09 Tamar Christina <tamar.christina@arm.com>
1138 PR target/82641
1139 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
1140 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
1142 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1144 PR target/PR84295
1145 * config/s390/s390.c (s390_set_current_function): Invoke
1146 s390_indirect_branch_settings also if fndecl didn't change.
1148 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1150 * config/rs6000/rs6000.md (blockage): Set length to zero.
1152 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1154 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
1156 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1158 PR sanitizer/84285
1159 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
1160 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
1161 -static-lib*san.
1163 PR debug/84252
1164 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
1165 PARALLEL incoming that failed vt_get_decl_and_offset check.
1167 PR middle-end/84237
1168 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
1169 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
1170 TREE_READONLY bit.
1171 (get_variable_section): For decls in named .bss* sections pass true as
1172 second argument to bss_initializer_p.
1174 2018-02-09 Marek Polacek <polacek@redhat.com>
1175 Jakub Jelinek <jakub@redhat.com>
1177 PR c++/83659
1178 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
1179 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
1180 Sync some changes from cxx_fold_indirect_ref.
1182 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1184 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
1185 markers.
1186 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
1187 (BLOCK_INLINE_ENTRY_LABEL): New.
1188 (dwarf2out_var_location): Disregard inline entry markers.
1189 (inline_entry_data): New struct.
1190 (inline_entry_data_hasher): New hashtable type.
1191 (inline_entry_data_hasher::hash): New.
1192 (inline_entry_data_hasher::equal): New.
1193 (inline_entry_data_table): New variable.
1194 (add_high_low_attributes): Add DW_AT_entry_pc and
1195 DW_AT_GNU_entry_view attributes if a pending entry is found
1196 in inline_entry_data_table. Add old entry_pc attribute only
1197 if debug nonbinding markers are disabled.
1198 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
1199 markers are enabled.
1200 (block_within_block_p, dwarf2out_inline_entry): New.
1201 (dwarf2out_finish): Check that no entries remained in
1202 inline_entry_data_table.
1203 * final.c (reemit_insn_block_notes): Handle inline entry notes.
1204 (final_scan_insn, notice_source_line): Likewise.
1205 (rest_of_clean_state): Skip inline entry markers.
1206 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
1207 markers.
1208 * gimple.c (gimple_build_debug_inline_entry): New.
1209 * gimple.h (enum gimple_debug_subcode): Add
1210 GIMPLE_DEBUG_INLINE_ENTRY.
1211 (gimple_build_debug_inline_entry): Declare.
1212 (gimple_debug_inline_entry_p): New.
1213 (gimple_debug_nonbind_marker_p): Adjust.
1214 * insn-notes.def (INLINE_ENTRY): New.
1215 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
1216 inline entry marker notes.
1217 (print_insn): Likewise.
1218 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
1219 (INSN_DEBUG_MARKER_KIND): Likewise.
1220 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
1221 * tree-inline.c (expand_call_inline): Build and insert
1222 debug_inline_entry stmt.
1223 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
1224 inline entry blocks early, if nonbind markers are enabled.
1225 (dump_scope_block): Dump fragment info.
1226 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
1227 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
1228 (gimple_build_debug_inline_entry): New.
1229 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
1230 Enable/disable inline entry points too.
1231 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
1232 (DEBUG_INSN): Describe inline entry markers.
1234 * common.opt (gvariable-location-views): New.
1235 (gvariable-location-views=incompat5): New.
1236 * config.in: Rebuilt.
1237 * configure: Rebuilt.
1238 * configure.ac: Test assembler for view support.
1239 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
1240 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
1241 * dwarf2out.c (var_loc_view): New typedef.
1242 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
1243 (dwarf2out_locviews_in_attribute): New.
1244 (dwarf2out_locviews_in_loclist): New.
1245 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
1246 (enum dw_line_info_opcode): Add LI_adv_address.
1247 (struct dw_line_info_table): Add view.
1248 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
1249 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
1250 (zero_view_p): New variable.
1251 (ZERO_VIEW_P): New macro.
1252 (output_asm_line_debug_info): New.
1253 (struct var_loc_node): Add view.
1254 (add_AT_view_list, AT_loc_list): New.
1255 (add_var_loc_to_decl): Add view param. Test it against last.
1256 (new_loc_list): Add view params. Record them.
1257 (AT_loc_list_ptr): Handle loc and view lists.
1258 (view_list_to_loc_list_val_node): New.
1259 (print_dw_val): Handle dw_val_class_view_list.
1260 (size_of_die): Likewise.
1261 (value_format): Likewise.
1262 (loc_list_has_views): New.
1263 (gen_llsym): Set vl_symbol too.
1264 (maybe_gen_llsym, skip_loc_list_entry): New.
1265 (dwarf2out_maybe_output_loclist_view_pair): New.
1266 (output_loc_list): Output view list or entries too.
1267 (output_view_list_offset): New.
1268 (output_die): Handle dw_val_class_view_list.
1269 (output_dwarf_version): New.
1270 (output_compilation_unit_header): Use it.
1271 (output_skeleton_debug_sections): Likewise.
1272 (output_rnglists, output_line_info): Likewise.
1273 (output_pubnames, output_aranges): Update version comments.
1274 (output_one_line_info_table): Output view numbers in asm comments.
1275 (dw_loc_list): Determine current endview, pass it to new_loc_list.
1276 Call maybe_gen_llsym.
1277 (loc_list_from_tree_1): Adjust.
1278 (add_AT_location_description): Create view list attribute if
1279 needed, check it's absent otherwise.
1280 (convert_cfa_to_fb_loc_list): Adjust.
1281 (maybe_emit_file): Call output_asm_line_debug_info for test.
1282 (dwarf2out_var_location): Reset views as needed. Precompute
1283 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
1284 attribute. Set view.
1285 (new_line_info_table): Reset next view.
1286 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
1287 (dwarf2out_source_line): Likewise. Output view resets and labels to
1288 the assembler, or select appropriate line info opcodes.
1289 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
1290 (optimize_string_length): Catch it. Adjust.
1291 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
1292 dw_val_class_view_list, and remove it if no longer needed.
1293 (hash_loc_list): Hash view numbers.
1294 (loc_list_hasher::equal): Compare them.
1295 (optimize_location_lists): Check whether a view list symbol is
1296 needed, and whether the locview attribute is present, and
1297 whether they match. Remove the locview attribute if no longer
1298 needed.
1299 (index_location_lists): Call skip_loc_list_entry for test.
1300 (dwarf2out_finish): Call output_asm_line_debug_info for test.
1301 Use output_dwarf_version.
1302 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
1303 (struct dw_val_node): Add val_view_list.
1304 * final.c (SEEN_NEXT_VIEW): New.
1305 (set_next_view_needed): New.
1306 (clear_next_view_needed): New.
1307 (maybe_output_next_view): New.
1308 (final_start_function): Rename to...
1309 (final_start_function_1): ... this. Take pointer to FIRST,
1310 add SEEN parameter. Emit param bindings in the initial view.
1311 (final_start_function): Reintroduce SEEN-less interface.
1312 (final): Rename to...
1313 (final_1): ... this. Take SEEN parameter. Output final pending
1314 next view at the end.
1315 (final): Reintroduce seen-less interface.
1316 (final_scan_insn): Output pending next view before switching
1317 sections or ending a block. Mark the next view as needed when
1318 outputting variable locations. Notify debug backend of section
1319 changes, and of location view changes.
1320 (rest_of_handle_final): Adjust.
1321 * toplev.c (process_options): Autodetect value for debug variable
1322 location views option. Warn on incompat5 without -gdwarf-5.
1323 * doc/invoke.texi (gvariable-location-views): New.
1324 (gvariable-location-views=incompat5): New.
1325 (gno-variable-location-views): New.
1327 2018-02-08 David Malcolm <dmalcolm@redhat.com>
1329 PR tree-optimization/84136
1330 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
1331 that the result of find_edge is non-NULL.
1333 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1335 PR target/83008
1336 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
1337 storing integer register in SImode. Fix cost of 256 and 512
1338 byte aligned SSE register store.
1340 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1342 * config/i386/i386.c (ix86_multiplication_cost): Fix
1343 multiplication cost for TARGET_AVX512DQ.
1345 2018-02-08 Marek Polacek <polacek@redhat.com>
1347 PR tree-optimization/84238
1348 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
1349 get_range_strlen.
1351 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1353 PR tree-optimization/84265
1354 * tree-vect-stmts.c (vectorizable_store): Don't treat
1355 VMAT_CONTIGUOUS accesses as grouped.
1356 (vectorizable_load): Likewise.
1358 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1360 PR tree-optimization/81635
1361 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
1362 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
1363 (test_round_for_mask): New functions.
1364 (wide_int_cc_tests): Call test_round_for_mask.
1365 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
1366 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
1367 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
1368 range returned by get_range_info.
1370 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1372 PR ipa/81360
1373 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
1374 * symtab.c: Include builtins.h
1375 (symtab_node::output_to_lto_symbol_table_p): Move here
1376 from lto-streamer-out.c:output_symbol_p.
1377 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
1378 (output_symbol_p): Move all logic to symtab.c
1379 (produce_symtab): Update.
1381 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1383 * config/s390/s390-opts.h (enum indirect_branch): Define.
1384 * config/s390/s390-protos.h (s390_return_addr_from_memory)
1385 (s390_indirect_branch_via_thunk)
1386 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
1387 (enum s390_indirect_branch_type): Define.
1388 * config/s390/s390.c (struct s390_frame_layout, struct
1389 machine_function): Remove.
1390 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
1391 (indirect_branch_table_label_no, indirect_branch_table_name):
1392 Define variables.
1393 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
1394 (enum s390_indirect_branch_option): Define.
1395 (s390_return_addr_from_memory): New function.
1396 (s390_handle_string_attribute): New function.
1397 (s390_attribute_table): Add new attribute handler.
1398 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
1399 (s390_indirect_branch_via_thunk): New function.
1400 (s390_indirect_branch_via_inline_thunk): New function.
1401 (s390_function_ok_for_sibcall): When jumping via thunk disallow
1402 sibling call optimization for non z10 compiles.
1403 (s390_emit_call): Force indirect branch target to be a single
1404 register. Add r1 clobber for non-z10 compiles.
1405 (s390_emit_epilogue): Emit return jump via return_use expander.
1406 (s390_reorg): Handle JUMP_INSNs as execute targets.
1407 (s390_option_override_internal): Perform validity checks for the
1408 new command line options.
1409 (s390_indirect_branch_attrvalue): New function.
1410 (s390_indirect_branch_settings): New function.
1411 (s390_set_current_function): Invoke s390_indirect_branch_settings.
1412 (s390_output_indirect_thunk_function): New function.
1413 (s390_code_end): Implement target hook.
1414 (s390_case_values_threshold): Implement target hook.
1415 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
1416 macros.
1417 * config/s390/s390.h (struct s390_frame_layout)
1418 (struct machine_function): Move here from s390.c.
1419 (TARGET_INDIRECT_BRANCH_NOBP_RET)
1420 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
1421 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
1422 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
1423 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
1424 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
1425 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
1426 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
1427 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
1428 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
1429 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
1430 (mnemonic attribute): Add values which aren't recognized
1431 automatically.
1432 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
1433 pattern for branch conversion. Fix mnemonic attribute.
1434 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
1435 indirect branch via thunk if requested.
1436 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
1437 ("*indirect_jump"): Disable for branch conversion using out of
1438 line thunks.
1439 ("indirect_jump_via_thunk<mode>_z10")
1440 ("indirect_jump_via_thunk<mode>")
1441 ("indirect_jump_via_inlinethunk<mode>_z10")
1442 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
1443 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
1444 ("casesi_jump_via_inlinethunk<mode>_z10")
1445 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
1446 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
1447 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
1448 ("*indirect2_jump"): Disable for branch conversion.
1449 ("casesi_jump"): Turn into expander and expand patterns for branch
1450 conversion.
1451 ("return_use"): New expander.
1452 ("*return"): Emit return via thunk and rename it to ...
1453 ("*return<mode>"): ... this one.
1454 * config/s390/s390.opt: Add new options and and enum for the
1455 option values.
1457 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1459 * lra-constraints.c (match_reload): Unconditionally use
1460 gen_lowpart_SUBREG, rather than selecting between that
1461 and equivalent gen_rtx_SUBREG code.
1463 2018-02-08 Richard Biener <rguenther@suse.de>
1465 PR tree-optimization/84233
1466 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
1467 changed flag instead of boguously re-using phi_inserted.
1469 2018-02-08 Martin Jambor <mjambor@suse.cz>
1471 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
1472 static local variables.
1474 2018-02-08 Richard Biener <rguenther@suse.de>
1476 PR tree-optimization/84278
1477 * tree-vect-stmts.c (vectorizable_store): When looking for
1478 smaller vector types to perform grouped strided loads/stores
1479 make sure the mode is supported by the target.
1480 (vectorizable_load): Likewise.
1482 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1484 * config/aarch64/aarch64.c (aarch64_components_for_bb):
1485 Increase LDP/STP opportunities by adding adjacent callee-saves.
1487 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1489 PR rtl-optimization/84068
1490 PR rtl-optimization/83459
1491 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
1493 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
1495 PR tree-optimization/84224
1496 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
1497 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
1498 non-zero arguments.
1500 2018-02-07 Iain Sandoe <iain@codesourcery.com>
1502 PR target/84113
1503 * config/rs6000/altivec.md (*restore_world): Remove LR use.
1504 * config/rs6000/predicates.md (restore_world_operation): Adjust op
1505 count, remove one USE.
1507 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
1509 * doc/install.texi (Configuration): Document the
1510 --with-long-double-format={ibm,ieee} PowerPC configuration
1511 options.
1513 PR target/84154
1514 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
1515 Convert from define_expand to be define_insn_and_split. Rework
1516 float/double/_Float128 conversions to QI/HI/SImode to work with
1517 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
1518 conversions to QI/HImode types did a store and then a load to
1519 truncate the value. For conversions to VSX registers, don't split
1520 the insn, instead emit the code directly. Use the code iterator
1521 any_fix to combine signed and unsigned conversions.
1522 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
1523 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
1524 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1525 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
1526 (fix_<mode>di2_hw): Likewise.
1527 (fixuns_<mode>di2_hw): Likewise.
1528 (fix_<mode>si2_hw): Likewise.
1529 (fixuns_<mode>si2_hw): Likewise.
1530 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
1531 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1532 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
1533 fix<uns>_trunc<SFDF:mode>si2_p8.
1534 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
1535 used.
1536 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1537 (fix<uns>_<mode>_mem): Likewise.
1538 (fctiw<u>z_<mode>_mem): Likewise.
1539 (fix<uns>_<mode>_mem): Likewise.
1540 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1541 the register allocator from doing a direct move to the GPRs to do
1542 a store, and instead use the ISA 3.0 store byte/half-word from
1543 vector register instruction. For IEEE 128-bit floating point,
1544 also optimize stores of 32-bit ints.
1545 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1547 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1549 * genextract.c (push_pathstr_operand): New function to support
1550 [a-zA-Z].
1551 (walk_rtx): Call push_pathstr_operand.
1552 (print_path): Support [a-zA-Z].
1554 2018-02-07 Richard Biener <rguenther@suse.de>
1556 PR tree-optimization/84037
1557 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1558 (cse_and_gimplify_to_preheader): Declare.
1559 (vect_get_place_in_interleaving_chain): Likewise.
1560 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1561 ivexpr_map.
1562 (_loop_vec_info::~_loop_vec_info): Delete it.
1563 (cse_and_gimplify_to_preheader): New function.
1564 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1565 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1566 (vectorizable_load): Likewise. For grouped stores always base
1567 the IV on the first element.
1568 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1569 condition before gimplifying.
1571 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1573 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1574 *DIV_EXPR and *MOD_EXPR.
1576 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1578 PR target/84248
1579 * config/i386/i386.c (ix86_option_override_internal): Mask out
1580 the CF_SET bit when checking -fcf-protection.
1582 2018-02-07 Tom de Vries <tom@codesourcery.com>
1584 PR libgomp/84217
1585 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1586 enough.
1588 2018-02-07 Richard Biener <rguenther@suse.de>
1590 PR tree-optimization/84204
1591 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1592 this place.
1594 PR tree-optimization/84205
1595 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1596 special-case isl_ast_op_zdiv_r.
1598 PR tree-optimization/84223
1599 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1600 Only add conditions from within the region.
1601 (gather_bbs::after_dom_children): Adjust.
1603 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1605 PR target/84209
1606 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1607 * config/avr/avr.md: Only post-reload split REG-REG moves if
1608 either register is REGERAL_REG_P.
1610 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1612 PR tree-optimization/84235
1613 * tree-ssa-scopedtables.c
1614 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1615 if the subtraction is performed in floating point type where NaNs are
1616 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1617 build 1. Formatting fix.
1619 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1621 PR target/84146
1622 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1623 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1624 and skip it regardless of bb boundaries. Use CALL_P macro,
1625 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1626 unnecessarily, formatting fix.
1628 2018-02-06 Michael Collison <michael.collison@arm.com>
1630 * config/arm/thumb2.md:
1631 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1632 (*thumb_mov_notscc): Ditto.
1634 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1636 PR target/84154
1637 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1638 unsigned_fix, not "s".
1640 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1642 * configure.ac (gcc_fn_eh_frame_ro): New function.
1643 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1644 correct .eh_frame permissions.
1645 * configure: Regenerate.
1647 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1649 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1650 irrelevant options.
1652 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1654 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1655 Display warning message for -mno-speculate-indirect-jumps.
1657 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1659 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1660 Undocumented.
1661 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1663 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1665 PR tree-optimization/84225
1666 * tree-eh.c (find_trapping_overflow): Only call
1667 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1669 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1671 PR target/84145
1672 * config/i386/i386.c: Reimplement the check of possible options
1673 -mibt/-mshstk conbination. Change error messages.
1674 * doc/invoke.texi: Fix a typo: remove extra '='.
1676 2018-02-06 Marek Polacek <polacek@redhat.com>
1678 PR tree-optimization/84228
1679 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1681 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1683 PR target/82641
1684 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1685 emitted arch directives.
1686 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1687 __ARM_FEATURE_COPROC before changing architectures.
1689 2018-02-06 Richard Biener <rguenther@suse.de>
1691 * config/i386/i386.c (print_reg): Fix typo.
1692 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1694 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1696 * configure: Regenerate.
1698 2018-02-05 Martin Sebor <msebor@redhat.com>
1700 PR tree-optimization/83369
1701 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1702 inlining context.
1704 2018-02-05 Martin Liska <mliska@suse.cz>
1706 * doc/invoke.texi: Cherry-pick upstream r323995.
1708 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1710 * ira.c (ira_init_register_move_cost): Adjust comment.
1712 2018-02-05 Martin Liska <mliska@suse.cz>
1714 PR gcov-profile/84137
1715 * doc/gcov.texi: Fix typo in documentation.
1717 2018-02-05 Martin Liska <mliska@suse.cz>
1719 PR gcov-profile/83879
1720 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1721 using dlopen functionality.
1723 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1725 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1726 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1727 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1728 _mm_maskz_range_round_ss): New intrinsics.
1729 (__builtin_ia32_rangesd128_round)
1730 (__builtin_ia32_rangess128_round): Remove.
1731 (__builtin_ia32_rangesd128_mask_round,
1732 __builtin_ia32_rangess128_mask_round): New builtins.
1733 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1734 __builtin_ia32_rangess128_round): Remove.
1735 (__builtin_ia32_rangesd128_mask_round,
1736 __builtin_ia32_rangess128_mask_round): New builtins.
1737 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1738 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1739 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1740 "<round_saeonly_constraint>")): Changed to ...
1741 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1742 "<round_saeonly_scalar_constraint>")): ... this.
1743 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1744 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1745 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1746 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1747 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1749 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1751 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1752 options.
1753 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1754 Remove all values except native, 8540 and 8548.
1756 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1758 * config/i386/i386.c (ix86_output_function_return): Pass
1759 INVALID_REGNUM, instead of -1, as invalid register number to
1760 indirect_thunk_name and output_indirect_thunk.
1762 2018-02-02 Julia Koval <julia.koval@intel.com>
1764 * config.gcc: Add -march=icelake.
1765 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1766 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1767 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1768 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1769 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1770 (processor_target_table): Add icelake.
1771 (ix86_option_override_internal): Handle new PTAs.
1772 (get_builtin_code_for_version): Handle icelake.
1773 (M_INTEL_COREI7_ICELAKE): New.
1774 (fold_builtin_cpu): Handle icelake.
1775 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1776 * doc/invoke.texi: Add -march=icelake.
1778 2018-02-02 Julia Koval <julia.koval@intel.com>
1780 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1781 to wide_int_bitmask.
1782 * wide-int-bitmask.h: New.
1784 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1786 PR target/84066
1787 * config/i386/i386.md: Replace Pmode with word_mode in
1788 builtin_setjmp_setup and builtin_longjmp to support x32.
1790 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1792 PR target/56010
1793 PR target/83743
1794 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1795 #include "opts.h".
1796 (rs6000_supported_cpu_names): New static variable.
1797 (linux_cpu_translation_table): Likewise.
1798 (elf_platform) <cpu>: Define new static variable and use it.
1799 Translate kernel AT_PLATFORM name to canonical name if needed.
1800 Error if platform name is unknown.
1802 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1804 PR target/84089
1805 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1807 2018-02-01 Jeff Law <law@redhat.com>
1809 PR target/84128
1810 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1811 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1812 the scratch if RELEASE_VIA_POP is false.
1813 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1814 If we have to save a temporary register, decrement SIZE appropriately.
1815 Pass new arguments to release_scratch_register_on_entry.
1816 (ix86_adjust_stack_and_probe): Likewise.
1817 (ix86_emit_probe_stack_range): Pass new arguments to
1818 release_scratch_register_on_entry.
1820 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1822 PR rtl-optimization/84157
1823 * combine.c (change_zero_ext): Use REG_P predicate in
1824 front of HARD_REGISTER_P predicate.
1826 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1828 * config/avr/avr.c (avr_option_override): Move disabling of
1829 -fdelete-null-pointer-checks to...
1830 * common/config/avr/avr-common.c (avr_option_optimization_table):
1831 ...here.
1833 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1835 PR tree-optimization/81635
1836 * tree-data-ref.c (split_constant_offset_1): For types that
1837 wrap on overflow, try to use range info to prove that wrapping
1838 cannot occur.
1840 2018-02-01 Renlin Li <renlin.li@arm.com>
1842 PR target/83370
1843 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1844 TAILCALL_ADDR_REGS.
1845 (aarch64_register_move_cost): Likewise.
1846 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1847 TAILCALL_ADDR_REGS.
1848 (REG_CLASS_NAMES): Likewise.
1849 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1850 TAILCALL_ADDR_REGS. Remove IP registers.
1851 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1853 2018-02-01 Richard Biener <rguenther@suse.de>
1855 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1856 for specifying RPO order and allow NULL for that.
1857 * domwalk.c (dom_walker::dom_walker): Likewise.
1858 (dom_walker::walk): Handle NULL RPO order.
1859 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1860 in RPO order.
1861 (rewrite_update_dom_walker): Likewise.
1862 (mark_def_dom_walker): Likewise.
1864 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1866 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1867 (aarch64_maybe_expand_sve_subreg_move): Declare.
1868 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1869 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1870 predicate.
1871 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1872 that are semantically a reverse operation.
1873 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1874 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1875 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1876 functions.
1877 (aarch64_can_change_mode_class): For big-endian, forbid changes
1878 between two SVE modes if they have different element sizes.
1880 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1882 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1883 the TImode handling for big-endian targets.
1885 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1887 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1888 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1889 not just bytes.
1890 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1891 Remove BSWAP handing for big-endian targets and use the form of
1892 LD1RQ appropariate for the mode.
1894 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1896 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1897 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1898 duplicated element.
1900 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1902 PR tearget/83845
1903 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1904 check for operands that need to go through aarch64_sve_reload_be.
1906 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1908 PR tree-optimization/81661
1909 PR tree-optimization/84117
1910 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1911 * tree-eh.c: Include gimplify.h.
1912 (find_trapping_overflow, replace_trapping_overflow,
1913 rewrite_to_non_trapping_overflow): New functions.
1914 * tree-vect-loop.c: Include tree-eh.h.
1915 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1916 * tree-data-ref.c: Include tree-eh.h.
1917 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1919 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1921 PR rtl-optimization/84123
1922 * combine.c (change_zero_ext): Check if hard register satisfies
1923 can_change_dest_mode before calling gen_lowpart_SUBREG.
1925 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1927 PR target/82444
1928 * ira.c (ira_init_register_move_cost): Remove assert.
1930 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1932 PR rtl-optimization/84071
1933 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1934 * doc/tm.texi: Regenerate.
1936 2018-01-31 Richard Biener <rguenther@suse.de>
1938 PR tree-optimization/84132
1939 * tree-data-ref.c (analyze_miv_subscript): Properly
1940 check whether evolution_function_is_affine_multivariate_p
1941 before calling gcd_of_steps_may_divide_p.
1943 2018-01-31 Julia Koval <julia.koval@intel.com>
1945 PR target/83618
1946 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1947 * config/i386/i386.md (rdpid_rex64) New.
1948 (rdpid): Make 32bit only.
1950 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1952 PR lto/84105
1953 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1954 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1956 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1958 Revert
1959 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1961 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1963 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1965 PR rtl-optimization/84071
1966 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1967 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1969 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1971 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1972 (arc_attribute_table): Add 'aux' attribute.
1973 (arc_in_small_data_p): Consider aux like variables.
1974 (arc_is_aux_reg_p): New function.
1975 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1976 (arc_get_aux_arg): New function.
1977 (prepare_move_operands): Handle aux-register access.
1978 (arc_handle_aux_attribute): New function.
1979 * doc/extend.texi (ARC Variable attributes): Add subsection.
1981 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1983 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1984 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1985 (arc_attribute_table): Add 'uncached' attribute.
1986 (arc_print_operand): Print '.di' flag for uncached memory
1987 accesses.
1988 (arc_in_small_data_p): Do not consider for small data the uncached
1989 types.
1990 (arc_is_uncached_mem_p): New function.
1991 * config/arc/predicates.md (compact_store_memory_operand): Check
1992 for uncached memory accesses.
