1 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
3 PR tree-optimization/81635
4 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
5 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
6 (test_round_for_mask): New functions.
7 (wide_int_cc_tests): Call test_round_for_mask.
8 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
9 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
10 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
11 range returned by get_range_info.
13 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
16 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
17 * symtab.c: Include builtins.h
18 (symtab_node::output_to_lto_symbol_table_p): Move here
19 from lto-streamer-out.c:output_symbol_p.
20 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
21 (output_symbol_p): Move all logic to symtab.c
22 (produce_symtab): Update.
24 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
26 * config/s390/s390-opts.h (enum indirect_branch): Define.
27 * config/s390/s390-protos.h (s390_return_addr_from_memory)
28 (s390_indirect_branch_via_thunk)
29 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
30 (enum s390_indirect_branch_type): Define.
31 * config/s390/s390.c (struct s390_frame_layout, struct
32 machine_function): Remove.
33 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
34 (indirect_branch_table_label_no, indirect_branch_table_name):
36 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
37 (enum s390_indirect_branch_option): Define.
38 (s390_return_addr_from_memory): New function.
39 (s390_handle_string_attribute): New function.
40 (s390_attribute_table): Add new attribute handler.
41 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
42 (s390_indirect_branch_via_thunk): New function.
43 (s390_indirect_branch_via_inline_thunk): New function.
44 (s390_function_ok_for_sibcall): When jumping via thunk disallow
45 sibling call optimization for non z10 compiles.
46 (s390_emit_call): Force indirect branch target to be a single
47 register. Add r1 clobber for non-z10 compiles.
48 (s390_emit_epilogue): Emit return jump via return_use expander.
49 (s390_reorg): Handle JUMP_INSNs as execute targets.
50 (s390_option_override_internal): Perform validity checks for the
51 new command line options.
52 (s390_indirect_branch_attrvalue): New function.
53 (s390_indirect_branch_settings): New function.
54 (s390_set_current_function): Invoke s390_indirect_branch_settings.
55 (s390_output_indirect_thunk_function): New function.
56 (s390_code_end): Implement target hook.
57 (s390_case_values_threshold): Implement target hook.
58 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
60 * config/s390/s390.h (struct s390_frame_layout)
61 (struct machine_function): Move here from s390.c.
62 (TARGET_INDIRECT_BRANCH_NOBP_RET)
63 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
64 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
65 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
66 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
67 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
68 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
69 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
70 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
71 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
72 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
73 (mnemonic attribute): Add values which aren't recognized
75 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
76 pattern for branch conversion. Fix mnemonic attribute.
77 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
78 indirect branch via thunk if requested.
79 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
80 ("*indirect_jump"): Disable for branch conversion using out of
82 ("indirect_jump_via_thunk<mode>_z10")
83 ("indirect_jump_via_thunk<mode>")
84 ("indirect_jump_via_inlinethunk<mode>_z10")
85 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
86 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
87 ("casesi_jump_via_inlinethunk<mode>_z10")
88 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
89 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
90 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
91 ("*indirect2_jump"): Disable for branch conversion.
92 ("casesi_jump"): Turn into expander and expand patterns for branch
94 ("return_use"): New expander.
95 ("*return"): Emit return via thunk and rename it to ...
96 ("*return<mode>"): ... this one.
97 * config/s390/s390.opt: Add new options and and enum for the
100 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
102 * lra-constraints.c (match_reload): Unconditionally use
103 gen_lowpart_SUBREG, rather than selecting between that
104 and equivalent gen_rtx_SUBREG code.
106 2018-02-08 Richard Biener <rguenther@suse.de>
108 PR tree-optimization/84233
109 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
110 changed flag instead of boguously re-using phi_inserted.
112 2018-02-08 Martin Jambor <mjambor@suse.cz>
114 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
115 static local variables.
117 2018-02-08 Richard Biener <rguenther@suse.de>
119 PR tree-optimization/84278
120 * tree-vect-stmts.c (vectorizable_store): When looking for
121 smaller vector types to perform grouped strided loads/stores
122 make sure the mode is supported by the target.