1993 (nonvol_nonimm_operand): Likewise.
1994 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1996 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1998 PR c/84100
1999 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
2000 falign-loops=): Add Optimization flag.
2002 2018-01-30 Jeff Law <law@redhat.com>
2004 PR target/84064
2005 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
2006 INT_REGISTERS_SAVED. Check it prior to calling
2007 get_scratch_register_on_entry.
2008 (ix86_adjust_stack_and_probe): Similarly.
2009 (ix86_emit_probe_stack_range): Similarly.
2010 (ix86_expand_prologue): Corresponding changes.
2012 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2014 PR target/40411
2015 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
2016 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
2018 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
2020 PR target/84112
2021 * lra-constraints.c (curr_insn_transform): Process AND in the
2022 address.
2024 2018-01-30 Jakub Jelinek <jakub@redhat.com>
2026 PR rtl-optimization/83986
2027 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
2028 dependence against last_pending_memory_flush in addition to
2029 pending_jump_insns.
2031 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
2033 PR tree-optimization/81611
2034 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
2035 copies.
2037 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2039 PR target/83758
2040 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
2041 a reg rtx.
2043 2018-01-30 Richard Biener <rguenther@suse.de>
2044 Jakub Jelinek <jakub@redhat.com>
2046 PR tree-optimization/84111
2047 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
2048 inner loops added during recursion, as they don't have up-to-date
2049 SSA form.
2051 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2053 PR ipa/81360
2054 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
2055 (can_inline_edge_by_limits_p): ... here.
2056 (can_early_inline_edge_p, check_callers,
2057 update_caller_keys, update_callee_keys, recursive_inlining,
2058 add_new_edges_to_heap, speculation_useful_p,
2059 inline_small_functions,
2060 inline_small_functions, flatten_function,
2061 inline_to_all_callers_1): Update.
2063 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2065 * profile-count.c (profile_count::combine_with_ipa_count): Handle
2066 zeros correctly.
2068 2018-01-30 Richard Biener <rguenther@suse.de>
2070 PR tree-optimization/83008
2071 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
2072 invariant and constant vector uses in stmts when they need
2073 more than one stmt.
2075 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2077 PR bootstrap/84017
2078 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
2079 * configure: Regenerate.
2081 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2083 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
2084 pattern.
2085 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
2086 Use gen_rtx_REG rather than gen_lowpart.
2088 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2090 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
2091 rather than 0 when creating partial subregs.
2093 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2095 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
2096 of usage.
2098 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
2100 PR target/81550
2101 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
2102 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
2103 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
2104 flags. This restores the settings used before the 2017-07-24.
2105 Turning off pre increment/decrement/modify allows IVOPTS to
2106 optimize DF/SF loops where the index is an int.
2108 2018-01-29 Richard Biener <rguenther@suse.de>
2109 Kelvin Nilsen <kelvin@gcc.gnu.org>
2111 PR bootstrap/80867
2112 * tree-vect-stmts.c (vectorizable_call): Don't call
2113 targetm.vectorize_builtin_md_vectorized_function if callee is
2114 NULL.
2116 2018-01-22 Carl Love <cel@us.ibm.com>
2118 * doc/extend.tex: Fix typo in second arg in
2119 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
2121 2018-01-29 Richard Biener <rguenther@suse.de>
2123 PR tree-optimization/84086
2124 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
2125 (flush_ssaname_freelist): When SSA names were released reset
2126 the SCEV hash table.
2128 2018-01-29 Richard Biener <rguenther@suse.de>
2130 PR tree-optimization/84057
2131 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
2132 removed paths when removing edges.
2134 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
2136 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
2137 -mfunction-return=@var{choice}.
2139 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
2141 PR diagnostic/84034
2142 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
2143 Handle CR like TAB.
2144 (layout::print_source_line): Likewise.
2145 (test_get_line_width_without_trailing_whitespace): Add test cases.
2147 2018-01-27 Jakub Jelinek <jakub@redhat.com>
2149 PR middle-end/84040
2150 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
2151 debug insns.
2153 2018-01-26 Jim Wilson <jimw@sifive.com>
2155 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
2157 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
2158 specified.
2160 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2162 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
2163 and CMP + SUB-immediate -> SUBS.
2165 2018-01-26 Martin Sebor <msebor@redhat.com>
2167 PR tree-optimization/83896
2168 * tree-ssa-strlen.c (get_string_len): Rename...
2169 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
2170 Avoid assuming length is constant.
2171 (handle_char_store): Use HOST_WIDE_INT for string length.
2173 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
2175 PR target/81763
2176 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
2177 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
2179 2018-01-26 Richard Biener <rguenther@suse.de>
2181 PR rtl-optimization/84003
2182 * dse.c (record_store): Only record redundant stores when
2183 the earlier store aliases at least all accesses the later one does.
2185 2018-01-26 Jakub Jelinek <jakub@redhat.com>
2187 PR rtl-optimization/83985
2188 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
2189 REG_CFA_RESTORE insns.
2190 (delete_unmarked_insns): Don't ignore separate shrink wrapping
2191 REG_CFA_RESTORE insns here.
2193 PR c/83989
2194 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
2195 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
2197 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2199 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
2200 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
2201 (arc_init): Likewise.
2202 (arc_override_options): Likewise.
2203 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
2204 value.
2205 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
2206 support.
2207 * config/arc/arc.h (TARGET_DBNZ): Define.
2208 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
2209 properly set the tune attribute.
2210 (dbnz): Use TARGET_DBNZ guard.
2211 * config/arc/arc.opt (mtune): Add core3 option.
2213 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2215 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
2216 recognize new pic like addresses.
2217 (arc_delegitimize_address): Clean up.
2219 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2221 * config/arc/arc-arches.def: Option mrf16 valid for all
2222 architectures.
2223 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
2224 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
2225 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
2226 * config/arc/arc-tables.opt: Regenerate.
2227 * config/arc/arc.c (arc_conditional_register_usage): Handle
2228 reduced register file case.
2229 (arc_file_start): Set must have build attributes.
2230 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
2231 mrf16 option value.
2232 * config/arc/arc.opt (mrf16): Add new option.
2233 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
2234 * config/arc/genmultilib.awk: Handle new mrf16 option.
2235 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
2236 * config/arc/t-multilib: Regenerate.
2237 * doc/invoke.texi (ARC Options): Document mrf16 option.
2239 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2241 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
2242 * config/arc/arc.c (arc_handle_secure_attribute): New function.
2243 (arc_attribute_table): Add 'secure_call' attribute.
2244 (arc_print_operand): Print secure call operand.
2245 (arc_function_ok_for_sibcall): Don't optimize tail calls when
2246 secure.
2247 (arc_is_secure_call_p): New function. * config/arc/arc.md
2248 (call_i): Add support for sjli instruction.
2249 (call_value_i): Likewise.
2250 * config/arc/constraints.md (Csc): New constraint.
2252 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2253 John Eric Martin <John.Martin@emmicro-us.com>
2255 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
2256 * config/arc/arc.c (_arc_jli_section): New struct.
2257 (arc_jli_section): New type.
2258 (rc_jli_sections): New static variable.
2259 (arc_handle_jli_attribute): New function.
2260 (arc_attribute_table): Add jli_always and jli_fixed attribute.
2261 (arc_file_end): New function.
2262 (TARGET_ASM_FILE_END): Define.
2263 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
2264 (arc_add_jli_section): New function.
2265 (jli_call_scan): Likewise.
2266 (arc_reorg): Call jli_call_scan.
2267 (arc_output_addsi): Remove 'S' from printing asm operand.
2268 (arc_is_jli_call_p): New function.
2269 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
2270 operand.
2271 (movhi_insn): Likewise.
2272 (movsi_insn): Likewise.
2273 (movsi_set_cc_insn): Likewise.
2274 (loadqi_update): Likewise.
2275 (load_zeroextendqisi_update): Likewise.
2276 (load_signextendqisi_update): Likewise.
2277 (loadhi_update): Likewise.
2278 (load_zeroextendhisi_update): Likewise.
2279 (load_signextendhisi_update): Likewise.
2280 (loadsi_update): Likewise.
2281 (loadsf_update): Likewise.
2282 (movsicc_insn): Likewise.
2283 (bset_insn): Likewise.
2284 (bxor_insn): Likewise.
2285 (bclr_insn): Likewise.
2286 (bmsk_insn): Likewise.
2287 (bicsi3_insn): Likewise.
2288 (cmpsi_cc_c_insn): Likewise.
2289 (movsi_ne): Likewise.
2290 (movsi_cond_exec): Likewise.
2291 (clrsbsi2): Likewise.
2292 (norm_f): Likewise.
2293 (normw): Likewise.
2294 (swap): Likewise.
2295 (divaw): Likewise.
2296 (flag): Likewise.
2297 (sr): Likewise.
2298 (kflag): Likewise.
2299 (ffs): Likewise.
2300 (ffs_f): Likewise.
2301 (fls): Likewise.
2302 (call_i): Remove 'S' asm letter, add jli instruction.
2303 (call_value_i): Likewise.
2304 * config/arc/arc.op (mjli-always): New option.
2305 * config/arc/constraints.md (Cji): New constraint.
2306 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
2307 operand.
2308 (subsf3_fpx): Likewise.
2309 (mulsf3_fpx): Likewise.
2310 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
2311 asm operand.
2312 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
2313 function attrbutes.
2314 * doc/invoke.texi (ARC): Document mjli-always option.
2316 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
2318 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
2319 avoid addition with 0 and use incw and decw where possible.
2321 2018-01-26 Richard Biener <rguenther@suse.de>
2323 PR tree-optimization/81082
2324 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
2325 association if it requires casting to unsigned.
2326 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
2327 from fold_plusminus_mult_expr to catch important cases late when
2328 range info is available.
2330 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2332 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
2333 * configure.ac (hidden_linkonce): New test.
2334 * configure: Regenerate.
2335 * config.in: Regenerate.
2337 2018-01-26 Julia Koval <julia.koval@intel.com>
2339 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
2340 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
2341 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
2342 _mm_mask_bitshuffle_epi64_mask): Fix type.
2343 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
2344 USI_FTYPE_V4DI_V4DI_USI): Remove.
2345 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
2346 __builtin_ia32_vpshufbitqmb256_mask,
2347 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
2348 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
2349 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
2351 2018-01-26 Alan Modra <amodra@gmail.com>
2353 PR target/84033
2354 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
2355 UNSPEC_VBPERMQ. Sort other unspecs.
2357 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
2359 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
2361 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
2363 PR middle-end/83055
2364 * predict.c (drop_profile): Do not push/pop cfun; update also
2365 node->count.
2366 (handle_missing_profiles): Fix logic looking for zero profiles.
2368 2018-01-25 Jakub Jelinek <jakub@redhat.com>
2370 PR middle-end/83977
2371 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
2372 on functions with #pragma omp declare simd or functions with simd
2373 attribute.
2374 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
2375 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
2376 Remove trailing \n from warning_at calls.
2378 2018-01-25 Tom de Vries <tom@codesourcery.com>
2380 PR target/84028
2381 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2382 for neutered workers.
2384 2018-01-24 Joseph Myers <joseph@codesourcery.com>
2386 PR target/68467
2387 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
2388 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
2390 2018-01-24 Jeff Law <law@redhat.com>
2392 PR target/83994
2393 * i386.c (get_probe_interval): Move to earlier point.
2394 (ix86_compute_frame_layout): If -fstack-clash-protection and
2395 the frame is larger than the probe interval, then use pushes
2396 to save registers rather than reg->mem moves.
2397 (ix86_expand_prologue): Remove conditional for int_registers_saved
2398 assertion.
2400 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
2402 PR target/84014
2403 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
2404 min/max for never referenced object.
2406 2018-01-24 Jakub Jelinek <jakub@redhat.com>
2408 PR middle-end/83977
2409 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
2410 here.
2411 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
2412 attributes from DECL_ATTRIBUTES (decl) without affecting
2413 DECL_ATTRIBUTES (current_function_decl).
2414 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
2415 functions with non-NULL DECL_ABSTRACT_ORIGIN.
2417 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
2419 PR tree-optimization/83979
2420 * fold-const.c (fold_comparison): Use constant_boolean_node
2421 instead of boolean_{true,false}_node.
2423 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
2425 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
2426 with zero counts.
2428 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2430 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2431 Simplify the clause that sets the length attribute.
2432 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2433 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
2434 clause that sets the length attribute.
2435 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2437 2018-01-24 Tom de Vries <tom@codesourcery.com>
2439 PR target/83589
2440 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
2441 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
2442 Add strict parameter.
2443 (prevent_branch_around_nothing): Insert dummy insn between branch to
2444 label and label with no ptx insn inbetween.
2445 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
2447 2018-01-24 Tom de Vries <tom@codesourcery.com>
2449 PR target/81352
2450 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2451 for neutered threads in warp.
2452 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
2454 2018-01-24 Richard Biener <rguenther@suse.de>
2456 PR tree-optimization/83176
2457 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
2458 operands.
2460 2018-01-24 Richard Biener <rguenther@suse.de>
2462 PR tree-optimization/82819
2463 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
2464 code generating pluses that are no-ops in the target precision.
2466 2018-01-24 Richard Biener <rguenther@suse.de>
2468 PR middle-end/84000
2469 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
2471 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2473 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
2474 to merge probabilities.
2475 * predict.c (probably_never_executed): Also mark as cold functions
2476 with global 0 profile and guessed local profile.
2477 * profile-count.c (profile_probability::combine_with_count): New
2478 member function.
2479 * profile-count.h (profile_probability::operator*,
2480 profile_probability::operator*=, profile_probability::operator/,
2481 profile_probability::operator/=): Reduce precision to adjusted
2482 and set value to guessed on contradictory divisions.
2483 (profile_probability::combine_with_freq): Remove.
2484 (profile_probability::combine_wiht_count): Declare.
2485 (profile_count::force_nonzero):: Set to adjusted.
2486 (profile_count::probability_in):: Set quality to adjusted.
2487 * tree-ssa-tail-merge.c (replace_block_by): Use
2488 combine_with_count.
2490 2018-01-23 Andrew Waterman <andrew@sifive.com>
2491 Jim Wilson <jimw@sifive.com>
2493 * config/riscv/riscv.c (riscv_stack_boundary): New.
2494 (riscv_option_override): Set riscv_stack_boundary. Handle
2495 riscv_preferred_stack_boundary_arg.
2496 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
2497 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
2498 (STACK_BOUNDARY): Set to riscv_stack_boundary.
2499 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
2500 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
2501 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
2503 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
2505 PR target/83905
2506 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
2507 of struct ix86_frame.
2508 (ix86_expand_epilogue): Likewise. Add a local variable for
2509 the reg_save_offset field in struct ix86_frame.
2511 2018-01-23 Bin Cheng <bin.cheng@arm.com>
2513 PR tree-optimization/82604
2514 * tree-loop-distribution.c (enum partition_kind): New enum item
2515 PKIND_PARTIAL_MEMSET.
2516 (partition_builtin_p): Support above new enum item.
2517 (generate_code_for_partition): Ditto.
2518 (compute_access_range): Differentiate cases that equality can be
2519 proven at all loops, the innermost loops or no loops.
2520 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
2521 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
2522 (finalize_partitions, distribute_loop): Don't fuse partition of
2523 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
2524 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
2525 parloop is enabled.
2527 2018-01-23 Martin Liska <mliska@suse.cz>
2529 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
2530 order to ignore the predictor.
2531 (PRED_POLYMORPHIC_CALL): Likewise.
2532 (PRED_RECURSIVE_CALL): Likewise.
2534 2018-01-23 Martin Liska <mliska@suse.cz>
2536 * tree-profile.c (tree_profiling): Print function header to
2537 aware reader which function we are working on.
2538 * value-prof.c (gimple_find_values_to_profile): Do not print
2539 not interesting value histograms.
2541 2018-01-23 Martin Liska <mliska@suse.cz>
2543 * profile-count.h (enum profile_quality): Add
2544 profile_uninitialized as the first value. Do not number values
2545 as they are zero based.
2546 (profile_count::verify): Update sanity check.
2547 (profile_probability::verify): Likewise.
2549 2018-01-23 Nathan Sidwell <nathan@acm.org>
2551 * doc/invoke.texi (ffor-scope): Deprecate.
2553 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2555 PR tree-optimization/83510
2556 * domwalk.c (set_all_edges_as_executable): New function.
2557 (dom_walker::dom_walker): Convert bool param
2558 "skip_unreachable_blocks" to enum reachability. Move setup of
2559 edge flags to set_all_edges_as_executable and only do it when
2560 reachability is REACHABLE_BLOCKS.
2561 * domwalk.h (enum dom_walker::reachability): New enum.
2562 (dom_walker::dom_walker): Convert bool param
2563 "skip_unreachable_blocks" to enum reachability.
2564 (set_all_edges_as_executable): New decl.
2565 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2566 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2567 "reachability".
2568 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2569 but converting true to REACHABLE_BLOCKS.
2570 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2571 * tree-vrp.c
2572 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2573 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2574 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2575 REACHABLE_BLOCKS.
2576 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2577 if check_all_array_refs will be called.
2579 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2581 * tree.c (selftest::test_location_wrappers): Add more test
2582 coverage.
2584 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2586 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2587 (selftest::test_bit_in_range): Likewise.
2589 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2591 PR testsuite/83888
2592 * doc/sourcebuild.texi (vect_float): Say that the selector
2593 only describes the situation when -funsafe-math-optimizations is on.
2594 (vect_float_strict): Document.
2596 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2598 PR tree-optimization/83965
2599 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2600 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2601 instead of checking only for a reduction.
2602 (vect_recog_widen_sum_pattern): Likewise.
2604 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2606 * predict.c (probably_never_executed): Only use precise profile info.
2607 (compute_function_frequency): Skip after inlining hack since we now
2608 have quality checking.
2610 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2612 * profile-count.h (profile_probability::very_unlikely,
2613 profile_probability::unlikely, profile_probability::even): Set
2614 precision to guessed.
2616 2018-01-23 Richard Biener <rguenther@suse.de>
2618 PR tree-optimization/83963
2619 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2620 Properly terminate dominator walk when crossing the exit edge not
2621 when visiting its source block.
2623 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2625 PR c++/83918
2626 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2627 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2629 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2631 PR tree-optimization/83957
2632 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2633 semicolon after for body surrounded by braces.
2635 PR tree-optimization/83081
2636 * profile-count.h (profile_probability::split): New method.
2637 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2638 Use profile_probability::split.
2639 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2640 when splitting a single conditional jump into 2.
2642 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2644 PR tree-optimization/69452
2645 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2646 decl.
2648 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2650 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2651 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2652 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2654 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2656 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2657 * config/rl78/rl78.md (movdi): New define_expand.
2658 * config/rl78/rl78.c (rl78_split_movdi): New function.
2660 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2662 PR target/83862
2663 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2664 no longer used.
2665 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2666 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2667 128-bit to produce an UNSPEC move to get the double word with the
2668 signbit and then a shift directly to do signbit.
2669 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2670 implementation with a new version that just does either a direct
2671 move or a regular move. Move memory interface to separate insns.
2672 Move insns so they are next to the expander.
2673 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2674 with signbit move. Split big and little endian case.
2675 (signbit<mode>2_dm_mem_le): Likewise.
2676 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2677 (signbit<mode>2_dm2): Likewise.
2679 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2681 * config/rl78/rl78.md (anddi3): New define_expand.
2683 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2685 * config/rl78/rl78.md (umindi3): New define_expand.
2687 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2689 * config/rl78/rl78.md (smindi3): New define_expand.
2691 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2693 * config/rl78/rl78.md (smaxdi3): New define_expand.
2695 2018-01-22 Carl Love <cel@us.ibm.com>
2697 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2698 LVX_V1TI): Add macro expansion.
2699 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2700 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2701 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2702 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2703 Change check to determine if the instruction is a byte reversing
2704 entry. Fix typo in comment.
2705 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2706 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2707 Add def_builtin calls for new builtins.
2708 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2709 Add define_insn expansion.
2711 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2713 * config/rl78/rl78.md (umaxdi3): New define_expand.
2715 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2717 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2718 for non-QImode registers.
2720 2018-01-22 Richard Biener <rguenther@suse.de>
2722 PR tree-optimization/83963
2723 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2724 including the loop exit block.
2725 (scop_detection::merge_sese): Likewise.
2726 (scop_detection::add_scop): Do it here instead.
2728 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2730 * doc/sourcebuild.texi (arm_softfloat): Document.
2732 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2734 PR gcc/77734
2735 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2736 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2737 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2739 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2740 David Edelsohn <dje.gcc@gmail.com>
2742 PR target/83946
2743 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2744 Change "crset eq" to "crset 2".
2745 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2746 (*call_indirect_aix<mode>_nospec): Likewise.
2747 (*call_value_indirect_aix<mode>_nospec): Likewise.
2748 (*call_indirect_elfv2<mode>_nospec): Likewise.
2749 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2750 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2751 change assembly output from . to $.
2752 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2753 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2754 (*tablejump<mode>_internal1_nospec): Likewise.
2756 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2758 PR target/80870
2759 * config/sh/sh_optimize_sett_clrt.cc:
2760 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2762 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2764 PR tree-optimization/83940
2765 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2766 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2767 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2768 use it to pass back the definition type.
2769 (vect_check_store_rhs): Likewise rhs_dt_out.
2770 (vect_build_gather_load_calls): Add a mask_dt argument and use
2771 it instead of a call to vect_is_simple_use.
2772 (vectorizable_store): Update calls to vect_check_load_store_mask
2773 and vect_check_store_rhs. Use the dt returned by the latter instead
2774 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2775 instead of calls to vect_is_simple_use. Pass the scalar rather
2776 than the vector operand to vect_is_simple_use when handling
2777 second and subsequent copies of an rhs value.
2778 (vectorizable_load): Update calls to vect_check_load_store_mask
2779 and vect_build_gather_load_calls. Use the cached mask_dt and
2780 gs_info.offset_dt instead of calls to vect_is_simple_use.
2782 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2784 PR middle-end/83945
2785 * tree-emutls.c: Include gimplify.h.
2786 (lower_emutls_2): New function.
2787 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2788 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2789 it before further processing.
2791 PR target/83930
2792 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2793 UINTVAL (trueop1) instead of INTVAL (op1).
2795 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2797 PR debug/81570
2798 PR debug/83728
2799 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2800 INCOMING_FRAME_SP_OFFSET if not defined.
2801 (scan_trace): Add ENTRY argument. If true and
2802 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2803 emit a note to adjust the CFA offset.
2804 (create_cfi_notes): Adjust scan_trace callers.
2805 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2806 INCOMING_FRAME_SP_OFFSET in the CIE.
2807 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2808 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2809 Likewise.
2810 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2811 * doc/tm.texi: Regenerated.
2813 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2815 PR rtl-optimization/83147
2816 * lra-constraints.c (remove_inheritance_pseudos): Use
2817 lra_substitute_pseudo_within_insn.
2819 2018-01-19 Tom de Vries <tom@codesourcery.com>
2820 Cesar Philippidis <cesar@codesourcery.com>
2822 PR target/83920
2823 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2825 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2827 PR target/83790
2828 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2829 spaces for function labels.
2831 2018-01-19 Martin Liska <mliska@suse.cz>
2833 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2834 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2835 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2836 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2837 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2838 (PRED_CONST_RETURN): Change from 69 to 65.
2839 (PRED_NULL_RETURN): Change from 91 to 71.
2840 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2841 (PRED_LOOP_GUARD): Change from 66 to 73.
2843 2018-01-19 Martin Liska <mliska@suse.cz>
2845 * predict.c (predict_insn_def): Add new assert.
2846 (struct branch_predictor): Change type to signed integer.
2847 (test_prediction_value_range): Amend test to cover
2848 PROB_UNINITIALIZED.
2849 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2850 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2851 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2852 (PRED_LOOP_IV_COMPARE): Likewise.
2853 * predict.h (PROB_UNINITIALIZED): Define new constant.
2855 2018-01-19 Martin Liska <mliska@suse.cz>
2857 * predict.c (dump_prediction): Add new format for
2858 analyze_brprob.py script which is enabled with -details
2859 suboption.
2860 * profile-count.h (precise_p): New function.
2862 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2864 PR tree-optimization/83922
2865 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2866 there are no statements that need masking.
2867 (vect_active_double_reduction_p): New function.
2868 (vect_analyze_loop_operations): Use it when handling phis that
2869 are not in the loop header.
2871 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2873 PR tree-optimization/83914
2874 * tree-vect-loop.c (vectorizable_induction): Don't convert
2875 init_expr or apply the peeling adjustment for inductions
2876 that are nested within the vectorized loop.
2878 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2880 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2881 instead of NEG.
2883 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2885 PR sanitizer/81715
2886 PR testsuite/83882
2887 * function.h (gimplify_parameters): Add gimple_seq * argument.
2888 * function.c: Include gimple.h and options.h.
2889 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2890 for the added local temporaries if needed.
2891 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2892 if there are any parameter cleanups, wrap whole body into a
2893 try/finally with the cleanups.
2895 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2897 PR target/82964
2898 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2899 Use GET_MODE_CLASS for scalar floating point.
2901 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2903 PR ipa/82256
2904 patch by PaX Team
2905 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2906 Fix call of call_cgraph_insertion_hooks.
2908 2018-01-18 Martin Sebor <msebor@redhat.com>
2910 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2912 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2914 PR ipa/83619
2915 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2916 frequencies.
2918 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2920 PR other/70268
2921 * common.opt: (-ffile-prefix-map): New option.
2922 * opts.c (common_handle_option): Defer it.
2923 * opts-global.c (handle_common_deferred_options): Handle it.
2924 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2925 * file-prefix-map.h: New file.
2926 (remap_debug_filename, add_debug_prefix_map): ...here.
2927 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2928 * final.c (debug_prefix_map, add_debug_prefix_map
2929 remap_debug_filename): Move to...
2930 * file-prefix-map.c: New file.
2931 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2932 generalize, get rid of alloca(), use strrchr() instead of strchr().
2933 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2934 Implement in terms of add_prefix_map().
2935 (remap_macro_filename, remap_debug_filename): Implement in term of
2936 remap_filename().
2937 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2938 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2939 * dbxout.c: Include file-prefix-map.h.
2940 * varasm.c: Likewise.
2941 * vmsdbgout.c: Likewise.
2942 * xcoffout.c: Likewise.
2943 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2944 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2945 * doc/invoke.texi (-ffile-prefix-map): Document.
2946 (-fdebug-prefix-map): Update description.
2948 2018-01-18 Martin Liska <mliska@suse.cz>
2950 * config/i386/i386.c (indirect_thunk_name): Document that also
2951 lfence is emitted.
2952 (output_indirect_thunk): Document why both instructions
2953 (pause and lfence) are generated.
2955 2018-01-18 Richard Biener <rguenther@suse.de>
2957 PR tree-optimization/83887
2958 * graphite-scop-detection.c
2959 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2960 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2961 (scop_detection::merge_sese): Re-implement with a flood-fill
2962 algorithm that properly finds a SESE region if it exists.
2964 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2966 PR c/61240
2967 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2968 pointer_diff optimizations use view_convert instead of convert.
2970 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2972 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2973 Generate different code for -mno-speculate-indirect-jumps.
2974 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2975 (*call_indirect_aix<mode>): Disable for
2976 -mno-speculate-indirect-jumps.
2977 (*call_indirect_aix<mode>_nospec): New define_insn.
2978 (*call_value_indirect_aix<mode>): Disable for
2979 -mno-speculate-indirect-jumps.
2980 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2981 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2982 -mno-speculate-indirect-jumps.
2983 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2985 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2987 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2988 long double type, set the flags for noting the default long double
2989 type, even if we don't pass or return a long double type.
2991 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2993 PR ipa/83051
2994 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2995 failure.
2997 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2999 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
3000 support for merge[hl].
3001 (fold_mergehl_helper): New helper function.
3002 (tree-vector-builder.h): New #include for tree_vector_builder usage.
3003 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
3004 (altivec_vmrglw_direct): Add xxmrglw insn.
3006 2018-01-17 Andrew Waterman <andrew@sifive.com>
3008 * config/riscv/riscv.c (riscv_conditional_register_usage): If
3009 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
3011 2018-01-17 David Malcolm <dmalcolm@redhat.com>
3013 PR lto/83121
3014 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
3015 call the lto_location_cache before reading the
3016 DECL_SOURCE_LOCATION of the types.
3018 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
3019 Richard Sandiford <richard.sandiford@linaro.org>
3021 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
3022 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
3023 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
3024 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
3025 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
3026 Add declaration.
3027 * config/aarch64/constraints.md (aarch64_movti_operand):
3028 Limit immediates.
3029 * config/aarch64/predicates.md (Uti): Add new constraint.
3031 2018-01-17 Carl Love <cel@us.ibm.com>
3033 * config/rs6000/vsx.md (define_expand xl_len_r,
3034 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
3035 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
3036 lxvll.
3037 (define_expand, define_insn): Move the shift left from the
3038 define_insn to the define_expand for lxvl and stxvl instructions.
3039 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
3040 and XL_LEN_R definitions to PURE.
3042 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
3044 * config/i386/i386.c (indirect_thunk_name): Declare regno
3045 as unsigned int. Compare regno with INVALID_REGNUM.
3046 (output_indirect_thunk): Ditto.
3047 (output_indirect_thunk_function): Ditto.
3048 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
3049 in the call to output_indirect_thunk_function.
3051 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
3053 PR middle-end/83884
3054 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
3055 rather than the size of inner_type to determine the stack slot size
3056 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
3058 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
3060 PR target/83546
3061 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
3062 to PTA_SILVERMONT.
3064 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
3066 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
3067 endian Linux systems to optionally enable multilibs for selecting
3068 the long double type if the user configured an explicit type.
3069 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
3070 have no long double multilibs if not defined.
3071 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
3072 warn if the user used -mabi={ieee,ibm}longdouble and we built
3073 multilibs for long double.
3074 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
3075 appropriate multilib option.
3076 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
3077 multilib options.
3078 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
3079 for building long double multilibs.
3080 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
3082 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
3084 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
3085 copies.
3087 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
3088 64 bits.
3089 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
3090 128 bits.
3092 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
3093 variables.
3095 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
3096 return value.
3098 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3100 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
3101 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
3103 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
3105 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
3106 different rtl trees depending on TARGET_64BIT.
3107 (rs6000_gen_lvx): Likewise.
3109 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3111 * config/visium/visium.md (nop): Tweak comment.
3112 (hazard_nop): Likewise.
3114 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3116 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
3117 -mspeculate-indirect-jumps.
3118 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
3119 for -mno-speculate-indirect-jumps.
3120 (*call_indirect_elfv2<mode>_nospec): New define_insn.
3121 (*call_value_indirect_elfv2<mode>): Disable for
3122 -mno-speculate-indirect-jumps.
3123 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
3124 (indirect_jump): Emit different RTL for
3125 -mno-speculate-indirect-jumps.
3126 (*indirect_jump<mode>): Disable for
3127 -mno-speculate-indirect-jumps.
3128 (*indirect_jump<mode>_nospec): New define_insn.
3129 (tablejump): Emit different RTL for
3130 -mno-speculate-indirect-jumps.
3131 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
3132 (tablejumpsi_nospec): New define_expand.
3133 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
3134 (tablejumpdi_nospec): New define_expand.
3135 (*tablejump<mode>_internal1): Disable for
3136 -mno-speculate-indirect-jumps.
3137 (*tablejump<mode>_internal1_nospec): New define_insn.
3138 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
3139 option.
3141 2018-01-16 Artyom Skrobov tyomitch@gmail.com
3143 * caller-save.c (insert_save): Drop unnecessary parameter. All
3144 callers updated.
3146 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3147 Richard Biener <rguenth@suse.de>
3149 PR libgomp/83590
3150 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
3151 return early, inline manually is_gimple_sizepos. Make sure if we
3152 call gimplify_expr we don't end up with a gimple constant.
3153 * tree.c (variably_modified_type_p): Don't return true for
3154 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
3155 * gimplify.h (is_gimple_sizepos): Remove.
3157 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3159 PR tree-optimization/83857
3160 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
3161 vectorizable_live_operation for pure SLP statements.
3162 (vectorizable_live_operation): Handle PHIs.
3164 2018-01-16 Richard Biener <rguenther@suse.de>
3166 PR tree-optimization/83867
3167 * tree-vect-stmts.c (vect_transform_stmt): Precompute
3168 nested_in_vect_loop_p since the scalar stmt may get invalidated.
3170 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3172 PR c/83844
3173 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
3174 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
3175 If off is not INTEGER_CST, issue a may not be aligned warning
3176 rather than isn't aligned. Use isn%'t rather than isn't.
3177 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
3178 into MULT_EXPR.
3179 <case MULT_EXPR>: Improve the case when bottom and one of the
3180 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
3181 operand, in that case check if the other operand is multiple of
3182 bottom divided by the INTEGER_CST operand.
3184 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3186 PR target/83858
3187 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
3188 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
3189 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
3190 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
3191 * config/pa/pa.c (pa_function_arg_advance): Likewise.
3192 (pa_function_arg, pa_arg_partial_bytes): Likewise.
3193 (pa_function_arg_size): New function.
3195 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3197 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
3198 in a separate statement.
3200 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3202 PR tree-optimization/83847
3203 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
3204 group gathers and scatters.
3206 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3208 PR rtl-optimization/86620
3209 * params.def (max-sched-ready-insns): Bump minimum value to 1.
3211 PR rtl-optimization/83213
3212 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
3213 to last if both are JUMP_INSNs.
3215 PR tree-optimization/83843
3216 * gimple-ssa-store-merging.c
3217 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
3218 store_immediate_info for bswap/nop orig_stores.
3220 2018-01-15 Andrew Waterman <andrew@sifive.com>
3222 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
3223 !TARGET_MUL.
3224 <UDIV>: Increase cost if !TARGET_DIV.
3226 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
3228 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
3229 (define_attr "cr_logical_3op"): New.
3230 (cceq_ior_compare): Adjust.
3231 (cceq_ior_compare_complement): Adjust.
3232 (*cceq_rev_compare): Adjust.
3233 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
3234 (is_cracked_insn): Adjust.
3235 (insn_must_be_first_in_group): Adjust.
3236 * config/rs6000/40x.md: Adjust.
3237 * config/rs6000/440.md: Adjust.
3238 * config/rs6000/476.md: Adjust.
3239 * config/rs6000/601.md: Adjust.
3240 * config/rs6000/603.md: Adjust.
3241 * config/rs6000/6xx.md: Adjust.
3242 * config/rs6000/7450.md: Adjust.
3243 * config/rs6000/7xx.md: Adjust.
3244 * config/rs6000/8540.md: Adjust.
3245 * config/rs6000/cell.md: Adjust.
3246 * config/rs6000/e300c2c3.md: Adjust.
3247 * config/rs6000/e500mc.md: Adjust.
3248 * config/rs6000/e500mc64.md: Adjust.
3249 * config/rs6000/e5500.md: Adjust.
3250 * config/rs6000/e6500.md: Adjust.
3251 * config/rs6000/mpc.md: Adjust.
3252 * config/rs6000/power4.md: Adjust.
3253 * config/rs6000/power5.md: Adjust.
3254 * config/rs6000/power6.md: Adjust.
3255 * config/rs6000/power7.md: Adjust.
3256 * config/rs6000/power8.md: Adjust.
3257 * config/rs6000/power9.md: Adjust.
3258 * config/rs6000/rs64.md: Adjust.
3259 * config/rs6000/titan.md: Adjust.
3261 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3263 * config/i386/predicates.md (indirect_branch_operand): Rewrite
3264 ix86_indirect_branch_register logic.
3266 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3268 * config/i386/constraints.md (Bs): Update
3269 ix86_indirect_branch_register check. Don't check
3270 ix86_indirect_branch_register with GOT_memory_operand.
3271 (Bw): Likewise.
3272 * config/i386/predicates.md (GOT_memory_operand): Don't check
3273 ix86_indirect_branch_register here.
3274 (GOT32_symbol_operand): Likewise.
3276 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3278 * config/i386/predicates.md (constant_call_address_operand):
3279 Rewrite ix86_indirect_branch_register logic.
3280 (sibcall_insn_operand): Likewise.
3282 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3284 * config/i386/constraints.md (Bs): Replace
3285 ix86_indirect_branch_thunk_register with
3286 ix86_indirect_branch_register.
3287 (Bw): Likewise.
3288 * config/i386/i386.md (indirect_jump): Likewise.
3289 (tablejump): Likewise.
3290 (*sibcall_memory): Likewise.
3291 (*sibcall_value_memory): Likewise.
3292 Peepholes of indirect call and jump via memory: Likewise.
3293 * config/i386/i386.opt: Likewise.
3294 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3295 (GOT_memory_operand): Likewise.
3296 (call_insn_operand): Likewise.
3297 (sibcall_insn_operand): Likewise.
3298 (GOT32_symbol_operand): Likewise.
3300 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3302 PR middle-end/83837
3303 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
3304 type rather than type addr's type points to.
3305 (expand_omp_atomic_mutex): Likewise.
3306 (expand_omp_atomic): Likewise.
3308 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3310 PR target/83839
3311 * config/i386/i386.c (output_indirect_thunk_function): Use
3312 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
3313 for __x86_return_thunk.
3315 2018-01-15 Richard Biener <rguenther@suse.de>
3317 PR middle-end/83850
3318 * expmed.c (extract_bit_field_1): Fix typo.
3320 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3322 PR target/83687
3323 * config/arm/iterators.md (VF): New mode iterator.
3324 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
3325 Remove integer-related logic from pattern.
3326 (neon_vabd<mode>_3): Likewise.
3328 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3330 PR middle-end/82694
3331 * common.opt (fstrict-overflow): No longer an alias.
3332 (fwrapv-pointer): New option.
3333 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
3334 also for pointer types based on flag_wrapv_pointer.
3335 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
3336 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
3337 opts->x_flag_wrapv got set.
3338 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
3339 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
3340 POINTER_TYPE_OVERFLOW_UNDEFINED.
3341 * match.pd: Likewise in address comparison pattern.
3342 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
3344 2018-01-15 Richard Biener <rguenther@suse.de>
3346 PR lto/83804
3347 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
3348 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
3349 Reset type names to their identifier if their TYPE_DECL doesn't
3350 have linkage (and thus is used for ODR and devirt).
3351 (save_debug_info_for_decl): Remove.
3352 (save_debug_info_for_type): Likewise.
3353 (add_tree_to_fld_list): Adjust.
3354 * tree-pretty-print.c (dump_generic_node): Make dumping of
3355 type names more robust.
3357 2018-01-15 Richard Biener <rguenther@suse.de>
3359 * BASE-VER: Bump to 8.0.1.
3361 2018-01-14 Martin Sebor <msebor@redhat.com>
3363 PR other/83508
3364 * builtins.c (check_access): Avoid warning when the no-warning bit
3365 is set.
3367 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
3369 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
3370 * ira-color (allocno_hard_regs_compare): Likewise.
3372 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
3374 PR target/83013
3375 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
3376 Use .pushsection/.popsection.
3378 2018-01-14 Martin Sebor <msebor@redhat.com>
3380 PR c++/81327
3381 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
3383 2018-01-14 Jakub Jelinek <jakub@redhat.com>
3385 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
3386 entry from extra_headers.
3387 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
3388 extra_headers, make the list bitwise identical to the i?86-*-* one.
3390 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3392 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
3393 -mcmodel=large with -mindirect-branch=thunk,
3394 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
3395 -mfunction-return=thunk-extern.
3396 * doc/invoke.texi: Document -mcmodel=large is incompatible with
3397 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
3398 -mfunction-return=thunk and -mfunction-return=thunk-extern.
3400 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3402 * config/i386/i386.c (print_reg): Print the name of the full
3403 integer register without '%'.
3404 (ix86_print_operand): Handle 'V'.
3405 * doc/extend.texi: Document 'V' modifier.
3407 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3409 * config/i386/constraints.md (Bs): Disallow memory operand for
3410 -mindirect-branch-register.
3411 (Bw): Likewise.
3412 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3413 (GOT_memory_operand): Likewise.
3414 (call_insn_operand): Likewise.
3415 (sibcall_insn_operand): Likewise.
3416 (GOT32_symbol_operand): Likewise.
3417 * config/i386/i386.md (indirect_jump): Call convert_memory_address
3418 for -mindirect-branch-register.
3419 (tablejump): Likewise.
3420 (*sibcall_memory): Likewise.
3421 (*sibcall_value_memory): Likewise.
3422 Disallow peepholes of indirect call and jump via memory for
3423 -mindirect-branch-register.
3424 (*call_pop): Replace m with Bw.
3425 (*call_value_pop): Likewise.
3426 (*sibcall_pop_memory): Replace m with Bs.
3427 * config/i386/i386.opt (mindirect-branch-register): New option.
3428 * doc/invoke.texi: Document -mindirect-branch-register option.
3430 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3432 * config/i386/i386-protos.h (ix86_output_function_return): New.
3433 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
3434 set function_return_type.
3435 (indirect_thunk_name): Add ret_p to indicate thunk for function
3436 return.
3437 (output_indirect_thunk_function): Pass false to
3438 indirect_thunk_name.
3439 (ix86_output_indirect_branch_via_reg): Likewise.
3440 (ix86_output_indirect_branch_via_push): Likewise.
3441 (output_indirect_thunk_function): Create alias for function
3442 return thunk if regno < 0.
3443 (ix86_output_function_return): New function.
3444 (ix86_handle_fndecl_attribute): Handle function_return.
3445 (ix86_attribute_table): Add function_return.
3446 * config/i386/i386.h (machine_function): Add
3447 function_return_type.
3448 * config/i386/i386.md (simple_return_internal): Use
3449 ix86_output_function_return.
3450 (simple_return_internal_long): Likewise.
3451 * config/i386/i386.opt (mfunction-return=): New option.
3452 (indirect_branch): Mention -mfunction-return=.
3453 * doc/extend.texi: Document function_return function attribute.
3454 * doc/invoke.texi: Document -mfunction-return= option.
3456 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3458 * config/i386/i386-opts.h (indirect_branch): New.
3459 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
3460 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
3461 with local indirect jump when converting indirect call and jump.
3462 (ix86_set_indirect_branch_type): New.
3463 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
3464 (indirectlabelno): New.
3465 (indirect_thunk_needed): Likewise.
3466 (indirect_thunk_bnd_needed): Likewise.
3467 (indirect_thunks_used): Likewise.
3468 (indirect_thunks_bnd_used): Likewise.
3469 (INDIRECT_LABEL): Likewise.
3470 (indirect_thunk_name): Likewise.
3471 (output_indirect_thunk): Likewise.
3472 (output_indirect_thunk_function): Likewise.
3473 (ix86_output_indirect_branch_via_reg): Likewise.
3474 (ix86_output_indirect_branch_via_push): Likewise.
3475 (ix86_output_indirect_branch): Likewise.
3476 (ix86_output_indirect_jmp): Likewise.
3477 (ix86_code_end): Call output_indirect_thunk_function if needed.
3478 (ix86_output_call_insn): Call ix86_output_indirect_branch if
3479 needed.
3480 (ix86_handle_fndecl_attribute): Handle indirect_branch.
3481 (ix86_attribute_table): Add indirect_branch.
3482 * config/i386/i386.h (machine_function): Add indirect_branch_type
3483 and has_local_indirect_jump.
3484 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
3485 to true.
3486 (tablejump): Likewise.
3487 (*indirect_jump): Use ix86_output_indirect_jmp.
3488 (*tablejump_1): Likewise.
3489 (simple_return_indirect_internal): Likewise.
3490 * config/i386/i386.opt (mindirect-branch=): New option.
3491 (indirect_branch): New.
3492 (keep): Likewise.
3493 (thunk): Likewise.
3494 (thunk-inline): Likewise.
3495 (thunk-extern): Likewise.
3496 * doc/extend.texi: Document indirect_branch function attribute.
3497 * doc/invoke.texi: Document -mindirect-branch= option.
3499 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
3501 PR ipa/83051
3502 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
3504 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
3506 * ipa-inline.c (want_inline_small_function_p): Return false if
3507 inlining has already failed with CIF_FINAL_ERROR.
3508 (update_caller_keys): Call want_inline_small_function_p before
3509 can_inline_edge_p.
3510 (update_callee_keys): Likewise.
3512 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
3514 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
3515 New function.
3516 (rs6000_quadword_masked_address_p): Likewise.
3517 (quad_aligned_load_p): Likewise.
3518 (quad_aligned_store_p): Likewise.
3519 (const_load_sequence_p): Add comment to describe the outer-most loop.
3520 (mimic_memory_attributes_and_flags): New function.
3521 (rs6000_gen_stvx): Likewise.
3522 (replace_swapped_aligned_store): Likewise.
3523 (rs6000_gen_lvx): Likewise.
3524 (replace_swapped_aligned_load): Likewise.
3525 (replace_swapped_load_constant): Capitalize argument name in
3526 comment describing this function.
3527 (rs6000_analyze_swaps): Add a third pass to search for vector loads
3528 and stores that access quad-word aligned addresses and replace
3529 with stvx or lvx instructions when appropriate.
3530 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
3531 New function prototype.
3532 (rs6000_quadword_masked_address_p): Likewise.
3533 (rs6000_gen_lvx): Likewise.
3534 (rs6000_gen_stvx): Likewise.
3535 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
3536 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3537 when memory address is aligned.
3538 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3539 this split to select lvx instruction when memory address is aligned.
3540 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3541 instruction when memory address is aligned.
3542 (*vsx_le_perm_load_v16qi): Likewise.
3543 (four unnamed splitters): Modify to select the stvx instruction
3544 when memory is aligned.
3546 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3548 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3549 which appears in the queue multiple times.
3551 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3552 Alan Hayward <alan.hayward@arm.com>
3553 David Sherwood <david.sherwood@arm.com>
3555 * tree-vectorizer.h (vec_lower_bound): New structure.
3556 (_loop_vec_info): Add check_nonzero and lower_bounds.
3557 (LOOP_VINFO_CHECK_NONZERO): New macro.
3558 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3559 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3560 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3561 fields. Make seg_len the distance travelled, not including the
3562 access size.
3563 (dr_direction_indicator): Declare.
3564 (dr_zero_step_indicator): Likewise.
3565 (dr_known_forward_stride_p): Likewise.
3566 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3567 tree-ssanames.h.
3568 (runtime_alias_check_p): Allow runtime alias checks with
3569 variable strides.
3570 (operator ==): Compare access_size and align.
3571 (prune_runtime_alias_test_list): Rework for new distinction between
3572 the access_size and seg_len.
3573 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3574 segment lengths.
3575 (get_segment_min_max): New function.
3576 (create_intersect_range_checks): Use it.
3577 (dr_step_indicator): New function.
3578 (dr_direction_indicator): Likewise.
3579 (dr_zero_step_indicator): Likewise.
3580 (dr_known_forward_stride_p): Likewise.
3581 * tree-loop-distribution.c (data_ref_segment_size): Return
3582 DR_STEP * (niters - 1).
3583 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3584 constructor.
3585 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3586 (vect_preserves_scalar_order_p): New function, split out from...
3587 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3588 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3589 (vect_vfa_access_size): New function.
3590 (vect_vfa_align): Likewise.
3591 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3592 (dump_lower_bound): New function.
3593 (vect_check_lower_bound): Likewise.
3594 (vect_small_gap_p): Likewise.
3595 (vectorizable_with_step_bound_p): Likewise.
3596 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3597 depencies if the vectorization factor is 1. Convert the checks
3598 for nonzero steps into checks on the bounds of DR_STEP. Try using
3599 a bunds check for variable steps if the minimum required step is
3600 relatively small. Update calls to the dr_with_seg_len
3601 constructor and to vect_compile_time_alias.
3602 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3603 function.
3604 (vect_loop_versioning): Call it.
3605 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3606 when retrying.
3607 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3609 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3610 Alan Hayward <alan.hayward@arm.com>
3611 David Sherwood <david.sherwood@arm.com>
3613 * doc/sourcebuild.texi (vect_scatter_store): Document.
3614 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3615 optabs.
3616 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3617 Document.
3618 * genopinit.c (main): Add supports_vec_scatter_store and
3619 supports_vec_scatter_store_cached to target_optabs.
3620 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3621 IFN_MASK_SCATTER_STORE.
3622 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3623 functions.
3624 * internal-fn.h (internal_store_fn_p): Declare.
3625 (internal_fn_stored_value_index): Likewise.
3626 * internal-fn.c (scatter_store_direct): New macro.
3627 (expand_scatter_store_optab_fn): New function.
3628 (direct_scatter_store_optab_supported_p): New macro.
3629 (internal_store_fn_p): New function.
3630 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3631 IFN_MASK_SCATTER_STORE.
3632 (internal_fn_mask_index): Likewise.
3633 (internal_fn_stored_value_index): New function.
3634 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3635 for scatter stores.
3636 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3637 * optabs-query.c (supports_vec_scatter_store_p): New function.
3638 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3639 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3640 true for scatter stores.
3641 (vect_gather_scatter_fn_p): Handle scatter stores too.
3642 (vect_check_gather_scatter): Consider using scatter stores if
3643 supports_vec_scatter_store_p.
3644 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3645 scatter stores too.
3646 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3647 internal_fn_stored_value_index.
3648 (check_load_store_masking): Handle scatter stores too.
3649 (vect_get_store_rhs): Make public.
3650 (vectorizable_call): Use internal_store_fn_p.
3651 (vectorizable_store): Handle scatter store internal functions.
3652 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3653 when deciding whether the end of the group has been reached.
3654 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3655 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3656 (mask_scatter_store<mode>): New insns.
3658 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3659 Alan Hayward <alan.hayward@arm.com>
3660 David Sherwood <david.sherwood@arm.com>
3662 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3663 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3664 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3665 function.
3666 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3667 Use vect_truncate_gather_scatter_offset if we can't treat the
3668 operation as a normal gather load or scatter store.
3669 (get_group_load_store_type): Take the gather_scatter_info
3670 as argument. Try using a gather load or scatter store for
3671 single-element groups.
3672 (get_load_store_type): Update calls to get_group_load_store_type
3673 and vect_use_strided_gather_scatters_p.
3675 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3676 Alan Hayward <alan.hayward@arm.com>
3677 David Sherwood <david.sherwood@arm.com>
3679 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3680 optional tree argument.
3681 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3682 null target hooks.
3683 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3684 but continue to use the current value as a fallback.
3685 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3686 to compare the updates.
3687 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3688 (get_load_store_type): Use it when handling a strided access.
3689 (vect_get_strided_load_store_ops): New function.
3690 (vect_get_data_ptr_increment): Likewise.
3691 (vectorizable_load): Handle strided gather loads. Always pass
3692 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3694 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3695 Alan Hayward <alan.hayward@arm.com>
3696 David Sherwood <david.sherwood@arm.com>
3698 * doc/md.texi (gather_load@var{m}): Document.
3699 (mask_gather_load@var{m}): Likewise.
3700 * genopinit.c (main): Add supports_vec_gather_load and
3701 supports_vec_gather_load_cached to target_optabs.
3702 * optabs-tree.c (init_tree_optimization_optabs): Use
3703 ggc_cleared_alloc to allocate target_optabs.
3704 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3705 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3706 functions.
3707 * internal-fn.h (internal_load_fn_p): Declare.
3708 (internal_gather_scatter_fn_p): Likewise.
3709 (internal_fn_mask_index): Likewise.
3710 (internal_gather_scatter_fn_supported_p): Likewise.
3711 * internal-fn.c (gather_load_direct): New macro.
3712 (expand_gather_load_optab_fn): New function.
3713 (direct_gather_load_optab_supported_p): New macro.
3714 (direct_internal_fn_optab): New function.
3715 (internal_load_fn_p): Likewise.
3716 (internal_gather_scatter_fn_p): Likewise.
3717 (internal_fn_mask_index): Likewise.
3718 (internal_gather_scatter_fn_supported_p): Likewise.
3719 * optabs-query.c (supports_at_least_one_mode_p): New function.
3720 (supports_vec_gather_load_p): Likewise.
3721 * optabs-query.h (supports_vec_gather_load_p): Declare.
3722 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3723 and memory_type field.
3724 (NUM_PATTERNS): Bump to 15.
3725 * tree-vect-data-refs.c: Include internal-fn.h.