123 (vectorizable_load): Likewise.
125 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
127 * config/aarch64/aarch64.c (aarch64_components_for_bb):
128 Increase LDP/STP opportunities by adding adjacent callee-saves.
130 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
132 PR rtl-optimization/84068
133 PR rtl-optimization/83459
134 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
136 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
138 PR tree-optimization/84224
139 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
140 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
143 2018-02-07 Iain Sandoe <iain@codesourcery.com>
145 * config/rs6000/altivec.md (*restore_world): Remove LR use.
146 * config/rs6000/predicates.md (restore_world_operation): Adjust op
147 count, remove one USE.
149 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
151 * doc/install.texi (Configuration): Document the
152 --with-long-double-format={ibm,ieee} PowerPC configuration
156 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
157 Convert from define_expand to be define_insn_and_split. Rework
158 float/double/_Float128 conversions to QI/HI/SImode to work with
159 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
160 conversions to QI/HImode types did a store and then a load to
161 truncate the value. For conversions to VSX registers, don't split
162 the insn, instead emit the code directly. Use the code iterator
163 any_fix to combine signed and unsigned conversions.
164 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
165 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
166 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
167 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
168 (fix_<mode>di2_hw): Likewise.
169 (fixuns_<mode>di2_hw): Likewise.
170 (fix_<mode>si2_hw): Likewise.
171 (fixuns_<mode>si2_hw): Likewise.
172 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
173 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
174 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
175 fix<uns>_trunc<SFDF:mode>si2_p8.
176 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
178 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
179 (fix<uns>_<mode>_mem): Likewise.
180 (fctiw<u>z_<mode>_mem): Likewise.
181 (fix<uns>_<mode>_mem): Likewise.
182 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
183 the register allocator from doing a direct move to the GPRs to do
184 a store, and instead use the ISA 3.0 store byte/half-word from
185 vector register instruction. For IEEE 128-bit floating point,
186 also optimize stores of 32-bit ints.
187 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
189 2018-02-07 Alan Hayward <alan.hayward@arm.com>
191 * genextract.c (push_pathstr_operand): New function to support
193 (walk_rtx): Call push_pathstr_operand.
194 (print_path): Support [a-zA-Z].
196 2018-02-07 Richard Biener <rguenther@suse.de>
198 PR tree-optimization/84037
199 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
200 (cse_and_gimplify_to_preheader): Declare.
201 (vect_get_place_in_interleaving_chain): Likewise.
202 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
204 (_loop_vec_info::~_loop_vec_info): Delete it.
205 (cse_and_gimplify_to_preheader): New function.
206 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
207 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
208 (vectorizable_load): Likewise. For grouped stores always base
209 the IV on the first element.
210 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
211 condition before gimplifying.
213 2018-02-07 Jakub Jelinek <jakub@redhat.com>
215 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
216 *DIV_EXPR and *MOD_EXPR.
218 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
221 * config/i386/i386.c (ix86_option_override_internal): Mask out
222 the CF_SET bit when checking -fcf-protection.
224 2018-02-07 Tom de Vries <tom@codesourcery.com>
227 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
230 2018-02-07 Richard Biener <rguenther@suse.de>
232 PR tree-optimization/84204
233 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
236 PR tree-optimization/84205
237 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
238 special-case isl_ast_op_zdiv_r.
240 PR tree-optimization/84223
241 * graphite-scop-detection.c (gather_bbs::before_dom_children):
242 Only add conditions from within the region.
243 (gather_bbs::after_dom_children): Adjust.
245 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
248 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
249 * config/avr/avr.md: Only post-reload split REG-REG moves if
250 either register is REGERAL_REG_P.
252 2018-02-07 Jakub Jelinek <jakub@redhat.com>
254 PR tree-optimization/84235
255 * tree-ssa-scopedtables.c
256 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
257 if the subtraction is performed in floating point type where NaNs are
258 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
259 build 1. Formatting fix.