3726 (vect_gather_scatter_fn_p): New function.
3727 (vect_describe_gather_scatter_call): Likewise.
3728 (vect_check_gather_scatter): Try using internal functions for
3729 gather loads. Recognize existing calls to a gather load function.
3730 (vect_analyze_data_refs): Consider using gather loads if
3731 supports_vec_gather_load_p.
3732 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3733 (vect_get_gather_scatter_offset_type): Likewise.
3734 (vect_convert_mask_for_vectype): Likewise.
3735 (vect_add_conversion_to_patterm): Likewise.
3736 (vect_try_gather_scatter_pattern): Likewise.
3737 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3738 (vect_vect_recog_func_ptrs): Add it.
3739 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3740 internal_fn_mask_index and internal_gather_scatter_fn_p.
3741 (check_load_store_masking): Take the gather_scatter_info as an
3742 argument and handle gather loads.
3743 (vect_get_gather_scatter_ops): New function.
3744 (vectorizable_call): Check internal_load_fn_p.
3745 (vectorizable_load): Likewise. Handle gather load internal
3746 functions.
3747 (vectorizable_store): Update call to check_load_store_masking.
3748 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3749 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3750 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3751 (aarch64_gather_scale_operand_d): New predicates.
3752 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3753 (mask_gather_load<mode>): New insns.
3755 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3756 Alan Hayward <alan.hayward@arm.com>
3757 David Sherwood <david.sherwood@arm.com>
3759 * optabs.def (fold_left_plus_optab): New optab.
3760 * doc/md.texi (fold_left_plus_@var{m}): Document.
3761 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3762 * internal-fn.c (fold_left_direct): Define.
3763 (expand_fold_left_optab_fn): Likewise.
3764 (direct_fold_left_optab_supported_p): Likewise.
3765 * fold-const-call.c (fold_const_fold_left): New function.
3766 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3767 * tree-parloops.c (valid_reduction_p): New function.
3768 (gather_scalar_reductions): Use it.
3769 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3770 (vect_finish_replace_stmt): Declare.
3771 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3772 (needs_fold_left_reduction_p): New function, split out from...
3773 (vect_is_simple_reduction): ...here. Accept reductions that
3774 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3775 (vect_force_simple_reduction): Also store the reduction type in
3776 the assignment's STMT_VINFO_REDUC_TYPE.
3777 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3778 (merge_with_identity): New function.
3779 (vect_expand_fold_left): Likewise.
3780 (vectorize_fold_left_reduction): Likewise.
3781 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3782 scalar phi in place for it. Check for target support and reject
3783 cases that would reassociate the operation. Defer the transform
3784 phase to vectorize_fold_left_reduction.
3785 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3786 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3787 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3789 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3791 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3792 call to ifc_temp_var.
3794 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3795 Alan Hayward <alan.hayward@arm.com>
3796 David Sherwood <david.sherwood@arm.com>
3798 * target.def (legitimize_address_displacement): Take the original
3799 offset as a poly_int.
3800 * targhooks.h (default_legitimize_address_displacement): Update
3801 accordingly.
3802 * targhooks.c (default_legitimize_address_displacement): Likewise.
3803 * doc/tm.texi: Regenerate.
3804 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3805 as an argument, moving assert of ad->disp == ad->disp_term to...
3806 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3807 Try calling targetm.legitimize_address_displacement before expanding
3808 the address rather than afterwards, and adjust for the new interface.
3809 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3810 Match the new hook interface. Handle SVE addresses.
3811 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3812 new hook interface.
3814 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3816 * Makefile.in (OBJS): Add early-remat.o.
3817 * target.def (select_early_remat_modes): New hook.
3818 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3819 * doc/tm.texi: Regenerate.
3820 * targhooks.h (default_select_early_remat_modes): Declare.
3821 * targhooks.c (default_select_early_remat_modes): New function.
3822 * timevar.def (TV_EARLY_REMAT): New timevar.
3823 * passes.def (pass_early_remat): New pass.
3824 * tree-pass.h (make_pass_early_remat): Declare.
3825 * early-remat.c: New file.
3826 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3827 function.
3828 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3830 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3831 Alan Hayward <alan.hayward@arm.com>
3832 David Sherwood <david.sherwood@arm.com>
3834 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3835 vfm1 with a bound_epilog parameter.
3836 (vect_do_peeling): Update calls accordingly, and move the prologue
3837 call earlier in the function. Treat the base bound_epilog as 0 for
3838 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3839 this base when peeling for gaps.
3840 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3841 with fully-masked loops.
3842 (vect_estimate_min_profitable_iters): Handle the single peeled
3843 iteration in that case.
3845 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3846 Alan Hayward <alan.hayward@arm.com>
3847 David Sherwood <david.sherwood@arm.com>
3849 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3850 single-element interleaving even if the size is not a power of 2.
3851 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3852 accesses for single-element interleaving if the group size is
3853 not a power of 2.
3855 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3856 Alan Hayward <alan.hayward@arm.com>
3857 David Sherwood <david.sherwood@arm.com>
3859 * doc/md.texi (fold_extract_last_@var{m}): Document.
3860 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3861 * optabs.def (fold_extract_last_optab): New optab.
3862 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3863 * internal-fn.c (fold_extract_direct): New macro.
3864 (expand_fold_extract_optab_fn): Likewise.
3865 (direct_fold_extract_optab_supported_p): Likewise.
3866 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3867 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3868 EXTRACT_LAST_REDUCTION.
3869 (get_initial_def_for_reduction): Do not create an initial vector
3870 for EXTRACT_LAST_REDUCTION reductions.
3871 (vectorizable_reduction): Leave the scalar phi in place for
3872 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3873 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3874 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3875 transform phase to vectorizable_condition.
3876 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3877 split out from...
3878 (vect_finish_stmt_generation): ...here.
3879 (vect_finish_replace_stmt): New function.
3880 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3881 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3882 pattern.
3883 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3885 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3886 Alan Hayward <alan.hayward@arm.com>
3887 David Sherwood <david.sherwood@arm.com>
3889 * doc/md.texi (extract_last_@var{m}): Document.
3890 * optabs.def (extract_last_optab): New optab.
3891 * internal-fn.def (EXTRACT_LAST): New internal function.
3892 * internal-fn.c (cond_unary_direct): New macro.
3893 (expand_cond_unary_optab_fn): Likewise.
3894 (direct_cond_unary_optab_supported_p): Likewise.
3895 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3896 loops using EXTRACT_LAST.
3897 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3898 (extract_last_<mode>): ...this optab.
3899 (vec_extract<mode><Vel>): Update accordingly.
3901 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3902 Alan Hayward <alan.hayward@arm.com>
3903 David Sherwood <david.sherwood@arm.com>
3905 * target.def (empty_mask_is_expensive): New hook.
3906 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3907 * doc/tm.texi: Regenerate.
3908 * targhooks.h (default_empty_mask_is_expensive): Declare.
3909 * targhooks.c (default_empty_mask_is_expensive): New function.
3910 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3911 if the target says that empty masks are expensive.
3912 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3913 New function.
3914 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3916 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3917 Alan Hayward <alan.hayward@arm.com>
3918 David Sherwood <david.sherwood@arm.com>
3920 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3921 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3922 (vect_use_loop_mask_for_alignment_p): New function.
3923 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3924 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3925 niters_skip argument. Make sure that the first niters_skip elements
3926 of the first iteration are inactive.
3927 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3928 Update call to vect_set_loop_masks_directly.
3929 (get_misalign_in_elems): New function, split out from...
3930 (vect_gen_prolog_loop_niters): ...here.
3931 (vect_update_init_of_dr): Take a code argument that specifies whether
3932 the adjustment should be added or subtracted.
3933 (vect_update_init_of_drs): Likewise.
3934 (vect_prepare_for_masked_peels): New function.
3935 (vect_do_peeling): Skip prologue peeling if we're using a mask
3936 instead. Update call to vect_update_inits_of_drs.
3937 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3938 mask_skip_niters.
3939 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3940 alignment. Do not include the number of peeled iterations in
3941 the minimum threshold in that case.
3942 (vectorizable_induction): Adjust the start value down by
3943 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3944 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3945 Take the number of skipped iterations into account when calculating
3946 the loop bounds.
3947 * tree-vect-stmts.c (vect_gen_while_not): New function.
3949 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3950 Alan Hayward <alan.hayward@arm.com>
3951 David Sherwood <david.sherwood@arm.com>
3953 * doc/sourcebuild.texi (vect_fully_masked): Document.
3954 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3955 default value to 0.
3956 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3957 split out from...
3958 (vect_analyze_loop_2): ...here. Don't check the vectorization
3959 factor against the number of loop iterations if the loop is
3960 fully-masked.
3962 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3963 Alan Hayward <alan.hayward@arm.com>
3964 David Sherwood <david.sherwood@arm.com>
3966 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3967 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3968 (dump_groups): Update accordingly.
3969 (iv_use::mem_type): New member variable.
3970 (address_p): New function.
3971 (record_use): Add a mem_type argument and initialize the new
3972 mem_type field.
3973 (record_group_use): Add a mem_type argument. Use address_p.
3974 Remove obsolete null checks of base_object. Update call to record_use.
3975 (find_interesting_uses_op): Update call to record_group_use.
3976 (find_interesting_uses_cond): Likewise.
3977 (find_interesting_uses_address): Likewise.
3978 (get_mem_type_for_internal_fn): New function.
3979 (find_address_like_use): Likewise.
3980 (find_interesting_uses_stmt): Try find_address_like_use before
3981 calling find_interesting_uses_op.
3982 (addr_offset_valid_p): Use the iv mem_type field as the type
3983 of the addressed memory.
3984 (add_autoinc_candidates): Likewise.
3985 (get_address_cost): Likewise.
3986 (split_small_address_groups_p): Use address_p.
3987 (split_address_groups): Likewise.
3988 (add_iv_candidate_for_use): Likewise.
3989 (autoinc_possible_for_pair): Likewise.
3990 (rewrite_groups): Likewise.
3991 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3992 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3993 (get_alias_ptr_type_for_ptr_address): New function.
3994 (rewrite_use_address): Rewrite address uses in calls that were
3995 identified by find_address_like_use.
3997 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3998 Alan Hayward <alan.hayward@arm.com>
3999 David Sherwood <david.sherwood@arm.com>
4001 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
4002 TARGET_MEM_REFs.
4003 * gimple-expr.h (is_gimple_addressable: Likewise.
4004 * gimple-expr.c (is_gimple_address): Likewise.
4005 * internal-fn.c (expand_call_mem_ref): New function.
4006 (expand_mask_load_optab_fn): Use it.
4007 (expand_mask_store_optab_fn): Likewise.
4009 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4010 Alan Hayward <alan.hayward@arm.com>
4011 David Sherwood <david.sherwood@arm.com>
4013 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
4014 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
4015 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
4016 (cond_umax@var{mode}): Document.
4017 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
4018 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
4019 (cond_umin_optab, cond_umax_optab): New optabs.
4020 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
4021 (COND_IOR, COND_XOR): New internal functions.
4022 * internal-fn.h (get_conditional_internal_fn): Declare.
4023 * internal-fn.c (cond_binary_direct): New macro.
4024 (expand_cond_binary_optab_fn): Likewise.
4025 (direct_cond_binary_optab_supported_p): Likewise.
4026 (get_conditional_internal_fn): New function.
4027 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
4028 Cope with reduction statements that are vectorized as calls rather
4029 than assignments.
4030 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
4031 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
4032 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
4033 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
4034 (UNSPEC_COND_EOR): New unspecs.
4035 (optab): Add mappings for them.
4036 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
4037 (sve_int_op, sve_fp_op): New int attributes.
4039 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4040 Alan Hayward <alan.hayward@arm.com>
4041 David Sherwood <david.sherwood@arm.com>
4043 * optabs.def (while_ult_optab): New optab.
4044 * doc/md.texi (while_ult@var{m}@var{n}): Document.
4045 * internal-fn.def (WHILE_ULT): New internal function.
4046 * internal-fn.h (direct_internal_fn_supported_p): New override
4047 that takes two types as argument.
4048 * internal-fn.c (while_direct): New macro.
4049 (expand_while_optab_fn): New function.
4050 (convert_optab_supported_p): Likewise.
4051 (direct_while_optab_supported_p): New macro.
4052 * wide-int.h (wi::udiv_ceil): New function.
4053 * tree-vectorizer.h (rgroup_masks): New structure.
4054 (vec_loop_masks): New typedef.
4055 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
4056 and fully_masked_p.
4057 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
4058 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
4059 (vect_max_vf): New function.
4060 (slpeel_make_loop_iterate_ntimes): Delete.
4061 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
4062 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
4063 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
4064 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
4065 internal-fn.h, stor-layout.h and optabs-query.h.
4066 (vect_set_loop_mask): New function.
4067 (add_preheader_seq): Likewise.
4068 (add_header_seq): Likewise.
4069 (interleave_supported_p): Likewise.
4070 (vect_maybe_permute_loop_masks): Likewise.
4071 (vect_set_loop_masks_directly): Likewise.
4072 (vect_set_loop_condition_masked): Likewise.
4073 (vect_set_loop_condition_unmasked): New function, split out from
4074 slpeel_make_loop_iterate_ntimes.
4075 (slpeel_make_loop_iterate_ntimes): Rename to..
4076 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
4077 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
4078 (vect_do_peeling): Update call accordingly.
4079 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
4080 loops.
4081 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4082 mask_compare_type, can_fully_mask_p and fully_masked_p.
4083 (release_vec_loop_masks): New function.
4084 (_loop_vec_info): Use it to free the loop masks.
4085 (can_produce_all_loop_masks_p): New function.
4086 (vect_get_max_nscalars_per_iter): Likewise.
4087 (vect_verify_full_masking): Likewise.
4088 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
4089 retries, and free the mask rgroups before retrying. Check loop-wide
4090 reasons for disallowing fully-masked loops. Make the final decision
4091 about whether use a fully-masked loop or not.
4092 (vect_estimate_min_profitable_iters): Do not assume that peeling
4093 for the number of iterations will be needed for fully-masked loops.
4094 (vectorizable_reduction): Disable fully-masked loops.
4095 (vectorizable_live_operation): Likewise.
4096 (vect_halve_mask_nunits): New function.
4097 (vect_double_mask_nunits): Likewise.
4098 (vect_record_loop_mask): Likewise.
4099 (vect_get_loop_mask): Likewise.
4100 (vect_transform_loop): Handle the case in which the final loop
4101 iteration might handle a partial vector. Call vect_set_loop_condition
4102 instead of slpeel_make_loop_iterate_ntimes.
4103 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
4104 (check_load_store_masking): New function.
4105 (prepare_load_store_mask): Likewise.
4106 (vectorizable_store): Handle fully-masked loops.
4107 (vectorizable_load): Likewise.
4108 (supportable_widening_operation): Use vect_halve_mask_nunits for
4109 booleans.
4110 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
4111 (vect_gen_while): New function.
4112 * config/aarch64/aarch64.md (umax<mode>3): New expander.
4113 (aarch64_uqdec<mode>): New insn.
4115 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4116 Alan Hayward <alan.hayward@arm.com>
4117 David Sherwood <david.sherwood@arm.com>
4119 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
4120 (reduc_xor_scal_optab): New optabs.
4121 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
4122 (reduc_xor_scal_@var{m}): Document.
4123 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
4124 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
4125 internal functions.
4126 * fold-const-call.c (fold_const_call): Handle them.
4127 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
4128 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
4129 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
4130 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
4131 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
4132 (UNSPEC_XORV): New unspecs.
4133 (optab): Add entries for them.
4134 (BITWISEV): New int iterator.
4135 (bit_reduc_op): New int attributes.
4137 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4138 Alan Hayward <alan.hayward@arm.com>
4139 David Sherwood <david.sherwood@arm.com>
4141 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
4142 * internal-fn.def (VEC_SHL_INSERT): New internal function.
4143 * optabs.def (vec_shl_insert_optab): New optab.
4144 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
4145 (duplicate_and_interleave): Likewise.
4146 * tree-vect-loop.c: Include internal-fn.h.
4147 (neutral_op_for_slp_reduction): New function, split out from
4148 get_initial_defs_for_reduction.
4149 (get_initial_def_for_reduction): Handle option 2 for variable-length
4150 vectors by loading the neutral value into a vector and then shifting
4151 the initial value into element 0.
4152 (get_initial_defs_for_reduction): Replace the code argument with
4153 the neutral value calculated by neutral_op_for_slp_reduction.
4154 Use gimple_build_vector for constant-length vectors.
4155 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
4156 but the first group_size elements have a neutral value.
4157 Use duplicate_and_interleave otherwise.
4158 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
4159 Update call to get_initial_defs_for_reduction. Handle SLP
4160 reductions for variable-length vectors by creating one vector
4161 result for each scalar result, with the elements associated
4162 with other scalar results stubbed out with the neutral value.
4163 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
4164 Require IFN_VEC_SHL_INSERT for double reductions on
4165 variable-length vectors, or SLP reductions that have
4166 a neutral value. Require can_duplicate_and_interleave_p
4167 support for variable-length unchained SLP reductions if there
4168 is no neutral value, such as for MIN/MAX reductions. Also require
4169 the number of vector elements to be a multiple of the number of
4170 SLP statements when doing variable-length unchained SLP reductions.
4171 Update call to vect_create_epilog_for_reduction.
4172 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
4173 and remove initial values.
4174 (duplicate_and_interleave): Make public.
4175 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
4176 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
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 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
4183 (can_duplicate_and_interleave_p): New function.
4184 (vect_get_and_check_slp_defs): Take the vector of statements
4185 rather than just the current one. Remove excess parentheses.
4186 Restriction rejectinon of vect_constant_def and vect_external_def
4187 for variable-length vectors to boolean types, or types for which
4188 can_duplicate_and_interleave_p is false.
4189 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
4190 (duplicate_and_interleave): New function.
4191 (vect_get_constant_vectors): Use gimple_build_vector for
4192 constant-length vectors and suitable variable-length constant
4193 vectors. Use duplicate_and_interleave for other variable-length
4194 vectors. Don't defer the update when inserting new statements.
4196 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4197 Alan Hayward <alan.hayward@arm.com>
4198 David Sherwood <david.sherwood@arm.com>
4200 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
4201 min_profitable_iters doesn't go negative.
4203 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4204 Alan Hayward <alan.hayward@arm.com>
4205 David Sherwood <david.sherwood@arm.com>
4207 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
4208 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
4209 * optabs.def (vec_mask_load_lanes_optab): New optab.
4210 (vec_mask_store_lanes_optab): Likewise.
4211 * internal-fn.def (MASK_LOAD_LANES): New internal function.
4212 (MASK_STORE_LANES): Likewise.
4213 * internal-fn.c (mask_load_lanes_direct): New macro.
4214 (mask_store_lanes_direct): Likewise.
4215 (expand_mask_load_optab_fn): Handle masked operations.
4216 (expand_mask_load_lanes_optab_fn): New macro.
4217 (expand_mask_store_optab_fn): Handle masked operations.
4218 (expand_mask_store_lanes_optab_fn): New macro.
4219 (direct_mask_load_lanes_optab_supported_p): Likewise.
4220 (direct_mask_store_lanes_optab_supported_p): Likewise.
4221 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
4222 parameter.
4223 (vect_load_lanes_supported): Likewise.
4224 * tree-vect-data-refs.c (strip_conversion): New function.
4225 (can_group_stmts_p): Likewise.
4226 (vect_analyze_data_ref_accesses): Use it instead of checking
4227 for a pair of assignments.
4228 (vect_store_lanes_supported): Take a masked_p parameter.
4229 (vect_load_lanes_supported): Likewise.
4230 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
4231 vect_store_lanes_supported and vect_load_lanes_supported.
4232 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
4233 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
4234 parameter. Don't allow gaps for masked accesses.
4235 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
4236 and vect_load_lanes_supported.
4237 (get_load_store_type): Take a masked_p parameter and update
4238 call to get_group_load_store_type.
4239 (vectorizable_store): Update call to get_load_store_type.
4240 Handle IFN_MASK_STORE_LANES.
4241 (vectorizable_load): Update call to get_load_store_type.
4242 Handle IFN_MASK_LOAD_LANES.
4244 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4245 Alan Hayward <alan.hayward@arm.com>
4246 David Sherwood <david.sherwood@arm.com>
4248 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
4249 modes for SVE.
4250 * config/aarch64/aarch64-protos.h
4251 (aarch64_sve_struct_memory_operand_p): Declare.
4252 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
4253 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
4254 (VPRED, vpred): Handle SVE structure modes.
4255 * config/aarch64/constraints.md (Utx): New constraint.
4256 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
4257 (aarch64_sve_struct_nonimmediate_operand): New predicates.
4258 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
4259 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
4260 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
4261 structure modes. Split into pieces after RA.
4262 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
4263 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
4264 New patterns.
4265 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
4266 SVE structure modes.
4267 (aarch64_classify_address): Likewise.
4268 (sizetochar): Move earlier in file.
4269 (aarch64_print_operand): Handle SVE register lists.
4270 (aarch64_array_mode): New function.
4271 (aarch64_sve_struct_memory_operand_p): Likewise.
4272 (TARGET_ARRAY_MODE): Redefine.
4274 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4275 Alan Hayward <alan.hayward@arm.com>
4276 David Sherwood <david.sherwood@arm.com>
4278 * target.def (array_mode): New target hook.
4279 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
4280 * doc/tm.texi: Regenerate.
4281 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
4282 * hooks.c (hook_optmode_mode_uhwi_none): New function.
4283 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
4284 targetm.array_mode.
4285 * stor-layout.c (mode_for_array): Likewise. Support polynomial
4286 type sizes.
4288 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4289 Alan Hayward <alan.hayward@arm.com>
4290 David Sherwood <david.sherwood@arm.com>
4292 * fold-const.c (fold_binary_loc): Check the argument types
4293 rather than the result type when testing for a vector operation.
4295 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4297 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
4298 * doc/tm.texi: Regenerate.
4300 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4301 Alan Hayward <alan.hayward@arm.com>
4302 David Sherwood <david.sherwood@arm.com>
4304 * doc/invoke.texi (-msve-vector-bits=): Document new option.
4305 (sve): Document new AArch64 extension.
4306 * doc/md.texi (w): Extend the description of the AArch64
4307 constraint to include SVE vectors.
4308 (Upl, Upa): Document new AArch64 predicate constraints.
4309 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
4310 enum.
4311 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
4312 (msve-vector-bits=): New option.
4313 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
4314 SVE when these are disabled.
4315 (sve): New extension.
4316 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
4317 modes. Adjust their number of units based on aarch64_sve_vg.
4318 (MAX_BITSIZE_MODE_ANY_MODE): Define.
4319 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
4320 aarch64_addr_query_type.
4321 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
4322 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
4323 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
4324 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
4325 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
4326 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
4327 (aarch64_simd_imm_zero_p): Delete.
4328 (aarch64_check_zero_based_sve_index_immediate): Declare.
4329 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4330 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4331 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4332 (aarch64_sve_float_mul_immediate_p): Likewise.
4333 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4334 rather than an rtx.
4335 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
4336 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
4337 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
4338 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
4339 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
4340 (aarch64_regmode_natural_size): Likewise.
4341 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
4342 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
4343 left one place.
4344 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
4345 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
4346 for VG and the SVE predicate registers.
4347 (V_ALIASES): Add a "z"-prefixed alias.
4348 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
4349 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
4350 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
4351 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
4352 (REG_CLASS_NAMES): Add entries for them.
4353 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
4354 and the predicate registers.
4355 (aarch64_sve_vg): Declare.
4356 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
4357 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
4358 (REGMODE_NATURAL_SIZE): Define.
4359 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
4360 SVE macros.
4361 * config/aarch64/aarch64.c: Include cfgrtl.h.
4362 (simd_immediate_info): Add a constructor for series vectors,
4363 and an associated step field.
4364 (aarch64_sve_vg): New variable.
4365 (aarch64_dbx_register_number): Handle VG and the predicate registers.
4366 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
4367 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
4368 (VEC_ANY_DATA, VEC_STRUCT): New constants.
4369 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
4370 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
4371 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
4372 (aarch64_get_mask_mode): New functions.
4373 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
4374 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
4375 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
4376 predicate modes and predicate registers. Explicitly restrict
4377 GPRs to modes of 16 bytes or smaller. Only allow FP registers
4378 to store a vector mode if it is recognized by
4379 aarch64_classify_vector_mode.
4380 (aarch64_regmode_natural_size): New function.
4381 (aarch64_hard_regno_caller_save_mode): Return the original mode
4382 for predicates.
4383 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
4384 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
4385 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
4386 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
4387 functions.
4388 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
4389 does not overlap dest if the function is frame-related. Handle
4390 SVE constants.
4391 (aarch64_split_add_offset): New function.
4392 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
4393 them aarch64_add_offset.
4394 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
4395 and update call to aarch64_sub_sp.
4396 (aarch64_add_cfa_expression): New function.
4397 (aarch64_expand_prologue): Pass extra temporary registers to the
4398 functions above. Handle the case in which we need to emit new
4399 DW_CFA_expressions for registers that were originally saved
4400 relative to the stack pointer, but now have to be expressed
4401 relative to the frame pointer.
4402 (aarch64_output_mi_thunk): Pass extra temporary registers to the
4403 functions above.
4404 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
4405 IP0 and IP1 values for SVE frames.
4406 (aarch64_expand_vec_series): New function.
4407 (aarch64_expand_sve_widened_duplicate): Likewise.
4408 (aarch64_expand_sve_const_vector): Likewise.
4409 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
4410 Handle SVE constants. Use emit_move_insn to move a force_const_mem
4411 into the register, rather than emitting a SET directly.
4412 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
4413 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
4414 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
4415 (offset_9bit_signed_scaled_p): New functions.
4416 (aarch64_replicate_bitmask_imm): New function.
4417 (aarch64_bitmask_imm): Use it.
4418 (aarch64_cannot_force_const_mem): Reject expressions involving
4419 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
4420 (aarch64_classify_index): Handle SVE indices, by requiring
4421 a plain register index with a scale that matches the element size.
4422 (aarch64_classify_address): Handle SVE addresses. Assert that
4423 the mode of the address is VOIDmode or an integer mode.
4424 Update call to aarch64_classify_symbol.
4425 (aarch64_classify_symbolic_expression): Update call to
4426 aarch64_classify_symbol.