261 2018-02-06 Jakub Jelinek <jakub@redhat.com>
264 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
265 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
266 and skip it regardless of bb boundaries. Use CALL_P macro,
267 don't test INSN_P (insn) together with CALL_P or JUMP_P check
268 unnecessarily, formatting fix.
270 2018-02-06 Michael Collison <michael.collison@arm.com>
272 * config/arm/thumb2.md:
273 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
274 (*thumb_mov_notscc): Ditto.
276 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
279 * config/rs6000/rs6000.md (su code attribute): Use "u" for
280 unsigned_fix, not "s".
282 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
284 * configure.ac (gcc_fn_eh_frame_ro): New function.
285 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
286 correct .eh_frame permissions.
287 * configure: Regenerate.
289 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
291 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
294 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
296 * config/rs6000/rs6000.c (rs6000_option_override_internal):
297 Display warning message for -mno-speculate-indirect-jumps.
299 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
301 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
303 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
305 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
307 PR tree-optimization/84225
308 * tree-eh.c (find_trapping_overflow): Only call
309 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
311 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
314 * config/i386/i386.c: Reimplement the check of possible options
315 -mibt/-mshstk conbination. Change error messages.
316 * doc/invoke.texi: Fix a typo: remove extra '='.
318 2018-02-06 Marek Polacek <polacek@redhat.com>
320 PR tree-optimization/84228
321 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
323 2018-02-06 Tamar Christina <tamar.christina@arm.com>
326 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
327 emitted arch directives.
328 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
329 __ARM_FEATURE_COPROC before changing architectures.
331 2018-02-06 Richard Biener <rguenther@suse.de>
333 * config/i386/i386.c (print_reg): Fix typo.
334 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
336 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
338 * configure: Regenerate.
340 2018-02-05 Martin Sebor <msebor@redhat.com>
342 PR tree-optimization/83369
343 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
346 2018-02-05 Martin Liska <mliska@suse.cz>
348 * doc/invoke.texi: Cherry-pick upstream r323995.
350 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
352 * ira.c (ira_init_register_move_cost): Adjust comment.
354 2018-02-05 Martin Liska <mliska@suse.cz>
356 PR gcov-profile/84137
357 * doc/gcov.texi: Fix typo in documentation.
359 2018-02-05 Martin Liska <mliska@suse.cz>
361 PR gcov-profile/83879
362 * doc/gcov.texi: Document necessity of --dynamic-list-data when
363 using dlopen functionality.
365 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
367 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
368 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
369 _mm_maskz_range_ss, _mm_mask_range_round_ss,
370 _mm_maskz_range_round_ss): New intrinsics.
371 (__builtin_ia32_rangesd128_round)
372 (__builtin_ia32_rangess128_round): Remove.
373 (__builtin_ia32_rangesd128_mask_round,
374 __builtin_ia32_rangess128_mask_round): New builtins.
375 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
376 __builtin_ia32_rangess128_round): Remove.
377 (__builtin_ia32_rangesd128_mask_round,
378 __builtin_ia32_rangess128_mask_round): New builtins.
379 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
380 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
381 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
382 "<round_saeonly_constraint>")): Changed to ...
383 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
384 "<round_saeonly_scalar_constraint>")): ... this.
385 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
386 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
387 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
388 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
389 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
391 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
393 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
395 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
396 Remove all values except native, 8540 and 8548.
398 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
400 * config/i386/i386.c (ix86_output_function_return): Pass
401 INVALID_REGNUM, instead of -1, as invalid register number to
402 indirect_thunk_name and output_indirect_thunk.
404 2018-02-02 Julia Koval <julia.koval@intel.com>
406 * config.gcc: Add -march=icelake.
407 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
408 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
409 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
410 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
411 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
412 (processor_target_table): Add icelake.
413 (ix86_option_override_internal): Handle new PTAs.
414 (get_builtin_code_for_version): Handle icelake.