4427 (aarch64_const_vec_all_in_range_p): New function.
4428 (aarch64_print_vector_float_operand): Likewise.
4429 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
4430 "vN" for FP registers with SVE modes. Handle (const ...) vectors
4431 and the FP immediates 1.0 and 0.5.
4432 (aarch64_print_address_internal): Handle SVE addresses.
4433 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
4434 (aarch64_regno_regclass): Handle predicate registers.
4435 (aarch64_secondary_reload): Handle big-endian reloads of SVE
4436 data modes.
4437 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
4438 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
4439 (aarch64_convert_sve_vector_bits): New function.
4440 (aarch64_override_options): Use it to handle -msve-vector-bits=.
4441 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4442 rather than an rtx.
4443 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
4444 Handle SVE vector and predicate modes. Accept VL-based constants
4445 that need only one temporary register, and VL offsets that require
4446 no temporary registers.
4447 (aarch64_conditional_register_usage): Mark the predicate registers
4448 as fixed if SVE isn't available.
4449 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
4450 Return true for SVE vector and predicate modes.
4451 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
4452 rather than an unsigned int. Handle SVE modes.
4453 (aarch64_preferred_simd_mode): Update call accordingly. Handle
4454 SVE modes.
4455 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
4456 if SVE is enabled.
4457 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4458 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4459 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4460 (aarch64_sve_float_mul_immediate_p): New functions.
4461 (aarch64_sve_valid_immediate): New function.
4462 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
4463 Explicitly reject structure modes. Check for INDEX constants.
4464 Handle PTRUE and PFALSE constants.
4465 (aarch64_check_zero_based_sve_index_immediate): New function.
4466 (aarch64_simd_imm_zero_p): Delete.
4467 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
4468 vector modes. Accept constants in the range of CNT[BHWD].
4469 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
4470 ask for an Advanced SIMD mode.
4471 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
4472 (aarch64_simd_vector_alignment): Handle SVE predicates.
4473 (aarch64_vectorize_preferred_vector_alignment): New function.
4474 (aarch64_simd_vector_alignment_reachable): Use it instead of
4475 the vector size.
4476 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
4477 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
4478 functions.
4479 (MAX_VECT_LEN): Delete.
4480 (expand_vec_perm_d): Add a vec_flags field.
4481 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
4482 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
4483 (aarch64_evpc_ext): Don't apply a big-endian lane correction
4484 for SVE modes.
4485 (aarch64_evpc_rev): Rename to...
4486 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
4487 (aarch64_evpc_rev_global): New function.
4488 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
4489 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
4490 MAX_VECT_LEN.
4491 (aarch64_evpc_sve_tbl): New function.
4492 (aarch64_expand_vec_perm_const_1): Update after rename of
4493 aarch64_evpc_rev. Handle SVE permutes too, trying
4494 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
4495 than aarch64_evpc_tbl.
4496 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
4497 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
4498 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
4499 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
4500 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
4501 (aarch64_expand_sve_vcond): New functions.
4502 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
4503 of aarch64_vector_mode_p.
4504 (aarch64_dwarf_poly_indeterminate_value): New function.
4505 (aarch64_compute_pressure_classes): Likewise.
4506 (aarch64_can_change_mode_class): Likewise.
4507 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
4508 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
4509 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
4510 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
4511 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
4512 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
4513 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
4514 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
4515 constraints.
4516 (Dn, Dl, Dr): Accept const as well as const_vector.
4517 (Dz): Likewise. Compare against CONST0_RTX.
4518 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
4519 of "vector" where appropriate.
4520 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
4521 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
4522 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
4523 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
4524 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
4525 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
4526 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
4527 (v_int_equiv): Extend to SVE modes.
4528 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
4529 mode attributes.
4530 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
4531 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
4532 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
4533 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
4534 (SVE_COND_FP_CMP): New int iterators.
4535 (perm_hilo): Handle the new unpack unspecs.
4536 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4537 attributes.
4538 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4539 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4540 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4541 (aarch64_equality_operator, aarch64_constant_vector_operand)
4542 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4543 (aarch64_sve_nonimmediate_operand): Likewise.
4544 (aarch64_sve_general_operand): Likewise.
4545 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4546 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4547 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4548 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4549 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4550 (aarch64_sve_float_arith_immediate): Likewise.
4551 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4552 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4553 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4554 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4555 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4556 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4557 (aarch64_sve_float_arith_operand): Likewise.
4558 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4559 (aarch64_sve_float_mul_operand): Likewise.
4560 (aarch64_sve_vec_perm_operand): Likewise.
4561 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4562 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4563 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4564 as well as const_vector.
4565 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4566 in file. Use CONST0_RTX and CONSTM1_RTX.
4567 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4568 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4569 Use aarch64_simd_imm_zero.
4570 * config/aarch64/aarch64-sve.md: New file.
4571 * config/aarch64/aarch64.md: Include it.
4572 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4573 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4574 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4575 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4576 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4577 (sve): New attribute.
4578 (enabled): Disable instructions with the sve attribute unless
4579 TARGET_SVE.
4580 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4581 aarch64_expand_mov_immediate.
4582 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4583 CNT[BHSD] immediates.
4584 (movti): Split CONST_POLY_INT moves into two halves.
4585 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4586 Split additions that need a temporary here if the destination
4587 is the stack pointer.
4588 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4589 (*add<mode>3_poly_1): New instruction.
4590 (set_clobber_cc): New expander.
4592 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4594 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4595 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4596 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4597 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4598 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4599 Change innermode from fixed_mode_size to machine_mode.
4600 (simplify_subreg): Update call accordingly. Handle a constant-sized
4601 subreg of a variable-length CONST_VECTOR.
4603 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4604 Alan Hayward <alan.hayward@arm.com>
4605 David Sherwood <david.sherwood@arm.com>
4607 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4608 (add_offset_to_base): New function, split out from...
4609 (create_mem_ref): ...here. When handling a scale other than 1,
4610 check first whether the address is valid without the offset.
4611 Add it into the base if so, leaving the index and scale as-is.
4613 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4615 PR c++/83778
4616 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4617 fold_for_warn before checking if arg2 is INTEGER_CST.
4619 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4621 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4622 (store_multiple_operation): Delete.
4623 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4624 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4625 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4626 guarded by TARGET_STRING.
4627 (rs6000_output_load_multiple): Delete.
4628 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4629 OPTION_MASK_STRING / TARGET_STRING handling.
4630 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4631 (const rs6000_opt_masks) <"string">: Change mask to 0.
4632 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4633 (MASK_STRING): Delete.
4634 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4635 parts. Simplify.
4636 (load_multiple): Delete.
4637 (*ldmsi8): Delete.
4638 (*ldmsi7): Delete.
4639 (*ldmsi6): Delete.
4640 (*ldmsi5): Delete.
4641 (*ldmsi4): Delete.
4642 (*ldmsi3): Delete.
4643 (store_multiple): Delete.
4644 (*stmsi8): Delete.
4645 (*stmsi7): Delete.
4646 (*stmsi6): Delete.
4647 (*stmsi5): Delete.
4648 (*stmsi4): Delete.
4649 (*stmsi3): Delete.
4650 (movmemsi_8reg): Delete.
4651 (corresponding unnamed define_insn): Delete.
4652 (movmemsi_6reg): Delete.
4653 (corresponding unnamed define_insn): Delete.
4654 (movmemsi_4reg): Delete.
4655 (corresponding unnamed define_insn): Delete.
4656 (movmemsi_2reg): Delete.
4657 (corresponding unnamed define_insn): Delete.
4658 (movmemsi_1reg): Delete.
4659 (corresponding unnamed define_insn): Delete.
4660 * config/rs6000/rs6000.opt (mno-string): New.
4661 (mstring): Replace by deprecation warning stub.
4662 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4664 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4666 * regrename.c (regrename_do_replace): If replacing the same
4667 reg multiple times, try to reuse last created gen_raw_REG.
4669 PR debug/81155
4670 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4671 main to workaround a bug in GDB.
4673 2018-01-12 Tom de Vries <tom@codesourcery.com>
4675 PR target/83737
4676 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4678 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4680 PR rtl-optimization/80481
4681 * ira-color.c (get_cap_member): New function.
4682 (allocnos_conflict_by_live_ranges_p): Use it.
4683 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4684 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4686 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4688 PR target/83628
4689 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4690 (*saddl_se_1): Ditto.
4691 (*ssubsi_1): Ditto.
4692 (*ssubl_se_1): Ditto.
4694 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4696 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4697 rather than wi::to_widest for DR_INITs.
4698 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4699 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4700 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4701 INTEGER_CSTs.
4702 (vect_analyze_group_access_1): Note that here.
4704 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4706 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4707 polynomial type sizes.
4709 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4711 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4712 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4713 (gimple_add_tmp_var): Likewise.
4715 2018-01-12 Martin Liska <mliska@suse.cz>
4717 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4718 (gimple_alloc_sizes): Likewise.
4719 (dump_gimple_statistics): Use PRIu64 in printf format.
4720 * gimple.h: Change uint64_t to int.
4722 2018-01-12 Martin Liska <mliska@suse.cz>
4724 * tree-core.h: Use uint64_t instead of int.
4725 * tree.c (tree_node_counts): Likewise.
4726 (tree_node_sizes): Likewise.
4727 (dump_tree_statistics): Use PRIu64 in printf format.
4729 2018-01-12 Martin Liska <mliska@suse.cz>
4731 * Makefile.in: As qsort_chk is implemented in vec.c, add
4732 vec.o to linkage of gencfn-macros.
4733 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4734 passing the info to record_node_allocation_statistics.
4735 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4736 and pass the info.
4737 * ggc-common.c (struct ggc_usage): Add operator== and use
4738 it in operator< and compare function.
4739 * mem-stats.h (struct mem_usage): Likewise.
4740 * vec.c (struct vec_usage): Remove operator< and compare
4741 function. Can be simply inherited.
4743 2018-01-12 Martin Jambor <mjambor@suse.cz>
4745 PR target/81616
4746 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4747 * tree-ssa-math-opts.c: Include domwalk.h.
4748 (convert_mult_to_fma_1): New function.
4749 (fma_transformation_info): New type.
4750 (fma_deferring_state): Likewise.
4751 (cancel_fma_deferring): New function.
4752 (result_of_phi): Likewise.
4753 (last_fma_candidate_feeds_initial_phi): Likewise.
4754 (convert_mult_to_fma): Added deferring logic, split actual
4755 transformation to convert_mult_to_fma_1.
4756 (math_opts_dom_walker): New type.
4757 (math_opts_dom_walker::after_dom_children): New method, body moved
4758 here from pass_optimize_widening_mul::execute, added deferring logic
4759 bits.
4760 (pass_optimize_widening_mul::execute): Moved most of code to
4761 math_opts_dom_walker::after_dom_children.
4762 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4763 * config/i386/i386.c (ix86_option_override_internal): Added
4764 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4766 2018-01-12 Richard Biener <rguenther@suse.de>
4768 PR debug/83157
4769 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4770 inline instance vars.
4772 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4774 PR target/81819
4775 * config/rx/rx.c (rx_is_restricted_memory_address):
4776 Handle SUBREG case.
4778 2018-01-12 Richard Biener <rguenther@suse.de>
4780 PR tree-optimization/80846
4781 * target.def (split_reduction): New target hook.
4782 * targhooks.c (default_split_reduction): New function.
4783 * targhooks.h (default_split_reduction): Declare.
4784 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4785 target requests first reduce vectors by combining low and high
4786 parts.
4787 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4788 (get_vectype_for_scalar_type_and_size): Export.
4789 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4790 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4791 * doc/tm.texi: Regenerate.
4792 * config/i386/i386.c (ix86_split_reduction): Implement
4793 TARGET_VECTORIZE_SPLIT_REDUCTION.
4795 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4797 PR target/83368
4798 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4799 in PIC mode except for TARGET_VXWORKS_RTP.
4800 * config/sparc/sparc.c: Include cfgrtl.h.
4801 (TARGET_INIT_PIC_REG): Define.
4802 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4803 (sparc_pic_register_p): New predicate.
4804 (sparc_legitimate_address_p): Use it.
4805 (sparc_legitimize_pic_address): Likewise.
4806 (sparc_delegitimize_address): Likewise.
4807 (sparc_mode_dependent_address_p): Likewise.
4808 (gen_load_pcrel_sym): Remove 4th parameter.
4809 (load_got_register): Adjust call to above. Remove obsolete stuff.
4810 (sparc_expand_prologue): Do not call load_got_register here.
4811 (sparc_flat_expand_prologue): Likewise.
4812 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4813 (sparc_use_pseudo_pic_reg): New function.
4814 (sparc_init_pic_reg): Likewise.
4815 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4816 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4818 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4820 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4821 Add item for branch_cost.
4823 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4825 PR rtl-optimization/83565
4826 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4827 not extend the result to a larger mode for rotate operations.
4828 (num_sign_bit_copies1): Likewise.
4830 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4832 PR target/40411
4833 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4834 -symbolic.
4835 Use values-Xc.o for -pedantic.
4836 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4838 2018-01-12 Martin Liska <mliska@suse.cz>
4840 PR ipa/83054
4841 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4842 New function.
4843 (possible_polymorphic_call_targets): Use it.
4844 (ipa_devirt): Likewise.
4846 2018-01-12 Martin Liska <mliska@suse.cz>
4848 * profile-count.h (enum profile_quality): Use 0 as invalid
4849 enum value of profile_quality.
4851 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4853 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4854 -mext-string options.
4856 2018-01-12 Richard Biener <rguenther@suse.de>
4858 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4859 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4860 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4861 Likewise.
4862 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4864 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4866 * configure.ac (--with-long-double-format): Add support for the
4867 configuration option to change the default long double format on
4868 PowerPC systems.
4869 * config.gcc (powerpc*-linux*-*): Likewise.
4870 * configure: Regenerate.
4871 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4872 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4873 used without modification.
4875 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4877 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4878 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4879 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4880 MISC_BUILTIN_SPEC_BARRIER.
4881 (rs6000_init_builtins): Likewise.
4882 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4883 enum value.
4884 (speculation_barrier): New define_insn.
4885 * doc/extend.texi: Document __builtin_speculation_barrier.
4887 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4889 PR target/83203
4890 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4891 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4892 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4893 iterators.
4894 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4895 integral modes instead of "ss" and "sd".
4896 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4897 vectors with 32-bit and 64-bit elements.
4898 (vecdupssescalarmodesuffix): New mode attribute.
4899 (vec_dup<mode>): Use it.
4901 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4903 PR target/83330
4904 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4905 frame if argument is passed on stack.
4907 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4909 PR target/82682
4910 * ree.c (combine_reaching_defs): Optimize also
4911 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4912 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4914 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4916 PR middle-end/83189
4917 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4919 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4921 PR middle-end/83718
4922 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4923 after they are computed.
4925 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4927 PR tree-optimization/83695
4928 * gimple-loop-linterchange.cc
4929 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4930 reset cached scev information after interchange.
4931 (pass_linterchange::execute): Remove call to scev_reset_htab.
4933 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4935 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4936 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4937 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4938 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4939 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4940 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4941 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4942 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4943 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4944 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4945 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4946 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4947 (V_lane_reg): Likewise.
4948 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4949 New define_expand.
4950 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4951 (vfmal_lane_low<mode>_intrinsic,
4952 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4953 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4954 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4955 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4956 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4957 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4959 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4961 * config/arm/arm-cpus.in (fp16fml): New feature.
4962 (ALL_SIMD): Add fp16fml.
4963 (armv8.2-a): Add fp16fml as an option.
4964 (armv8.3-a): Likewise.
4965 (armv8.4-a): Add fp16fml as part of fp16.
4966 * config/arm/arm.h (TARGET_FP16FML): Define.
4967 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4968 when appropriate.
4969 * config/arm/arm-modes.def (V2HF): Define.
4970 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4971 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4972 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4973 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4974 vfmsl_low, vfmsl_high): New set of builtins.
4975 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4976 (vfml_op): New code attribute.
4977 (VFMLHALVES): New int iterator.
4978 (VFML, VFMLSEL): New mode attributes.
4979 (V_reg): Define mapping for V2HF.
4980 (V_hi, V_lo): New mode attributes.
4981 (VF_constraint): Likewise.
4982 (vfml_half, vfml_half_selector): New int attributes.
4983 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4984 define_expand.
4985 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4986 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4987 New define_insn.
4988 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4989 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4990 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4991 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4992 documentation.
4993 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4994 Document new effective target and option set.
4996 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4998 * config/arm/arm-cpus.in (armv8_4): New feature.
4999 (ARMv8_4a): New fgroup.
5000 (armv8.4-a): New arch.
5001 * config/arm/arm-tables.opt: Regenerate.
5002 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
5003 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
5004 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
5005 Add matching rules for -march=armv8.4-a and extensions.
5006 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
5008 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
5010 PR target/81821
5011 * config/rx/rx.md (BW): New mode attribute.
5012 (sync_lock_test_and_setsi): Add mode suffix to insn output.
5014 2018-01-11 Richard Biener <rguenther@suse.de>
5016 PR tree-optimization/83435
5017 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
5018 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
5019 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
5021 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5022 Alan Hayward <alan.hayward@arm.com>
5023 David Sherwood <david.sherwood@arm.com>
5025 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
5026 field.
5027 (aarch64_classify_address): Initialize it. Track polynomial offsets.
5028 (aarch64_print_address_internal): Use it to check for a zero offset.
5030 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5031 Alan Hayward <alan.hayward@arm.com>
5032 David Sherwood <david.sherwood@arm.com>
5034 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
5035 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
5036 Return a poly_int64 rather than a HOST_WIDE_INT.
5037 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
5038 rather than a HOST_WIDE_INT.
5039 * config/aarch64/aarch64.h (aarch64_frame): Protect with
5040 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
5041 hard_fp_offset, frame_size, initial_adjust, callee_offset and
5042 final_offset from HOST_WIDE_INT to poly_int64.
5043 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
5044 to_constant when getting the number of units in an Advanced SIMD
5045 mode.
5046 (aarch64_builtin_vectorized_function): Check for a constant number
5047 of units.
5048 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
5049 GET_MODE_SIZE.
5050 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
5051 attribute instead of GET_MODE_NUNITS.
5052 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
5053 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
5054 GET_MODE_SIZE for fixed-size registers.
5055 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
5056 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
5057 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
5058 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
5059 (aarch64_print_operand, aarch64_print_address_internal)
5060 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
5061 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
5062 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
5063 Handle polynomial GET_MODE_SIZE.
5064 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
5065 wider than SImode without modification.
5066 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
5067 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
5068 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
5069 passing and returning SVE modes.
5070 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
5071 rather than GEN_INT.
5072 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
5073 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
5074 (aarch64_allocate_and_probe_stack_space): Likewise.
5075 (aarch64_layout_frame): Cope with polynomial offsets.
5076 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
5077 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
5078 polynomial offsets.
5079 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
5080 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
5081 poly_int64 rather than a HOST_WIDE_INT.
5082 (aarch64_get_separate_components, aarch64_process_components)
5083 (aarch64_expand_prologue, aarch64_expand_epilogue)
5084 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
5085 (aarch64_anchor_offset): New function, split out from...
5086 (aarch64_legitimize_address): ...here.
5087 (aarch64_builtin_vectorization_cost): Handle polynomial
5088 TYPE_VECTOR_SUBPARTS.
5089 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
5090 GET_MODE_NUNITS.
5091 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
5092 number of elements from the PARALLEL rather than the mode.
5093 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
5094 rather than GET_MODE_BITSIZE.
5095 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
5096 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
5097 (aarch64_expand_vec_perm_const_1): Handle polynomial
5098 d->perm.length () and d->perm elements.
5099 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
5100 Apply to_constant to d->perm elements.
5101 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
5102 polynomial CONST_VECTOR_NUNITS.
5103 (aarch64_move_pointer): Take amount as a poly_int64 rather
5104 than an int.
5105 (aarch64_progress_pointer): Avoid temporary variable.
5106 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
5107 the mode attribute instead of GET_MODE.
5109 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5110 Alan Hayward <alan.hayward@arm.com>
5111 David Sherwood <david.sherwood@arm.com>
5113 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
5114 x exists before using it.
5115 (aarch64_add_constant_internal): Rename to...
5116 (aarch64_add_offset_1): ...this. Replace regnum with separate
5117 src and dest rtxes. Handle the case in which they're different,
5118 including when the offset is zero. Replace scratchreg with an rtx.
5119 Use 2 additions if there is no spare register into which we can
5120 move a 16-bit constant.
5121 (aarch64_add_constant): Delete.
5122 (aarch64_add_offset): Replace reg with separate src and dest
5123 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
5124 Use aarch64_add_offset_1.
5125 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
5126 an rtx rather than an int. Take the delta as a poly_int64
5127 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
5128 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
5129 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
5130 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
5131 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
5132 and aarch64_add_sp.
5133 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
5134 aarch64_add_constant.
5136 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5138 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
5139 Use scalar_float_mode.
5141 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5143 * config/aarch64/aarch64-simd.md
5144 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
5145 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
5146 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
5147 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
5148 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
5149 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
5150 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
5151 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
5152 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
5153 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
5155 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5157 PR target/83514
5158 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
5159 targ_options->x_arm_arch_string is non NULL.
5161 2018-01-11 Tamar Christina <tamar.christina@arm.com>
5163 * config/aarch64/aarch64.h
5164 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
5166 2018-01-11 Sudakshina Das <sudi.das@arm.com>
5168 PR target/82096
5169 * expmed.c (emit_store_flag_force): Swap if const op0
5170 and change VOIDmode to mode of op0.
5172 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5174 PR rtl-optimization/83761
5175 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
5176 than bytes to mode_for_size.
5178 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5180 PR middle-end/83189
5181 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
5182 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
5183 profile.
5185 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5187 PR middle-end/83575
5188 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
5189 when in layout mode.
5190 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
5191 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
5192 partition fixup.
5194 2018-01-10 Michael Collison <michael.collison@arm.com>
5196 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
5197 * config/aarch64/aarch64-option-extension.def: Add
5198 AARCH64_OPT_EXTENSION of 'fp16fml'.
5199 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5200 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
5201 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
5202 * config/aarch64/constraints.md (Ui7): New constraint.
5203 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
5204 (VFMLA_SEL_W): Ditto.
5205 (f16quad): Ditto.
5206 (f16mac1): Ditto.
5207 (VFMLA16_LOW): New int iterator.
5208 (VFMLA16_HIGH): Ditto.
5209 (UNSPEC_FMLAL): New unspec.
5210 (UNSPEC_FMLSL): Ditto.
5211 (UNSPEC_FMLAL2): Ditto.
5212 (UNSPEC_FMLSL2): Ditto.
5213 (f16mac): New code attribute.
5214 * config/aarch64/aarch64-simd-builtins.def
5215 (aarch64_fmlal_lowv2sf): Ditto.
5216 (aarch64_fmlsl_lowv2sf): Ditto.
5217 (aarch64_fmlalq_lowv4sf): Ditto.
5218 (aarch64_fmlslq_lowv4sf): Ditto.
5219 (aarch64_fmlal_highv2sf): Ditto.
5220 (aarch64_fmlsl_highv2sf): Ditto.
5221 (aarch64_fmlalq_highv4sf): Ditto.
5222 (aarch64_fmlslq_highv4sf): Ditto.
5223 (aarch64_fmlal_lane_lowv2sf): Ditto.
5224 (aarch64_fmlsl_lane_lowv2sf): Ditto.
5225 (aarch64_fmlal_laneq_lowv2sf): Ditto.
5226 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
5227 (aarch64_fmlalq_lane_lowv4sf): Ditto.
5228 (aarch64_fmlsl_lane_lowv4sf): Ditto.
5229 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
5230 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
5231 (aarch64_fmlal_lane_highv2sf): Ditto.
5232 (aarch64_fmlsl_lane_highv2sf): Ditto.
5233 (aarch64_fmlal_laneq_highv2sf): Ditto.
5234 (aarch64_fmlsl_laneq_highv2sf): Ditto.
5235 (aarch64_fmlalq_lane_highv4sf): Ditto.
5236 (aarch64_fmlsl_lane_highv4sf): Ditto.
5237 (aarch64_fmlalq_laneq_highv4sf): Ditto.
5238 (aarch64_fmlsl_laneq_highv4sf): Ditto.
5239 * config/aarch64/aarch64-simd.md:
5240 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
5241 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5242 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
5243 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5244 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
5245 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
5246 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
5247 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
5248 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
5249 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
5250 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
5251 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
5252 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
5253 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
5254 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
5255 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
5256 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
5257 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
5258 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
5259 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
5260 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
5261 (vfmlsl_low_u32): Ditto.
5262 (vfmlalq_low_u32): Ditto.
5263 (vfmlslq_low_u32): Ditto.
5264 (vfmlal_high_u32): Ditto.
5265 (vfmlsl_high_u32): Ditto.
5266 (vfmlalq_high_u32): Ditto.
5267 (vfmlslq_high_u32): Ditto.
5268 (vfmlal_lane_low_u32): Ditto.
5269 (vfmlsl_lane_low_u32): Ditto.
5270 (vfmlal_laneq_low_u32): Ditto.
5271 (vfmlsl_laneq_low_u32): Ditto.
5272 (vfmlalq_lane_low_u32): Ditto.
5273 (vfmlslq_lane_low_u32): Ditto.
5274 (vfmlalq_laneq_low_u32): Ditto.
5275 (vfmlslq_laneq_low_u32): Ditto.
5276 (vfmlal_lane_high_u32): Ditto.
5277 (vfmlsl_lane_high_u32): Ditto.
5278 (vfmlal_laneq_high_u32): Ditto.
5279 (vfmlsl_laneq_high_u32): Ditto.
5280 (vfmlalq_lane_high_u32): Ditto.
5281 (vfmlslq_lane_high_u32): Ditto.
5282 (vfmlalq_laneq_high_u32): Ditto.
5283 (vfmlslq_laneq_high_u32): Ditto.
5284 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
5285 (AARCH64_FL_FOR_ARCH8_4): New.
5286 (AARCH64_ISA_F16FML): New ISA flag.
5287 (TARGET_F16FML): New feature flag for fp16fml.
5288 (doc/invoke.texi): Document new fp16fml option.
5290 2018-01-10 Michael Collison <michael.collison@arm.com>
5292 * config/aarch64/aarch64-builtins.c:
5293 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
5294 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5295 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
5296 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
5297 (AARCH64_ISA_SHA3): New ISA flag.
5298 (TARGET_SHA3): New feature flag for sha3.
5299 * config/aarch64/iterators.md (sha512_op): New int attribute.
5300 (CRYPTO_SHA512): New int iterator.
5301 (UNSPEC_SHA512H): New unspec.
5302 (UNSPEC_SHA512H2): Ditto.
5303 (UNSPEC_SHA512SU0): Ditto.
5304 (UNSPEC_SHA512SU1): Ditto.
5305 * config/aarch64/aarch64-simd-builtins.def
5306 (aarch64_crypto_sha512hqv2di): New builtin.
5307 (aarch64_crypto_sha512h2qv2di): Ditto.
5308 (aarch64_crypto_sha512su0qv2di): Ditto.
5309 (aarch64_crypto_sha512su1qv2di): Ditto.
5310 (aarch64_eor3qv8hi): Ditto.
5311 (aarch64_rax1qv2di): Ditto.
5312 (aarch64_xarqv2di): Ditto.
5313 (aarch64_bcaxqv8hi): Ditto.
5314 * config/aarch64/aarch64-simd.md:
5315 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
5316 (aarch64_crypto_sha512su0qv2di): Ditto.
5317 (aarch64_crypto_sha512su1qv2di): Ditto.
5318 (aarch64_eor3qv8hi): Ditto.
5319 (aarch64_rax1qv2di): Ditto.
5320 (aarch64_xarqv2di): Ditto.
5321 (aarch64_bcaxqv8hi): Ditto.
5322 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
5323 (vsha512h2q_u64): Ditto.
5324 (vsha512su0q_u64): Ditto.
5325 (vsha512su1q_u64): Ditto.
5326 (veor3q_u16): Ditto.
5327 (vrax1q_u64): Ditto.
5328 (vxarq_u64): Ditto.
5329 (vbcaxq_u16): Ditto.
5330 * config/arm/types.md (crypto_sha512): New type attribute.
5331 (crypto_sha3): Ditto.
5332 (doc/invoke.texi): Document new sha3 option.
5334 2018-01-10 Michael Collison <michael.collison@arm.com>
5336 * config/aarch64/aarch64-builtins.c:
5337 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
5338 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5339 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
5340 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
5341 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
5342 (AARCH64_ISA_SM4): New ISA flag.
5343 (TARGET_SM4): New feature flag for sm4.
5344 * config/aarch64/aarch64-simd-builtins.def
5345 (aarch64_sm3ss1qv4si): Ditto.
5346 (aarch64_sm3tt1aq4si): Ditto.
5347 (aarch64_sm3tt1bq4si): Ditto.
5348 (aarch64_sm3tt2aq4si): Ditto.
5349 (aarch64_sm3tt2bq4si): Ditto.
5350 (aarch64_sm3partw1qv4si): Ditto.
5351 (aarch64_sm3partw2qv4si): Ditto.
5352 (aarch64_sm4eqv4si): Ditto.
5353 (aarch64_sm4ekeyqv4si): Ditto.
5354 * config/aarch64/aarch64-simd.md:
5355 (aarch64_sm3ss1qv4si): Ditto.
5356 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
5357 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
5358 (aarch64_sm4eqv4si): Ditto.
5359 (aarch64_sm4ekeyqv4si): Ditto.
5360 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
5361 (sm3part_op): Ditto.
5362 (CRYPTO_SM3TT): Ditto.
5363 (CRYPTO_SM3PART): Ditto.
5364 (UNSPEC_SM3SS1): New unspec.
5365 (UNSPEC_SM3TT1A): Ditto.
5366 (UNSPEC_SM3TT1B): Ditto.
5367 (UNSPEC_SM3TT2A): Ditto.
5368 (UNSPEC_SM3TT2B): Ditto.
5369 (UNSPEC_SM3PARTW1): Ditto.
5370 (UNSPEC_SM3PARTW2): Ditto.
5371 (UNSPEC_SM4E): Ditto.
5372 (UNSPEC_SM4EKEY): Ditto.
5373 * config/aarch64/constraints.md (Ui2): New constraint.
5374 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
5375 * config/arm/types.md (crypto_sm3): New type attribute.
5376 (crypto_sm4): Ditto.
5377 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
5378 (vsm3tt1aq_u32): Ditto.
5379 (vsm3tt1bq_u32): Ditto.
5380 (vsm3tt2aq_u32): Ditto.
5381 (vsm3tt2bq_u32): Ditto.
5382 (vsm3partw1q_u32): Ditto.
5383 (vsm3partw2q_u32): Ditto.
5384 (vsm4eq_u32): Ditto.
5385 (vsm4ekeyq_u32): Ditto.
5386 (doc/invoke.texi): Document new sm4 option.
5388 2018-01-10 Michael Collison <michael.collison@arm.com>
5390 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
5391 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
5392 (AARCH64_FL_FOR_ARCH8_4): New.
5393 (AARCH64_FL_V8_4): New flag.
5394 (doc/invoke.texi): Document new armv8.4-a option.
5396 2018-01-10 Michael Collison <michael.collison@arm.com>
5398 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5399 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
5400 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
5401 * config/aarch64/aarch64-option-extension.def: Add
5402 AARCH64_OPT_EXTENSION of 'sha2'.
5403 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
5404 (crypto): Disable sha2 and aes if crypto disabled.
5405 (crypto): Enable aes and sha2 if enabled.
5406 (simd): Disable sha2 and aes if simd disabled.
5407 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
5408 New flags.
5409 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
5410 (TARGET_SHA2): New feature flag for sha2.
5411 (TARGET_AES): New feature flag for aes.
5412 * config/aarch64/aarch64-simd.md:
5413 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
5414 conditional on TARGET_AES.
5415 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
5416 (aarch64_crypto_sha1hsi): Make pattern conditional
5417 on TARGET_SHA2.
5418 (aarch64_crypto_sha1hv4si): Ditto.
5419 (aarch64_be_crypto_sha1hv4si): Ditto.
5420 (aarch64_crypto_sha1su1v4si): Ditto.
5421 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
5422 (aarch64_crypto_sha1su0v4si): Ditto.
5423 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
5424 (aarch64_crypto_sha256su0v4si): Ditto.
5425 (aarch64_crypto_sha256su1v4si): Ditto.
5426 (doc/invoke.texi): Document new aes and sha2 options.
5428 2018-01-10 Martin Sebor <msebor@redhat.com>
5430 PR tree-optimization/83781
5431 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
5432 as string arrays.
5434 2018-01-11 Martin Sebor <msebor@gmail.com>
5435 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5437 PR tree-optimization/83501
5438 PR tree-optimization/81703
5440 * tree-ssa-strlen.c (get_string_cst): Rename...
5441 (get_string_len): ...to this. Handle global constants.
5442 (handle_char_store): Adjust.
5444 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
5445 Jim Wilson <jimw@sifive.com>
5447 * config/riscv/riscv-protos.h (riscv_output_return): New.
5448 * config/riscv/riscv.c (struct machine_function): New naked_p field.
5449 (riscv_attribute_table, riscv_output_return),
5450 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
5451 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
5452 (riscv_compute_frame_info): Only compute frame->mask if not a naked
5453 function.
5454 (riscv_expand_prologue): Add early return for naked function.
5455 (riscv_expand_epilogue): Likewise.
5456 (riscv_function_ok_for_sibcall): Return false for naked function.
5457 (riscv_set_current_function): New.
5458 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
5459 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
5460 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
5461 * doc/extend.texi (RISC-V Function Attributes): New.
5463 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
5465 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
5466 check for 128-bit long double before checking TCmode.
5467 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
5468 128-bit long doubles before checking TFmode or TCmode.
5469 (FLOAT128_IBM_P): Likewise.
5471 2018-01-10 Martin Sebor <msebor@redhat.com>
5473 PR tree-optimization/83671
5474 * builtins.c (c_strlen): Unconditionally return zero for the empty
5475 string.
5476 Use -Warray-bounds for warnings.
5477 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
5478 for non-constant array indices with COMPONENT_REF, arrays of
5479 arrays, and pointers to arrays.
5480 (gimple_fold_builtin_strlen): Determine and set length range for
5481 non-constant character arrays.
5483 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
5485 PR middle-end/81897
5486 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
5487 empty blocks.
5489 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
5491 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
5493 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5495 PR target/83399
5496 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
5497 VECTOR_MEM_ALTIVEC_OR_VSX_P.
5498 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
5499 indexed_or_indirect_operand predicate.
5500 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
5501 (*vsx_le_perm_load_v8hi): Likewise.
5502 (*vsx_le_perm_load_v16qi): Likewise.
5503 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
5504 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
5505 (*vsx_le_perm_store_v8hi): Likewise.
5506 (*vsx_le_perm_store_v16qi): Likewise.
5507 (eight unnamed splitters): Likewise.
5509 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5511 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
5512 * config/rs6000/emmintrin.h: Likewise.
5513 * config/rs6000/mmintrin.h: Likewise.
5514 * config/rs6000/xmmintrin.h: Likewise.
5516 2018-01-10 David Malcolm <dmalcolm@redhat.com>
5518 PR c++/43486
5519 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
5520 "public_flag".
5521 * tree.c (tree_nop_conversion): Return true for location wrapper
5522 nodes.
5523 (maybe_wrap_with_location): New function.
5524 (selftest::check_strip_nops): New function.
5525 (selftest::test_location_wrappers): New function.
5526 (selftest::tree_c_tests): Call it.
5527 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
5528 (maybe_wrap_with_location): New decl.
5529 (EXPR_LOCATION_WRAPPER_P): New macro.
5530 (location_wrapper_p): New inline function.
5531 (tree_strip_any_location_wrapper): New inline function.
5533 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
5535 PR target/83735
5536 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5537 stack_realign_offset for the largest alignment of stack slot
5538 actually used.
5539 (ix86_find_max_used_stack_alignment): New function.
5540 (ix86_finalize_stack_frame_flags): Use it. Set
5541 max_used_stack_alignment if we don't realign stack.
5542 * config/i386/i386.h (machine_function): Add
5543 max_used_stack_alignment.
5545 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5547 * config/arm/arm.opt (-mbranch-cost): New option.
5548 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5549 account.
5551 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5553 PR target/83629
5554 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5555 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5557 2018-01-10 Richard Biener <rguenther@suse.de>
5559 PR debug/83765
5560 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5561 early out so it also covers the case where we have a non-NULL
5562 origin.
5564 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5566 PR tree-optimization/83753
5567 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5568 for non-strided grouped accesses if the number of elements is 1.
5570 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5572 PR target/81616
5573 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5574 * i386.h (TARGET_USE_GATHER): Define.
5575 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5577 2018-01-10 Martin Liska <mliska@suse.cz>
5579 PR bootstrap/82831
5580 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5581 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5582 partitioning.
5583 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5584 CLEANUP_NO_PARTITIONING is not set.
5586 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5588 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5589 for vectors, as a partial revert of r254296.
5590 * rtl.h (const_vec_p): Delete.
5591 (const_vec_duplicate_p): Don't test for vector CONSTs.
5592 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5593 * expmed.c (make_tree): Likewise.
5595 Revert:
5596 * common.md (E, F): Use CONSTANT_P instead of checking for
5597 CONST_VECTOR.
5598 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5599 checking for CONST_VECTOR.
5601 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5603 PR middle-end/83575
5604 * predict.c (force_edge_cold): Handle in more sane way edges
5605 with no prediction.
5607 2018-01-09 Carl Love <cel@us.ibm.com>
5609 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5610 V4SI, V4SF types.
5611 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5612 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5613 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5614 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5615 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5616 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5617 * config/rs6000/rs6000-protos.h: Add extern defition for
5618 rs6000_generate_float2_double_code.
5619 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5620 function.
5621 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5622 (float2_v2df): Add define_expand.
5624 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5626 PR target/83628
5627 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5628 op_mode in the force_to_mode call.
5630 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5632 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5633 instead of checking each element individually.
5634 (aarch64_evpc_uzp): Likewise.
5635 (aarch64_evpc_zip): Likewise.
5636 (aarch64_evpc_ext): Likewise.
5637 (aarch64_evpc_rev): Likewise.
5638 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5639 instead of checking each element individually. Return true without
5640 generating rtl if
5641 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5642 whether all selected elements come from the same input, instead of
5643 checking each element individually. Remove calls to gen_rtx_REG,
5644 start_sequence and end_sequence and instead assert that no rtl is
5645 generated.
5647 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5649 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5650 order of HIGH and CONST checks.
5652 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5654 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5655 if the destination isn't an SSA_NAME.
5657 2018-01-09 Richard Biener <rguenther@suse.de>
5659 PR tree-optimization/83668
5660 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5661 move prologue...
5662 (canonicalize_loop_form): ... here, renamed from ...
5663 (canonicalize_loop_closed_ssa_form): ... this and amended to
5664 swap successor edges for loop exit blocks to make us use
5665 the RPO order we need for initial schedule generation.
5667 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5669 PR tree-optimization/64811
5670 * match.pd: When optimizing comparisons with Inf, avoid
5671 introducing or losing exceptions from comparisons with NaN.
5673 2018-01-09 Martin Liska <mliska@suse.cz>
5675 PR sanitizer/82517
5676 * asan.c (shadow_mem_size): Add gcc_assert.
5678 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5680 Don't save registers in main().
5682 PR target/83738
5683 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5684 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5685 * config/avr/avr.c (avr_set_current_function): Don't error if
5686 naked, OS_task or OS_main are specified at the same time.
5687 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5688 OS_main.
5689 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5690 attribute.
5691 * common/config/avr/avr-common.c (avr_option_optimization_table):
5692 Switch on -mmain-is-OS_task for optimizing compilations.
5694 2018-01-09 Richard Biener <rguenther@suse.de>
5696 PR tree-optimization/83572
5697 * graphite.c: Include cfganal.h.
5698 (graphite_transform_loops): Connect infinite loops to exit
5699 and remove fake edges at the end.
5701 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5703 * ipa-inline.c (edge_badness): Revert accidental checkin.
5705 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5707 PR ipa/80763
5708 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5709 symbols; not inline clones.
5711 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5713 PR target/83507
5714 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5715 hard registers. Formatting fixes.
5717 PR preprocessor/83722
5718 * gcc.c (try_generate_repro): Pass
5719 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5720 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5721 do_report_bug.
5723 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5724 Kito Cheng <kito.cheng@gmail.com>
5726 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5727 (riscv_leaf_function_p): Delete.
5728 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5730 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5732 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5733 function.
5734 (do_ifelse): New function.
5735 (do_isel): New function.
5736 (do_sub3): New function.
5737 (do_add3): New function.
5738 (do_load_mask_compare): New function.
5739 (do_overlap_load_compare): New function.
5740 (expand_compare_loop): New function.
5741 (expand_block_compare): Call expand_compare_loop() when appropriate.
5742 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5743 option description.
5744 (-mblock-compare-inline-loop-limit): New option.
5746 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5748 PR target/83677
5749 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5750 Reverse order of second and third operands in first alternative.
5751 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5752 of first and second elements in UNSPEC_VPERMR vector.
5753 (altivec_expand_vec_perm_le): Likewise.
5755 2018-01-08 Jeff Law <law@redhat.com>
5757 PR rtl-optimizatin/81308
5758 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5759 (process_switch): If group_case_labels makes a change, then set
5760 cfg_altered.
5761 (pass_convert_switch::execute): If a switch is converted, then
5762 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5764 PR rtl-optimization/81308
5765 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5766 splitting insns.
5768 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5770 PR target/83663 - Revert r255946
5771 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5772 generation for cases where splatting a value is not useful.
5773 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5774 across a vec_duplicate and a paradoxical subreg forming a vector
5775 mode to a vec_concat.
5777 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5779 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5780 -march=armv8.3-a variants.
5781 * config/arm/t-multilib: Likewise.
5782 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5784 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5786 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5787 to generate rtl.
5788 (cceq_ior_compare_complement): Give it a name so I can use it, and
5789 change boolean_or_operator predicate to boolean_operator so it can
5790 be used to generate a crand.
5791 (eqne): New code iterator.
5792 (bd/bd_neg): New code_attrs.
5793 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5794 a single define_insn.
5795 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5796 decrement (bdnzt/bdnzf/bdzt/bdzf).
5797 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5798 with the new names of the branch decrement patterns, and added the
5799 names of the branch decrement conditional patterns.
5801 2018-01-08 Richard Biener <rguenther@suse.de>
5803 PR tree-optimization/83563
5804 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5805 cache.
5807 2018-01-08 Richard Biener <rguenther@suse.de>
5809 PR middle-end/83713
5810 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5812 2018-01-08 Richard Biener <rguenther@suse.de>
5814 PR tree-optimization/83685
5815 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5816 references to abnormals.
5818 2018-01-08 Richard Biener <rguenther@suse.de>
5820 PR lto/83719
5821 * dwarf2out.c (output_indirect_strings): Handle empty
5822 skeleton_debug_str_hash.
5823 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5825 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5827 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5828 (emit_store_direct): Likewise.
5829 (arc_trampoline_adjust_address): Likewise.
5830 (arc_asm_trampoline_template): New function.
5831 (arc_initialize_trampoline): Use asm_trampoline_template.
5832 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5833 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5834 * config/arc/arc.md (flush_icache): Delete pattern.
5836 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5838 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5839 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5840 munaligned-access.
5842 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5844 PR target/83681
5845 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5846 by not USED_FOR_TARGET.
5847 (make_pass_resolve_sw_modes): Likewise.
5849 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5851 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5852 USED_FOR_TARGET.
5854 2018-01-08 Richard Biener <rguenther@suse.de>
5856 PR middle-end/83580
5857 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5859 2018-01-08 Richard Biener <rguenther@suse.de>
5861 PR middle-end/83517
5862 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5864 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5866 PR middle-end/81897
5867 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5868 basic blocks with a small number of successors.
5869 (convert_control_dep_chain_into_preds): Improve handling of
5870 forwarder blocks.
5871 (dump_predicates): Split apart into...
5872 (dump_pred_chain): ...here...
5873 (dump_pred_info): ...and here.
5874 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5875 (can_chain_union_be_invalidated_p): Improve check for invalidation
5876 of paths.
5877 (uninit_uses_cannot_happen): Avoid unnecessary if
5878 convert_control_dep_chain_into_preds yielded nothing.
5880 2018-01-06 Martin Sebor <msebor@redhat.com>
5882 PR tree-optimization/83640
5883 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5884 subtracting negative offset from size.
5885 (builtin_access::overlap): Adjust offset bounds of the access to fall
5886 within the size of the object if possible.
5888 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5890 PR rtl-optimization/83699
5891 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5892 extract_bit_field_as_subreg to cases in which the extracted
5893 value is also a vector.
5895 * lra-constraints.c (process_alt_operands): Test for the equivalence
5896 substitutions when detecting a possible reload cycle.
5898 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5900 PR debug/83480
5901 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5902 by default if flag_selective_schedling{,2}. Formatting fixes.
5904 PR rtl-optimization/83682
5905 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5906 if it has non-VECTOR_MODE element mode.
5907 (vec_duplicate_p): Likewise.
5909 PR middle-end/83694
5910 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5911 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5913 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5915 PR target/83604
5916 * config/i386/i386-builtin.def
5917 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5918 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5919 Require also OPTION_MASK_ISA_AVX512F in addition to
5920 OPTION_MASK_ISA_GFNI.
5921 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5922 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5923 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5924 to OPTION_MASK_ISA_GFNI.
5925 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5926 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5927 OPTION_MASK_ISA_AVX512BW.
5928 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5929 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5930 addition to OPTION_MASK_ISA_GFNI.
5931 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5932 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5933 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5934 to OPTION_MASK_ISA_GFNI.
5935 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5936 a requirement for all ISAs rather than any of them with a few
5937 exceptions.
5938 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5939 processing.
5940 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5941 bitmasks to be enabled with 3 exceptions, instead of requiring any
5942 enabled ISA with lots of exceptions.
5943 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5944 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5945 Change avx512bw in isa attribute to avx512f.
5946 * config/i386/sgxintrin.h: Add license boilerplate.
5947 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5948 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5949 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5950 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5951 defined.
5952 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5953 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5954 temporarily sse2 rather than sse if not enabled already.
5956 PR target/83604
5957 * config/i386/sse.md (VI248_VLBW): Rename to ...
5958 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5959 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5960 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5961 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5962 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5963 mode iterator instead of VI248_VLBW.
5965 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5967 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5968 (record_modified): Skip clobbers; add debug output.
5969 (param_change_prob): Use sreal frequencies.
5971 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5973 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5974 punt for user-aligned variables.
5976 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5978 * tree-chrec.c (chrec_contains_symbols): Return true for
5979 POLY_INT_CST.
5981 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5983 PR target/82439
5984 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5985 of (x|y) == x for BICS pattern.
5987 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5989 PR tree-optimization/83605
5990 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5991 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5992 can throw.
5994 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5996 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5997 * config/epiphany/rtems.h: New file.
5999 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6000 Uros Bizjak <ubizjak@gmail.com>
6002 PR target/83554
6003 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
6004 QIreg_operand instead of register_operand predicate.
6005 * config/i386/i386.c (ix86_rop_should_change_byte_p,
6006 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
6007 comments instead of -fmitigate[-_]rop.
6009 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
6011 PR bootstrap/81926
6012 * cgraphunit.c (symbol_table::compile): Switch to text_section
6013 before calling assembly_start debug hook.
6014 * run-rtl-passes.c (run_rtl_passes): Likewise.
6015 Include output.h.
6017 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6019 * tree-vrp.c (extract_range_from_binary_expr_1): Check
6020 range_int_cst_p rather than !symbolic_range_p before calling
6021 extract_range_from_multiplicative_op_1.
6023 2018-01-04 Jeff Law <law@redhat.com>
6025 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
6026 redundant test in assertion.
6028 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6030 * doc/rtl.texi: Document machine_mode wrapper classes.
6032 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6034 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
6035 using tree_to_uhwi.
6037 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6039 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
6040 the VEC_PERM_EXPR fold to fail.
6042 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6044 PR debug/83585
6045 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
6046 to switched_sections.
6048 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6050 PR target/83680
6051 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
6052 test for d.testing.
6054 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
6056 PR target/83387
6057 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
6058 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
6060 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6062 PR debug/83666
6063 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
6064 is BLKmode and bitpos not zero or mode change is needed.
6066 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6068 PR target/83675
6069 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
6070 TARGET_VIS2.
6072 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
6074 PR target/83628
6075 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
6076 instead of MULT rtx. Update all corresponding splitters.
6077 (*saddl_se): Ditto.
6078 (*ssub<modesuffix>): Ditto.
6079 (*ssubl_se): Ditto.
6080 (*cmp_sadd_di): Update split patterns.
6081 (*cmp_sadd_si): Ditto.
6082 (*cmp_sadd_sidi): Ditto.
6083 (*cmp_ssub_di): Ditto.
6084 (*cmp_ssub_si): Ditto.
6085 (*cmp_ssub_sidi): Ditto.
6086 * config/alpha/predicates.md (const23_operand): New predicate.
6087 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
6088 Look for ASHIFT, not MULT inner operand.
6089 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
6091 2018-01-04 Martin Liska <mliska@suse.cz>
6093 PR gcov-profile/83669
6094 * gcov.c (output_intermediate_file): Add version to intermediate
6095 gcov file.
6096 * doc/gcov.texi: Document new field 'version' in intermediate
6097 file format. Fix location of '-k' option of gcov command.
6099 2018-01-04 Martin Liska <mliska@suse.cz>
6101 PR ipa/82352
6102 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
6104 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6106 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
6108 2018-01-03 Martin Sebor <msebor@redhat.com>
6110 PR tree-optimization/83655
6111 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
6112 checking calls with invalid arguments.
6114 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6116 * tree-vect-stmts.c (vect_get_store_rhs): New function.
6117 (vectorizable_mask_load_store): Delete.
6118 (vectorizable_call): Return false for masked loads and stores.
6119 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
6120 instead of gimple_assign_rhs1.
6121 (vectorizable_load): Handle IFN_MASK_LOAD.
6122 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
6124 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6126 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
6127 split out from..,
6128 (vectorizable_mask_load_store): ...here.
6129 (vectorizable_load): ...and here.
6131 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6133 * tree-vect-stmts.c (vect_build_all_ones_mask)
6134 (vect_build_zero_merge_argument): New functions, split out from...
6135 (vectorizable_load): ...here.
6137 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6139 * tree-vect-stmts.c (vect_check_store_rhs): New function,
6140 split out from...
6141 (vectorizable_mask_load_store): ...here.
6142 (vectorizable_store): ...and here.
6144 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6146 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
6147 split out from...
6148 (vectorizable_mask_load_store): ...here.
6150 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6152 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
6153 (vect_model_store_cost): Take a vec_load_store_type instead of a
6154 vect_def_type.
6155 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
6156 (vect_model_store_cost): Take a vec_load_store_type instead of a
6157 vect_def_type.
6158 (vectorizable_mask_load_store): Update accordingly.
6159 (vectorizable_store): Likewise.
6160 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
6162 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6164 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
6165 IFN_MASK_LOAD calls here rather than...
6166 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
6168 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6169 Alan Hayward <alan.hayward@arm.com>
6170 David Sherwood <david.sherwood@arm.com>
6172 * expmed.c (extract_bit_field_1): For vector extracts,
6173 fall back to extract_bit_field_as_subreg if vec_extract
6174 isn't available.
6176 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6177 Alan Hayward <alan.hayward@arm.com>
6178 David Sherwood <david.sherwood@arm.com>
6180 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
6181 they are variable or constant sized.
6182 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
6183 slots for constant-sized data.
6185 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6186 Alan Hayward <alan.hayward@arm.com>
6187 David Sherwood <david.sherwood@arm.com>
6189 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
6190 handling COND_EXPRs with boolean comparisons, try to find a better
6191 basis for the mask type than the boolean itself.
6193 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6195 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
6196 is calculated and how it can be overridden.
6197 * genmodes.c (max_bitsize_mode_any_mode): New variable.
6198 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
6199 if defined.
6200 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
6201 if nonzero.
6203 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6204 Alan Hayward <alan.hayward@arm.com>
6205 David Sherwood <david.sherwood@arm.com>
6207 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
6208 Remove the mode argument.
6209 (aarch64_simd_valid_immediate): Remove the mode and inverse
6210 arguments.
6211 * config/aarch64/iterators.md (bitsize): New iterator.
6212 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
6213 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
6214 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
6215 aarch64_simd_valid_immediate.
6216 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
6217 (aarch64_reg_or_bic_imm): Likewise.
6218 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
6219 with an insn_type enum and msl with a modifier_type enum.
6220 Replace element_width with a scalar_mode. Change the shift
6221 to unsigned int. Add constructors for scalar_float_mode and
6222 scalar_int_mode elements.
6223 (aarch64_vect_float_const_representable_p): Delete.
6224 (aarch64_can_const_movi_rtx_p)
6225 (aarch64_simd_scalar_immediate_valid_for_move)
6226 (aarch64_simd_make_constant): Update call to
6227 aarch64_simd_valid_immediate.
6228 (aarch64_advsimd_valid_immediate_hs): New function.
6229 (aarch64_advsimd_valid_immediate): Likewise.
6230 (aarch64_simd_valid_immediate): Remove mode and inverse
6231 arguments. Rewrite to use the above. Use const_vec_duplicate_p
6232 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
6233 and aarch64_float_const_representable_p on the result.
6234 (aarch64_output_simd_mov_immediate): Remove mode argument.
6235 Update call to aarch64_simd_valid_immediate and use of
6236 simd_immediate_info.
6237 (aarch64_output_scalar_simd_mov_immediate): Update call
6238 accordingly.
6240 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6241 Alan Hayward <alan.hayward@arm.com>
6242 David Sherwood <david.sherwood@arm.com>
6244 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
6245 (mode_nunits): Likewise CONST_MODE_NUNITS.
6246 * machmode.def (ADJUST_NUNITS): Document.
6247 * genmodes.c (mode_data::need_nunits_adj): New field.
6248 (blank_mode): Update accordingly.
6249 (adj_nunits): New variable.
6250 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
6251 parameter.
6252 (emit_mode_size_inline): Set need_bytesize_adj for all modes
6253 listed in adj_nunits.
6254 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
6255 listed in adj_nunits. Don't emit case statements for such modes.
6256 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
6257 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
6258 nothing if adj_nunits is nonnull.
6259 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
6260 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
6261 (emit_mode_fbit): Update use of print_maybe_const_decl.
6262 (emit_move_size): Likewise. Treat the array as non-const
6263 if adj_nunits.
6264 (emit_mode_adjustments): Handle adj_nunits.
6266 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6268 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
6269 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
6270 (VECTOR_MODES): Use it.
6271 (make_vector_modes): Take the prefix as an argument.
6273 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6274 Alan Hayward <alan.hayward@arm.com>
6275 David Sherwood <david.sherwood@arm.com>
6277 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
6278 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
6279 for MODE_VECTOR_BOOL.
6280 * machmode.def (VECTOR_BOOL_MODE): Document.
6281 * genmodes.c (VECTOR_BOOL_MODE): New macro.
6282 (make_vector_bool_mode): New function.
6283 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
6284 MODE_VECTOR_BOOL.
6285 * lto-streamer-in.c (lto_input_mode_table): Likewise.
6286 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
6287 Likewise.
6288 * stor-layout.c (int_mode_for_mode): Likewise.
6289 * tree.c (build_vector_type_for_mode): Likewise.
6290 * varasm.c (output_constant_pool_2): Likewise.
6291 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
6292 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
6293 for MODE_VECTOR_BOOL.
6294 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
6295 of mode class checks.
6296 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
6297 instead of a list of mode class checks.
6298 (expand_vector_scalar_condition): Likewise.
6299 (type_for_widest_vector_mode): Handle BImode as an inner mode.
6301 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6302 Alan Hayward <alan.hayward@arm.com>
6303 David Sherwood <david.sherwood@arm.com>
6305 * machmode.h (mode_size): Change from unsigned short to
6306 poly_uint16_pod.
6307 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
6308 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6309 or if measurement_type is not polynomial.
6310 (fixed_size_mode::includes_p): Check for constant-sized modes.
6311 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
6312 return a poly_uint16 rather than an unsigned short.
6313 (emit_mode_size): Change the type of mode_size from unsigned short
6314 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
6315 (emit_mode_adjustments): Cope with polynomial vector sizes.
6316 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6317 for GET_MODE_SIZE.
6318 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6319 for GET_MODE_SIZE.
6320 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
6321 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
6322 * caller-save.c (setup_save_areas): Likewise.
6323 (replace_reg_with_saved_mem): Likewise.
6324 * calls.c (emit_library_call_value_1): Likewise.
6325 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
6326 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
6327 (gen_lowpart_for_combine): Likewise.
6328 * convert.c (convert_to_integer_1): Likewise.
6329 * cse.c (equiv_constant, cse_insn): Likewise.
6330 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
6331 (cselib_subst_to_values): Likewise.
6332 * dce.c (word_dce_process_block): Likewise.
6333 * df-problems.c (df_word_lr_mark_ref): Likewise.
6334 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
6335 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
6336 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
6337 (rtl_for_decl_location): Likewise.
6338 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
6339 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
6340 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
6341 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
6342 (expand_expr_real_1): Likewise.
6343 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
6344 (pad_below): Likewise.
6345 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6346 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
6347 * ira.c (get_subreg_tracking_sizes): Likewise.
6348 * ira-build.c (ira_create_allocno_objects): Likewise.
6349 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
6350 (ira_sort_regnos_for_alter_reg): Likewise.
6351 * ira-costs.c (record_operand_costs): Likewise.
6352 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
6353 (resolve_simple_move): Likewise.
6354 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
6355 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
6356 (lra_constraints): Likewise.
6357 (CONST_POOL_OK_P): Reject variable-sized modes.
6358 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
6359 (add_pseudo_to_slot, lra_spill): Likewise.
6360 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6361 * optabs-query.c (get_best_extraction_insn): Likewise.
6362 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6363 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
6364 (expand_mult_highpart, valid_multiword_target_p): Likewise.
6365 * recog.c (offsettable_address_addr_space_p): Likewise.
6366 * regcprop.c (maybe_mode_change): Likewise.
6367 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
6368 * regrename.c (build_def_use): Likewise.
6369 * regstat.c (dump_reg_info): Likewise.
6370 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
6371 (find_reloads, find_reloads_subreg_address): Likewise.
6372 * reload1.c (eliminate_regs_1): Likewise.
6373 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
6374 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
6375 (simplify_binary_operation_1, simplify_subreg): Likewise.
6376 * targhooks.c (default_function_arg_padding): Likewise.
6377 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
6378 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
6379 (verify_gimple_assign_ternary): Likewise.
6380 * tree-inline.c (estimate_move_cost): Likewise.
6381 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6382 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
6383 (get_address_cost_ainc): Likewise.
6384 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
6385 (vect_supportable_dr_alignment): Likewise.
6386 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6387 (vectorizable_reduction): Likewise.
6388 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
6389 (vectorizable_operation, vectorizable_load): Likewise.
6390 * tree.c (build_same_sized_truth_vector_type): Likewise.
6391 * valtrack.c (cleanup_auto_inc_dec): Likewise.
6392 * var-tracking.c (emit_note_insn_var_location): Likewise.
6393 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
6394 (ADDR_VEC_ALIGN): Likewise.
6396 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6397 Alan Hayward <alan.hayward@arm.com>
6398 David Sherwood <david.sherwood@arm.com>
6400 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
6401 unsigned short.
6402 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6403 or if measurement_type is polynomial.
6404 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
6405 * combine.c (make_extraction): Likewise.
6406 * dse.c (find_shift_sequence): Likewise.
6407 * dwarf2out.c (mem_loc_descriptor): Likewise.
6408 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
6409 (extract_bit_field, extract_low_bits): Likewise.
6410 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
6411 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
6412 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
6413 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
6414 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6415 * reload.c (find_reloads): Likewise.
6416 * reload1.c (alter_reg): Likewise.
6417 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
6418 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
6419 * tree-if-conv.c (predicate_mem_writes): Likewise.
6420 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
6421 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
6422 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
6423 * valtrack.c (dead_debug_insert_temp): Likewise.
6424 * varasm.c (mergeable_constant_section): Likewise.
6425 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
6427 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6428 Alan Hayward <alan.hayward@arm.com>
6429 David Sherwood <david.sherwood@arm.com>
6431 * expr.c (expand_assignment): Cope with polynomial mode sizes
6432 when assigning to a CONCAT.
6434 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6435 Alan Hayward <alan.hayward@arm.com>
6436 David Sherwood <david.sherwood@arm.com>
6438 * machmode.h (mode_precision): Change from unsigned short to
6439 poly_uint16_pod.
6440 (mode_to_precision): Return a poly_uint16 rather than an unsigned
6441 short.
6442 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
6443 or if measurement_type is not polynomial.
6444 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
6445 in which the mode is already known to be a scalar_int_mode.
6446 * genmodes.c (emit_mode_precision): Change the type of mode_precision
6447 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
6448 initializer.
6449 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6450 for GET_MODE_PRECISION.
6451 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6452 for GET_MODE_PRECISION.
6453 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
6454 as polynomial.
6455 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
6456 (expand_field_assignment, make_extraction): Likewise.
6457 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
6458 (get_last_value): Likewise.
6459 * convert.c (convert_to_integer_1): Likewise.
6460 * cse.c (cse_insn): Likewise.
6461 * expr.c (expand_expr_real_1): Likewise.
6462 * lra-constraints.c (simplify_operand_subreg): Likewise.
6463 * optabs-query.c (can_atomic_load_p): Likewise.
6464 * optabs.c (expand_atomic_load): Likewise.
6465 (expand_atomic_store): Likewise.
6466 * ree.c (combine_reaching_defs): Likewise.
6467 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
6468 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
6469 * tree.h (type_has_mode_precision_p): Likewise.
6470 * ubsan.c (instrument_si_overflow): Likewise.
6472 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6473 Alan Hayward <alan.hayward@arm.com>
6474 David Sherwood <david.sherwood@arm.com>
6476 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
6477 polynomial numbers of units.
6478 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
6479 (valid_vector_subparts_p): New function.
6480 (build_vector_type): Remove temporary shim and take the number
6481 of units as a poly_uint64 rather than an int.
6482 (build_opaque_vector_type): Take the number of units as a
6483 poly_uint64 rather than an int.
6484 * tree.c (build_vector_from_ctor): Handle polynomial
6485 TYPE_VECTOR_SUBPARTS.
6486 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
6487 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
6488 (build_vector_from_val): If the number of units is variable,
6489 use build_vec_duplicate_cst for constant operands and
6490 VEC_DUPLICATE_EXPR otherwise.
6491 (make_vector_type): Remove temporary is_constant ().
6492 (build_vector_type, build_opaque_vector_type): Take the number of
6493 units as a poly_uint64 rather than an int.
6494 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
6495 VECTOR_CST_NELTS.
6496 * cfgexpand.c (expand_debug_expr): Likewise.
6497 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
6498 (store_constructor, expand_expr_real_1): Likewise.
6499 (const_scalar_mask_from_tree): Likewise.
6500 * fold-const-call.c (fold_const_reduction): Likewise.
6501 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
6502 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
6503 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
6504 (fold_relational_const): Likewise.
6505 (native_interpret_vector): Likewise. Change the size from an
6506 int to an unsigned int.
6507 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
6508 TYPE_VECTOR_SUBPARTS.
6509 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
6510 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
6511 duplicating a non-constant operand into a variable-length vector.
6512 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
6513 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
6514 * ipa-icf.c (sem_variable::equals): Likewise.
6515 * match.pd: Likewise.
6516 * omp-simd-clone.c (simd_clone_subparts): Likewise.
6517 * print-tree.c (print_node): Likewise.
6518 * stor-layout.c (layout_type): Likewise.
6519 * targhooks.c (default_builtin_vectorization_cost): Likewise.
6520 * tree-cfg.c (verify_gimple_comparison): Likewise.
6521 (verify_gimple_assign_binary): Likewise.
6522 (verify_gimple_assign_ternary): Likewise.
6523 (verify_gimple_assign_single): Likewise.
6524 * tree-pretty-print.c (dump_generic_node): Likewise.
6525 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6526 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
6527 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
6528 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
6529 (vect_shift_permute_load_chain): Likewise.
6530 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
6531 (expand_vector_condition, optimize_vector_constructor): Likewise.
6532 (lower_vec_perm, get_compute_type): Likewise.
6533 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6534 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
6535 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
6536 (vect_recog_mask_conversion_pattern): Likewise.
6537 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6538 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6539 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6540 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6541 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6542 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6543 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6544 (supportable_widening_operation): Likewise.
6545 (supportable_narrowing_operation): Likewise.
6546 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6547 Likewise.
6548 * varasm.c (output_constant): Likewise.
6550 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6551 Alan Hayward <alan.hayward@arm.com>
6552 David Sherwood <david.sherwood@arm.com>
6554 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6555 so that both the length == 3 and length != 3 cases set up their
6556 own permute vectors. Add comments explaining why we know the
6557 number of elements is constant.
6558 (vect_permute_load_chain): Likewise.
6560 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6561 Alan Hayward <alan.hayward@arm.com>
6562 David Sherwood <david.sherwood@arm.com>
6564 * machmode.h (mode_nunits): Change from unsigned char to
6565 poly_uint16_pod.
6566 (ONLY_FIXED_SIZE_MODES): New macro.
6567 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6568 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6569 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6570 New typedefs.
6571 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6572 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6573 or if measurement_type is not polynomial.
6574 * genmodes.c (ZERO_COEFFS): New macro.
6575 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6576 poly_uint16.
6577 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6578 Use ZERO_COEFFS when emitting initializers.
6579 * data-streamer.h (bp_pack_poly_value): New function.
6580 (bp_unpack_poly_value): Likewise.
6581 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6582 for GET_MODE_NUNITS.
6583 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6584 for GET_MODE_NUNITS.
6585 * tree.c (make_vector_type): Remove temporary shim and make
6586 the real function take the number of units as a poly_uint64
6587 rather than an int.
6588 (build_vector_type_for_mode): Handle polynomial nunits.
6589 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6590 * emit-rtl.c (const_vec_series_p_1): Likewise.
6591 (gen_rtx_CONST_VECTOR): Likewise.
6592 * fold-const.c (test_vec_duplicate_folding): Likewise.
6593 * genrecog.c (validate_pattern): Likewise.
6594 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6595 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6596 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6597 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6598 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6599 * rtlanal.c (subreg_get_info): Likewise.
6600 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6601 (vect_grouped_load_supported): Likewise.
6602 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6603 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6604 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6605 (simplify_const_unary_operation, simplify_binary_operation_1)
6606 (simplify_const_binary_operation, simplify_ternary_operation)
6607 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6608 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6609 instead of CONST_VECTOR_NUNITS.
6610 * varasm.c (output_constant_pool_2): Likewise.
6611 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6612 explicit-encoded elements in the XVEC for variable-length vectors.
6614 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6616 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6618 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6619 Alan Hayward <alan.hayward@arm.com>
6620 David Sherwood <david.sherwood@arm.com>
6622 * coretypes.h (fixed_size_mode): Declare.
6623 (fixed_size_mode_pod): New typedef.
6624 * builtins.h (target_builtins::x_apply_args_mode)
6625 (target_builtins::x_apply_result_mode): Change type to
6626 fixed_size_mode_pod.
6627 * builtins.c (apply_args_size, apply_result_size, result_vector)
6628 (expand_builtin_apply_args_1, expand_builtin_apply)
6629 (expand_builtin_return): Update accordingly.
6631 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6633 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6634 * cselib.c (cselib_hash_rtx): Likewise.
6635 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6636 CONST_VECTOR encoding.
6638 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6639 Jeff Law <law@redhat.com>
6641 PR target/83641
6642 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6643 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6644 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6645 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6647 PR target/83641
6648 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6649 explicitly probe *sp in a noreturn function if there were any callee
6650 register saves or frame pointer is needed.
6652 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6654 PR debug/83621
6655 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6656 BLKmode for ternary, binary or unary expressions.
6658 PR debug/83645
6659 * var-tracking.c (delete_vta_debug_insn): New inline function.
6660 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6661 insns from get_insns () to NULL instead of each bb separately.
6662 Use delete_vta_debug_insn. No longer static.
6663 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6664 delete_vta_debug_insns callers.
6665 * rtl.h (delete_vta_debug_insns): Declare.
6666 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6667 instead of variable_tracking_main.
6669 2018-01-03 Martin Sebor <msebor@redhat.com>
6671 PR tree-optimization/83603
6672 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6673 arguments past the endof the argument list in functions declared
6674 without a prototype.
6675 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6676 Avoid checking when arguments are null.
6678 2018-01-03 Martin Sebor <msebor@redhat.com>
6680 PR c/83559
6681 * doc/extend.texi (attribute const): Fix a typo.
6682 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6683 issuing -Wsuggest-attribute for void functions.
6685 2018-01-03 Martin Sebor <msebor@redhat.com>
6687 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6688 offset_int::from instead of wide_int::to_shwi.
6689 (maybe_diag_overlap): Remove assertion.
6690 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6691 * gimple-ssa-sprintf.c (format_directive): Same.
6692 (parse_directive): Same.
6693 (sprintf_dom_walker::compute_format_length): Same.
6694 (try_substitute_return_value): Same.
6696 2018-01-03 Jeff Law <law@redhat.com>
6698 PR middle-end/83654
6699 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6700 non-constant residual for zero at runtime and avoid probing in
6701 that case. Reorganize code for trailing problem to mirror handling
6702 of the residual.
6704 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6706 PR tree-optimization/83501
6707 * tree-ssa-strlen.c (get_string_cst): New.
6708 (handle_char_store): Call get_string_cst.
6710 2018-01-03 Martin Liska <mliska@suse.cz>
6712 PR tree-optimization/83593
6713 * tree-ssa-strlen.c: Include tree-cfg.h.
6714 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6715 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6716 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6717 to false.
6718 (strlen_dom_walker::before_dom_children): Call
6719 gimple_purge_dead_eh_edges. Dump tranformation with details
6720 dump flags.
6721 (strlen_dom_walker::before_dom_children): Update call by adding
6722 new argument cleanup_eh.
6723 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6725 2018-01-03 Martin Liska <mliska@suse.cz>
6727 PR ipa/83549
6728 * cif-code.def (VARIADIC_THUNK): New enum value.
6729 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6730 thunks.
6732 2018-01-03 Jan Beulich <jbeulich@suse.com>
6734 * sse.md (mov<mode>_internal): Tighten condition for when to use
6735 vmovdqu<ssescalarsize> for TI and OI modes.
6737 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6739 Update copyright years.
6741 2018-01-03 Martin Liska <mliska@suse.cz>
6743 PR ipa/83594
6744 * ipa-visibility.c (function_and_variable_visibility): Skip
6745 functions with noipa attribure.
6747 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6749 * gcc.c (process_command): Update copyright notice dates.
6750 * gcov-dump.c (print_version): Ditto.
6751 * gcov.c (print_version): Ditto.
6752 * gcov-tool.c (print_version): Ditto.
6753 * gengtype.c (create_file): Ditto.
6754 * doc/cpp.texi: Bump @copying's copyright year.
6755 * doc/cppinternals.texi: Ditto.
6756 * doc/gcc.texi: Ditto.
6757 * doc/gccint.texi: Ditto.
6758 * doc/gcov.texi: Ditto.
6759 * doc/install.texi: Ditto.
6760 * doc/invoke.texi: Ditto.
6762 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6764 * vector-builder.h (vector_builder::m_full_nelts): Change from
6765 unsigned int to poly_uint64.
6766 (vector_builder::full_nelts): Update prototype accordingly.
6767 (vector_builder::new_vector): Likewise.
6768 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6769 (vector_builder::operator ==): Likewise.
6770 (vector_builder::finalize): Likewise.
6771 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6772 Take the number of elements as a poly_uint64 rather than an
6773 unsigned int.
6774 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6775 from unsigned int to poly_uint64.
6776 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6777 (vec_perm_indices::new_vector): Likewise.
6778 (vec_perm_indices::length): Likewise.
6779 (vec_perm_indices::nelts_per_input): Likewise.
6780 (vec_perm_indices::input_nelts): Likewise.
6781 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6782 number of elements per input as a poly_uint64 rather than an
6783 unsigned int. Use the original encoding for variable-length
6784 vectors, rather than clamping each individual element.
6785 For the second and subsequent elements in each pattern,
6786 clamp the step and base before clamping their sum.
6787 (vec_perm_indices::series_p): Handle polynomial element counts.
6788 (vec_perm_indices::all_in_range_p): Likewise.
6789 (vec_perm_indices_to_tree): Likewise.
6790 (vec_perm_indices_to_rtx): Likewise.
6791 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6792 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6793 (tree_vector_builder::new_binary_operation): Handle polynomial
6794 element counts. Return false if we need to know the number
6795 of elements at compile time.
6796 * fold-const.c (fold_vec_perm): Punt if the number of elements
6797 isn't known at compile time.
6799 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6801 * vec-perm-indices.h (vec_perm_builder): Change element type
6802 from HOST_WIDE_INT to poly_int64.
6803 (vec_perm_indices::element_type): Update accordingly.
6804 (vec_perm_indices::clamp): Handle polynomial element_types.
6805 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6806 (vec_perm_indices::all_in_range_p): Likewise.
6807 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6808 than shwi trees.
6809 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6810 polynomial vec_perm_indices element types.
6811 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6812 * fold-const.c (fold_vec_perm): Likewise.
6813 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6814 * tree-vect-generic.c (lower_vec_perm): Likewise.
6815 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6816 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6817 element type to HOST_WIDE_INT.
6819 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6820 Alan Hayward <alan.hayward@arm.com>
6821 David Sherwood <david.sherwood@arm.com>
6823 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6824 rather than an int. Use plus_constant.
6825 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6826 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6828 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6829 Alan Hayward <alan.hayward@arm.com>
6830 David Sherwood <david.sherwood@arm.com>
6832 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6833 a HOST_WIDE_INT to a poly_int64.
6835 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6836 Alan Hayward <alan.hayward@arm.com>
6837 David Sherwood <david.sherwood@arm.com>
6839 * calls.c (load_register_parameters): Cope with polynomial
6840 mode sizes. Require a constant size for BLKmode parameters
6841 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6842 forces a parameter to be padded at the lsb end in order to
6843 fill a complete number of words, require the parameter size
6844 to be ordered wrt UNITS_PER_WORD.
6846 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6847 Alan Hayward <alan.hayward@arm.com>
6848 David Sherwood <david.sherwood@arm.com>
6850 * reload1.c (spill_stack_slot_width): Change element type
6851 from unsigned int to poly_uint64_pod.
6852 (alter_reg): Treat mode sizes as polynomial.
6854 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6855 Alan Hayward <alan.hayward@arm.com>
6856 David Sherwood <david.sherwood@arm.com>
6858 * reload.c (complex_word_subreg_p): New function.
6859 (reload_inner_reg_of_subreg, push_reload): Use it.
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 * lra-constraints.c (process_alt_operands): Reject matched
6866 operands whose sizes aren't ordered.
6867 (match_reload): Refer to this check here.
6869 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6870 Alan Hayward <alan.hayward@arm.com>
6871 David Sherwood <david.sherwood@arm.com>
6873 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6874 that the mode size is in the set {1, 2, 4, 8, 16}.
6876 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6877 Alan Hayward <alan.hayward@arm.com>
6878 David Sherwood <david.sherwood@arm.com>
6880 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6881 Use plus_constant instead of gen_rtx_PLUS.
6883 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6884 Alan Hayward <alan.hayward@arm.com>
6885 David Sherwood <david.sherwood@arm.com>
6887 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6888 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6889 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6890 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6891 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6892 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6893 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6894 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6895 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6896 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6897 a poly_int64.
6898 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6899 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6900 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6901 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6902 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6903 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6904 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6905 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6906 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6907 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6908 function.
6909 * expr.c (emit_move_resolve_push): Treat the input and result
6910 of PUSH_ROUNDING as a poly_int64.
6911 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6912 (emit_push_insn): Likewise.
6913 * lra-eliminations.c (mark_not_eliminable): Likewise.
6914 * recog.c (push_operand): Likewise.
6915 * reload1.c (elimination_effects): Likewise.
6916 * rtlanal.c (nonzero_bits1): Likewise.
6917 * calls.c (store_one_arg): Likewise. Require the padding to be
6918 known at compile time.
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 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6925 Use plus_constant instead of gen_rtx_PLUS.
6927 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6928 Alan Hayward <alan.hayward@arm.com>
6929 David Sherwood <david.sherwood@arm.com>
6931 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6932 rather than an int.
6934 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6935 Alan Hayward <alan.hayward@arm.com>
6936 David Sherwood <david.sherwood@arm.com>
6938 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6939 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6940 via stack temporaries. Treat the mode size as polynomial too.
6942 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6943 Alan Hayward <alan.hayward@arm.com>
6944 David Sherwood <david.sherwood@arm.com>
6946 * expr.c (expand_expr_real_2): When handling conversions involving
6947 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6948 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6949 as a poly_uint64 too.
6951 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6952 Alan Hayward <alan.hayward@arm.com>
6953 David Sherwood <david.sherwood@arm.com>
6955 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6957 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6958 Alan Hayward <alan.hayward@arm.com>
6959 David Sherwood <david.sherwood@arm.com>
6961 * combine.c (can_change_dest_mode): Handle polynomial
6962 REGMODE_NATURAL_SIZE.
6963 * expmed.c (store_bit_field_1): Likewise.
6964 * expr.c (store_constructor): Likewise.
6965 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6966 and polynomial REGMODE_NATURAL_SIZE.
6967 (gen_lowpart_common): Likewise.
6968 * reginfo.c (record_subregs_of_mode): Likewise.
6969 * rtlanal.c (read_modify_subreg_p): Likewise.
6971 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6972 Alan Hayward <alan.hayward@arm.com>
6973 David Sherwood <david.sherwood@arm.com>
6975 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6976 numbers of elements.
6978 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6979 Alan Hayward <alan.hayward@arm.com>
6980 David Sherwood <david.sherwood@arm.com>
6982 * match.pd: Cope with polynomial numbers of vector elements.
6984 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6985 Alan Hayward <alan.hayward@arm.com>
6986 David Sherwood <david.sherwood@arm.com>
6988 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6989 in a POINTER_PLUS_EXPR.
6991 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6992 Alan Hayward <alan.hayward@arm.com>
6993 David Sherwood <david.sherwood@arm.com>
6995 * omp-simd-clone.c (simd_clone_subparts): New function.
6996 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6997 (ipa_simd_modify_function_body): Likewise.
6999 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7000 Alan Hayward <alan.hayward@arm.com>
7001 David Sherwood <david.sherwood@arm.com>
7003 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
7004 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
7005 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
7006 (expand_vector_condition, vector_element): Likewise.
7007 (subparts_gt): New function.
7008 (get_compute_type): Use subparts_gt.
7009 (count_type_subparts): Delete.
7010 (expand_vector_operations_1): Use subparts_gt instead of
7011 count_type_subparts.
7013 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7014 Alan Hayward <alan.hayward@arm.com>
7015 David Sherwood <david.sherwood@arm.com>
7017 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
7018 (vect_compile_time_alias): ...this new function. Do the calculation
7019 on poly_ints rather than trees.
7020 (vect_prune_runtime_alias_test_list): Update call accordingly.
7022 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7023 Alan Hayward <alan.hayward@arm.com>
7024 David Sherwood <david.sherwood@arm.com>
7026 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
7027 numbers of units.
7028 (vect_schedule_slp_instance): Likewise.
7030 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7031 Alan Hayward <alan.hayward@arm.com>
7032 David Sherwood <david.sherwood@arm.com>
7034 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
7035 constant and extern definitions for variable-length vectors.
7036 (vect_get_constant_vectors): Note that the number of units
7037 is known to be constant.
7039 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7040 Alan Hayward <alan.hayward@arm.com>
7041 David Sherwood <david.sherwood@arm.com>
7043 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
7044 of units as polynomial. Choose between WIDE and NARROW based
7045 on multiple_p.
7047 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7048 Alan Hayward <alan.hayward@arm.com>
7049 David Sherwood <david.sherwood@arm.com>
7051 * tree-vect-stmts.c (simd_clone_subparts): New function.
7052 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
7054 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7055 Alan Hayward <alan.hayward@arm.com>
7056 David Sherwood <david.sherwood@arm.com>
7058 * tree-vect-stmts.c (vectorizable_call): Treat the number of
7059 vectors as polynomial. Use build_index_vector for
7060 IFN_GOMP_SIMD_LANE.
7062 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7063 Alan Hayward <alan.hayward@arm.com>
7064 David Sherwood <david.sherwood@arm.com>
7066 * tree-vect-stmts.c (get_load_store_type): Treat the number of
7067 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
7068 for variable-length vectors.
7069 (vectorizable_mask_load_store): Treat the number of units as
7070 polynomial, asserting that it is constant if the condition has
7071 already been enforced.
7072 (vectorizable_store, vectorizable_load): Likewise.
7074 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7075 Alan Hayward <alan.hayward@arm.com>
7076 David Sherwood <david.sherwood@arm.com>
7078 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
7079 of units as polynomial. Punt if we can't tell at compile time
7080 which vector contains the final result.
7082 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7083 Alan Hayward <alan.hayward@arm.com>
7084 David Sherwood <david.sherwood@arm.com>
7086 * tree-vect-loop.c (vectorizable_induction): Treat the number
7087 of units as polynomial. Punt on SLP inductions. Use an integer
7088 VEC_SERIES_EXPR for variable-length integer reductions. Use a
7089 cast of such a series for variable-length floating-point
7090 reductions.
7092 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7093 Alan Hayward <alan.hayward@arm.com>
7094 David Sherwood <david.sherwood@arm.com>
7096 * tree.h (build_index_vector): Declare.
7097 * tree.c (build_index_vector): New function.
7098 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
7099 of units as polynomial, forcibly converting it to a constant if
7100 vectorizable_reduction has already enforced the condition.
7101 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
7102 to create a {1,2,3,...} vector.
7103 (vectorizable_reduction): Treat the number of units as polynomial.
7104 Choose vectype_in based on the largest scalar element size rather
7105 than the smallest number of units. Enforce the restrictions
7106 relied on above.
7108 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7109 Alan Hayward <alan.hayward@arm.com>
7110 David Sherwood <david.sherwood@arm.com>
7112 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
7113 number of units as polynomial.
7115 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7116 Alan Hayward <alan.hayward@arm.com>
7117 David Sherwood <david.sherwood@arm.com>
7119 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
7120 * target.def (autovectorize_vector_sizes): Return the vector sizes
7121 by pointer, using vector_sizes rather than a bitmask.
7122 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
7123 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
7124 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
7125 Likewise.
7126 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
7127 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
7128 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
7129 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
7130 * omp-general.c (omp_max_vf): Likewise.
7131 * omp-low.c (omp_clause_aligned_alignment): Likewise.
7132 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
7133 * tree-vect-loop.c (vect_analyze_loop): Likewise.
7134 * tree-vect-slp.c (vect_slp_bb): Likewise.
7135 * doc/tm.texi: Regenerate.
7136 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
7137 to a poly_uint64.
7138 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
7139 the vector size as a poly_uint64 rather than an unsigned int.
7140 (current_vector_size): Change from an unsigned int to a poly_uint64.
7141 (get_vectype_for_scalar_type): Update accordingly.
7142 * tree.h (build_truth_vector_type): Take the size and number of
7143 units as a poly_uint64 rather than an unsigned int.
7144 (build_vector_type): Add a temporary overload that takes
7145 the number of units as a poly_uint64 rather than an unsigned int.
7146 * tree.c (make_vector_type): Likewise.
7147 (build_truth_vector_type): Take the number of units as a poly_uint64
7148 rather than an unsigned int.
7150 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7151 Alan Hayward <alan.hayward@arm.com>
7152 David Sherwood <david.sherwood@arm.com>
7154 * target.def (get_mask_mode): Take the number of units and length
7155 as poly_uint64s rather than unsigned ints.
7156 * targhooks.h (default_get_mask_mode): Update accordingly.
7157 * targhooks.c (default_get_mask_mode): Likewise.
7158 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
7159 * doc/tm.texi: Regenerate.
7161 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7162 Alan Hayward <alan.hayward@arm.com>
7163 David Sherwood <david.sherwood@arm.com>
7165 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
7166 * omp-general.c (omp_max_vf): Likewise.
7167 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
7168 (expand_omp_simd): Handle polynomial safelen.
7169 * omp-low.c (omplow_simd_context): Add a default constructor.
7170 (omplow_simd_context::max_vf): Change from int to poly_uint64.
7171 (lower_rec_simd_input_clauses): Update accordingly.
7172 (lower_rec_input_clauses): Likewise.
7174 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7175 Alan Hayward <alan.hayward@arm.com>
7176 David Sherwood <david.sherwood@arm.com>
7178 * tree-vectorizer.h (vect_nunits_for_cost): New function.
7179 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
7180 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
7181 (vect_analyze_slp_cost): Likewise.
7182 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
7183 (vect_model_load_cost): Likewise.
7185 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7186 Alan Hayward <alan.hayward@arm.com>
7187 David Sherwood <david.sherwood@arm.com>
7189 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
7190 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
7191 from an unsigned int * to a poly_uint64_pod *.
7192 (calculate_unrolling_factor): New function.
7193 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
7195 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7196 Alan Hayward <alan.hayward@arm.com>
7197 David Sherwood <david.sherwood@arm.com>
7199 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
7200 from an unsigned int to a poly_uint64.
7201 (_loop_vec_info::slp_unrolling_factor): Likewise.
7202 (_loop_vec_info::vectorization_factor): Change from an int
7203 to a poly_uint64.
7204 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
7205 (vect_get_num_vectors): New function.
7206 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
7207 (vect_get_num_copies): Use vect_get_num_vectors.
7208 (vect_analyze_data_ref_dependences): Change max_vf from an int *
7209 to an unsigned int *.
7210 (vect_analyze_data_refs): Change min_vf from an int * to a
7211 poly_uint64 *.
7212 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7213 than an unsigned HOST_WIDE_INT.
7214 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
7215 (vect_analyze_data_ref_dependence): Change max_vf from an int *
7216 to an unsigned int *.
7217 (vect_analyze_data_ref_dependences): Likewise.
7218 (vect_compute_data_ref_alignment): Handle polynomial vf.
7219 (vect_enhance_data_refs_alignment): Likewise.
7220 (vect_prune_runtime_alias_test_list): Likewise.
7221 (vect_shift_permute_load_chain): Likewise.
7222 (vect_supportable_dr_alignment): Likewise.
7223 (dependence_distance_ge_vf): Take the vectorization factor as a
7224 poly_uint64 rather than an unsigned HOST_WIDE_INT.
7225 (vect_analyze_data_refs): Change min_vf from an int * to a
7226 poly_uint64 *.
7227 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
7228 vfm1 as a poly_uint64 rather than an int. Make the same change
7229 for the returned bound_scalar.
7230 (vect_gen_vector_loop_niters): Handle polynomial vf.
7231 (vect_do_peeling): Likewise. Update call to
7232 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
7233 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
7234 be constant.
7235 * tree-vect-loop.c (vect_determine_vectorization_factor)
7236 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
7237 (vect_get_known_peeling_cost): Likewise.
7238 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
7239 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
7240 (vect_transform_loop): Likewise. Use the lowest possible VF when
7241 updating the upper bounds of the loop.
7242 (vect_min_worthwhile_factor): Make static. Return an unsigned int
7243 rather than an int.
7244 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
7245 polynomial unroll factors.
7246 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
7247 (vect_make_slp_decision): Likewise.
7248 (vect_supported_load_permutation_p): Likewise, and polynomial
7249 vf too.
7250 (vect_analyze_slp_cost): Handle polynomial vf.
7251 (vect_slp_analyze_node_operations): Likewise.
7252 (vect_slp_analyze_bb_1): Likewise.
7253 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7254 than an unsigned HOST_WIDE_INT.
7255 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
7256 (vectorizable_load): Handle polynomial vf.
7257 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
7258 a poly_uint64.
7259 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
7261 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7262 Alan Hayward <alan.hayward@arm.com>
7263 David Sherwood <david.sherwood@arm.com>
7265 * match.pd: Handle bit operations involving three constants
7266 and try to fold one pair.
7268 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7270 * tree-vect-loop-manip.c: Include gimple-fold.h.
7271 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
7272 niters_maybe_zero parameters. Handle other cases besides a step of 1.
7273 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
7274 Add a path that uses a step of VF instead of 1, but disable it
7275 for now.
7276 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
7277 and niters_no_overflow parameters. Update calls to
7278 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
7279 Create a new SSA name if the latter choses to use a ste other
7280 than zero, and return it via niters_vector_mult_vf_var.
7281 * tree-vect-loop.c (vect_transform_loop): Update calls to
7282 vect_do_peeling, vect_gen_vector_loop_niters and
7283 slpeel_make_loop_iterate_ntimes.
7284 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
7285 (vect_gen_vector_loop_niters): Update declarations after above changes.
7287 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
7289 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
7290 128-bit round to integer instructions.
7291 (ceil<mode>2): Likewise.
7292 (btrunc<mode>2): Likewise.
7293 (round<mode>2): Likewise.
7295 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7297 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
7298 unaligned VSX load/store on P8/P9.
7299 (expand_block_clear): Allow the use of unaligned VSX
7300 load/store on P8/P9.
7302 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
7304 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
7305 New function.
7306 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
7307 swap associated with both a load and a store.
7309 2018-01-02 Andrew Waterman <andrew@sifive.com>
7311 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
7312 * config/riscv/riscv.md (clear_cache): Use it.
7314 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
7316 * web.c: Remove out-of-date comment.
7318 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7320 * expr.c (fixup_args_size_notes): Check that any existing
7321 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
7322 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
7323 (emit_single_push_insn): ...here.
7325 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7327 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
7328 (const_vector_encoded_nelts): New function.
7329 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
7330 (const_vector_int_elt, const_vector_elt): Declare.
7331 * emit-rtl.c (const_vector_int_elt_1): New function.
7332 (const_vector_elt): Likewise.
7333 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
7334 of CONST_VECTOR_ELT.
7336 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7338 * expr.c: Include rtx-vector-builder.h.
7339 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
7340 directly on the tree encoding.
7341 (const_vector_from_tree): Likewise.
7342 * optabs.c: Include rtx-vector-builder.h.
7343 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
7344 sequence of "u" values.
7345 * vec-perm-indices.c: Include rtx-vector-builder.h.
7346 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
7347 directly on the vec_perm_indices encoding.
7349 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7351 * doc/rtl.texi (const_vector): Describe new encoding scheme.
7352 * Makefile.in (OBJS): Add rtx-vector-builder.o.
7353 * rtx-vector-builder.h: New file.
7354 * rtx-vector-builder.c: Likewise.
7355 * rtl.h (rtx_def::u2): Add a const_vector field.
7356 (CONST_VECTOR_NPATTERNS): New macro.
7357 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
7358 (CONST_VECTOR_DUPLICATE_P): Likewise.
7359 (CONST_VECTOR_STEPPED_P): Likewise.
7360 (CONST_VECTOR_ENCODED_ELT): Likewise.
7361 (const_vec_duplicate_p): Check for a duplicated vector encoding.
7362 (unwrap_const_vec_duplicate): Likewise.
7363 (const_vec_series_p): Check for a non-duplicated vector encoding.
7364 Say that the function only returns true for integer vectors.
7365 * emit-rtl.c: Include rtx-vector-builder.h.
7366 (gen_const_vec_duplicate_1): Delete.
7367 (gen_const_vector): Call gen_const_vec_duplicate instead of
7368 gen_const_vec_duplicate_1.
7369 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
7370 (gen_const_vec_duplicate): Use rtx_vector_builder.
7371 (gen_const_vec_series): Likewise.
7372 (gen_rtx_CONST_VECTOR): Likewise.
7373 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
7374 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7375 Build a new vector rather than modifying a CONST_VECTOR in-place.
7376 (handle_special_swappables): Update call accordingly.
7377 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
7378 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7379 Build a new vector rather than modifying a CONST_VECTOR in-place.
7380 (handle_special_swappables): Update call accordingly.
7382 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7384 * simplify-rtx.c (simplify_const_binary_operation): Use
7385 CONST_VECTOR_ELT instead of XVECEXP.
7387 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7389 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
7390 the selector elements to be different from the data elements
7391 if the selector is a VECTOR_CST.
7392 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
7393 ssizetype for the selector.
7395 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7397 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
7398 before testing each element individually.
7399 * tree-vect-generic.c (lower_vec_perm): Likewise.
7401 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7403 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
7404 * selftest-run-tests.c (selftest::run_tests): Call it.
7405 * vector-builder.h (vector_builder::operator ==): New function.
7406 (vector_builder::operator !=): Likewise.
7407 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
7408 (vec_perm_indices::all_from_input_p): New function.
7409 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7410 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
7411 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
7412 instead of reading the VECTOR_CST directly. Detect whether both
7413 vector inputs are the same before constructing the vec_perm_indices,
7414 and update the number of inputs argument accordingly. Use the
7415 utility functions added above. Only construct sel2 if we need to.
7417 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7419 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
7420 the broadcast of the low byte.
7421 (expand_mult_highpart): Use an explicit encoding for the permutes.
7422 * optabs-query.c (can_mult_highpart_p): Likewise.
7423 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
7424 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7425 (vectorizable_bswap): Likewise.
7426 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
7427 explicit encoding for the power-of-2 permutes.
7428 (vect_permute_store_chain): Likewise.
7429 (vect_grouped_load_supported): Likewise.
7430 (vect_permute_load_chain): Likewise.
7432 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7434 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
7435 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
7436 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
7437 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7438 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
7439 (vect_gen_perm_mask_any): Likewise.
7441 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7443 * int-vector-builder.h: New file.
7444 * vec-perm-indices.h: Include int-vector-builder.h.
7445 (vec_perm_indices): Redefine as an int_vector_builder.
7446 (auto_vec_perm_indices): Delete.
7447 (vec_perm_builder): Redefine as a stand-alone class.
7448 (vec_perm_indices::vec_perm_indices): New function.
7449 (vec_perm_indices::clamp): Likewise.
7450 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
7451 (vec_perm_indices::new_vector): New function.
7452 (vec_perm_indices::new_expanded_vector): Update for new
7453 vec_perm_indices class.
7454 (vec_perm_indices::rotate_inputs): New function.
7455 (vec_perm_indices::all_in_range_p): Operate directly on the
7456 encoded form, without computing elided elements.
7457 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
7458 encoding. Update for new vec_perm_indices class.
7459 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
7460 the given vec_perm_builder.
7461 (expand_vec_perm_var): Update vec_perm_builder constructor.
7462 (expand_mult_highpart): Use vec_perm_builder instead of
7463 auto_vec_perm_indices.
7464 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
7465 vec_perm_indices instead of auto_vec_perm_indices. Use a single
7466 or double series encoding as appropriate.
7467 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
7468 vec_perm_indices instead of auto_vec_perm_indices.
7469 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7470 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7471 (vect_permute_store_chain): Likewise.
7472 (vect_grouped_load_supported): Likewise.
7473 (vect_permute_load_chain): Likewise.
7474 (vect_shift_permute_load_chain): Likewise.
7475 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7476 (vect_transform_slp_perm_load): Likewise.
7477 (vect_schedule_slp_instance): Likewise.
7478 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7479 (vectorizable_mask_load_store): Likewise.
7480 (vectorizable_bswap): Likewise.
7481 (vectorizable_store): Likewise.
7482 (vectorizable_load): Likewise.
7483 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
7484 vec_perm_indices instead of auto_vec_perm_indices. Use
7485 tree_to_vec_perm_builder to read the vector from a tree.
7486 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
7487 vec_perm_builder instead of a vec_perm_indices.
7488 (have_whole_vector_shift): Use vec_perm_builder and
7489 vec_perm_indices instead of auto_vec_perm_indices. Leave the
7490 truncation to calc_vec_perm_mask_for_shift.
7491 (vect_create_epilog_for_reduction): Likewise.
7492 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
7493 from auto_vec_perm_indices to vec_perm_indices.
7494 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7495 instead of changing individual elements.
7496 (aarch64_vectorize_vec_perm_const): Use new_vector to install
7497 the vector in d.perm.
7498 * config/arm/arm.c (expand_vec_perm_d::perm): Change
7499 from auto_vec_perm_indices to vec_perm_indices.
7500 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7501 instead of changing individual elements.
7502 (arm_vectorize_vec_perm_const): Use new_vector to install
7503 the vector in d.perm.
7504 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
7505 Update vec_perm_builder constructor.
7506 (rs6000_expand_interleave): Likewise.
7507 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
7508 (rs6000_expand_interleave): Likewise.
7510 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7512 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
7513 to qimode could truncate the indices.
7514 * optabs.c (expand_vec_perm_var): Likewise.
7516 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7518 * Makefile.in (OBJS): Add vec-perm-indices.o.
7519 * vec-perm-indices.h: New file.
7520 * vec-perm-indices.c: Likewise.
7521 * target.h (vec_perm_indices): Replace with a forward class
7522 declaration.
7523 (auto_vec_perm_indices): Move to vec-perm-indices.h.
7524 * optabs.h: Include vec-perm-indices.h.
7525 (expand_vec_perm): Delete.
7526 (selector_fits_mode_p, expand_vec_perm_var): Declare.
7527 (expand_vec_perm_const): Declare.
7528 * target.def (vec_perm_const_ok): Replace with...
7529 (vec_perm_const): ...this new hook.
7530 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
7531 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
7532 * doc/tm.texi: Regenerate.
7533 * optabs.def (vec_perm_const): Delete.
7534 * doc/md.texi (vec_perm_const): Likewise.
7535 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
7536 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7537 expand_vec_perm for constant permutation vectors. Assert that
7538 the mode of variable permutation vectors is the integer equivalent
7539 of the mode that is being permuted.
7540 * optabs-query.h (selector_fits_mode_p): Declare.
7541 * optabs-query.c: Include vec-perm-indices.h.
7542 (selector_fits_mode_p): New function.
7543 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7544 is defined, instead of checking whether the vec_perm_const_optab
7545 exists. Use targetm.vectorize.vec_perm_const instead of
7546 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7547 fit in the vector mode before using a variable permute.
7548 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7549 vec_perm_indices instead of an rtx.
7550 (expand_vec_perm): Replace with...
7551 (expand_vec_perm_const): ...this new function. Take the selector
7552 as a vec_perm_indices rather than an rtx. Also take the mode of
7553 the selector. Update call to shift_amt_for_vec_perm_mask.
7554 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7555 Use vec_perm_indices::new_expanded_vector to expand the original
7556 selector into bytes. Check whether the indices fit in the vector
7557 mode before using a variable permute.
7558 (expand_vec_perm_var): Make global.
7559 (expand_mult_highpart): Use expand_vec_perm_const.
7560 * fold-const.c: Includes vec-perm-indices.h.
7561 * tree-ssa-forwprop.c: Likewise.
7562 * tree-vect-data-refs.c: Likewise.
7563 * tree-vect-generic.c: Likewise.
7564 * tree-vect-loop.c: Likewise.
7565 * tree-vect-slp.c: Likewise.
7566 * tree-vect-stmts.c: Likewise.
7567 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7568 Delete.
7569 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7570 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7571 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7572 (aarch64_vectorize_vec_perm_const): ...this new function.
7573 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7574 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7575 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7576 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7577 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7578 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7579 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7580 into...
7581 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7582 check for NEON modes.
7583 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7584 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7585 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7586 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7587 into...
7588 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7589 the old VEC_PERM_CONST conditions.
7590 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7591 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7592 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7593 (ia64_vectorize_vec_perm_const_ok): Merge into...
7594 (ia64_vectorize_vec_perm_const): ...this new function.
7595 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7596 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7597 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7598 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7599 * config/mips/mips.c (mips_expand_vec_perm_const)
7600 (mips_vectorize_vec_perm_const_ok): Merge into...
7601 (mips_vectorize_vec_perm_const): ...this new function.
7602 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7603 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7604 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7605 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7606 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7607 (rs6000_expand_vec_perm_const): Delete.
7608 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7609 Delete.
7610 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7611 (altivec_expand_vec_perm_const_le): Take each operand individually.
7612 Operate on constant selectors rather than rtxes.
7613 (altivec_expand_vec_perm_const): Likewise. Update call to
7614 altivec_expand_vec_perm_const_le.
7615 (rs6000_expand_vec_perm_const): Delete.
7616 (rs6000_vectorize_vec_perm_const_ok): Delete.
7617 (rs6000_vectorize_vec_perm_const): New function.
7618 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7619 an element count and rtx array.
7620 (rs6000_expand_extract_even): Update call accordingly.
7621 (rs6000_expand_interleave): Likewise.
7622 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7623 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7624 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7625 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7626 (rs6000_expand_vec_perm_const): Delete.
7627 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7628 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7629 (altivec_expand_vec_perm_const_le): Take each operand individually.
7630 Operate on constant selectors rather than rtxes.
7631 (altivec_expand_vec_perm_const): Likewise. Update call to
7632 altivec_expand_vec_perm_const_le.
7633 (rs6000_expand_vec_perm_const): Delete.
7634 (rs6000_vectorize_vec_perm_const_ok): Delete.
7635 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7636 reference to the SPE evmerge intructions.
7637 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7638 an element count and rtx array.
7639 (rs6000_expand_extract_even): Update call accordingly.
7640 (rs6000_expand_interleave): Likewise.
7641 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7642 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7643 new function.
7644 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7646 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7648 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7649 vector mode and that that mode matches the mode of the data
7650 being permuted.
7651 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7652 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7653 directly using expand_vec_perm_1 when forcing selectors into
7654 registers.
7655 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7657 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7659 * optabs-query.h (can_vec_perm_p): Delete.
7660 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7661 * optabs-query.c (can_vec_perm_p): Split into...
7662 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7663 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7664 particular selector is valid.
7665 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7666 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7667 (vect_grouped_load_supported): Likewise.
7668 (vect_shift_permute_load_chain): Likewise.
7669 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7670 (vect_transform_slp_perm_load): Likewise.
7671 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7672 (vectorizable_bswap): Likewise.
7673 (vect_gen_perm_mask_checked): Likewise.
7674 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7675 implementations of variable permutation vectors into account
7676 when deciding which selector to use.
7677 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7678 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7679 with a false third argument.
7680 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7681 to test whether the constant selector is valid and can_vec_perm_var_p
7682 to test whether a variable selector is valid.
7684 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7686 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7687 * optabs-query.c (can_vec_perm_p): Likewise.
7688 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7689 instead of vec_perm_indices.
7690 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7691 (vect_gen_perm_mask_checked): Likewise,
7692 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7693 (vect_gen_perm_mask_checked): Likewise,
7695 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7697 * optabs-query.h (qimode_for_vec_perm): Declare.
7698 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7699 (qimode_for_vec_perm): ...this new function.
7700 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7702 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7704 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7705 does not have a conditional at the top.
7707 2018-01-02 Richard Biener <rguenther@suse.de>
7709 * ipa-inline.c (big_speedup_p): Fix expression.
7711 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7713 PR target/81616
7714 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7715 for generic 4->6.
7717 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7719 PR target/81616
7720 Generic tuning.
7721 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7722 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7723 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7724 cond_taken_branch_cost 3->4.
7726 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7728 PR tree-optimization/83581
7729 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7730 TODO_cleanup_cfg if any changes have been made.
7732 PR middle-end/83608
7733 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7734 convert_modes if target mode has the right side, but different mode
7735 class.
7737 PR middle-end/83609
7738 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7739 last argument when extracting from CONCAT. If either from_real or
7740 from_imag is NULL, use expansion through memory. If result is not
7741 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7742 the parts directly to inner mode, if even that fails, use expansion
7743 through memory.
7745 PR middle-end/83623
7746 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7747 check for bswap in mode rather than HImode and use that in expand_unop
7748 too.
7749 \f
7750 Copyright (C) 2018 Free Software Foundation, Inc.
7752 Copying and distribution of this file, with or without modification,
7753 are permitted in any medium without royalty provided the copyright
7754 notice and this notice are preserved.