415 (M_INTEL_COREI7_ICELAKE): New.
416 (fold_builtin_cpu): Handle icelake.
417 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
418 * doc/invoke.texi: Add -march=icelake.
420 2018-02-02 Julia Koval <julia.koval@intel.com>
422 * config/i386/i386.c (ix86_option_override_internal): Change flags type
424 * wide-int-bitmask.h: New.
426 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
429 * config/i386/i386.md: Replace Pmode with word_mode in
430 builtin_setjmp_setup and builtin_longjmp to support x32.
432 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
436 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
438 (rs6000_supported_cpu_names): New static variable.
439 (linux_cpu_translation_table): Likewise.
440 (elf_platform) <cpu>: Define new static variable and use it.
441 Translate kernel AT_PLATFORM name to canonical name if needed.
442 Error if platform name is unknown.
444 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
447 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
449 2018-02-01 Jeff Law <law@redhat.com>
452 * config/i386/i386.c (release_scratch_register_on_entry): Add new
453 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
454 the scratch if RELEASE_VIA_POP is false.
455 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
456 If we have to save a temporary register, decrement SIZE appropriately.
457 Pass new arguments to release_scratch_register_on_entry.
458 (ix86_adjust_stack_and_probe): Likewise.
459 (ix86_emit_probe_stack_range): Pass new arguments to
460 release_scratch_register_on_entry.
462 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
464 PR rtl-optimization/84157
465 * combine.c (change_zero_ext): Use REG_P predicate in
466 front of HARD_REGISTER_P predicate.
468 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
470 * config/avr/avr.c (avr_option_override): Move disabling of
471 -fdelete-null-pointer-checks to...
472 * common/config/avr/avr-common.c (avr_option_optimization_table):
475 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
477 PR tree-optimization/81635
478 * tree-data-ref.c (split_constant_offset_1): For types that
479 wrap on overflow, try to use range info to prove that wrapping
482 2018-02-01 Renlin Li <renlin.li@arm.com>
485 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
487 (aarch64_register_move_cost): Likewise.
488 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
490 (REG_CLASS_NAMES): Likewise.
491 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
492 TAILCALL_ADDR_REGS. Remove IP registers.
493 * config/aarch64/aarch64.md (Ucs): Update register constraint.
495 2018-02-01 Richard Biener <rguenther@suse.de>
497 * domwalk.h (dom_walker::dom_walker): Add additional constructor
498 for specifying RPO order and allow NULL for that.
499 * domwalk.c (dom_walker::dom_walker): Likewise.
500 (dom_walker::walk): Handle NULL RPO order.
501 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
503 (rewrite_update_dom_walker): Likewise.
504 (mark_def_dom_walker): Likewise.
506 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
508 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
509 (aarch64_maybe_expand_sve_subreg_move): Declare.
510 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
511 * config/aarch64/predicates.md (aarch64_any_register_operand): New
513 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
514 that are semantically a reverse operation.
515 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
516 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
517 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
519 (aarch64_can_change_mode_class): For big-endian, forbid changes
520 between two SVE modes if they have different element sizes.
522 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
524 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
525 the TImode handling for big-endian targets.
527 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
529 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
530 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
532 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
533 Remove BSWAP handing for big-endian targets and use the form of
534 LD1RQ appropariate for the mode.
536 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
538 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
539 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
542 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
545 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
546 check for operands that need to go through aarch64_sve_reload_be.
548 2018-02-01 Jakub Jelinek <jakub@redhat.com>
550 PR tree-optimization/81661
551 PR tree-optimization/84117
552 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
553 * tree-eh.c: Include gimplify.h.
554 (find_trapping_overflow, replace_trapping_overflow,
555 rewrite_to_non_trapping_overflow): New functions.
556 * tree-vect-loop.c: Include tree-eh.h.
557 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
558 * tree-data-ref.c: Include tree-eh.h.
559 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
561 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
563 PR rtl-optimization/84123
564 * combine.c (change_zero_ext): Check if hard register satisfies
565 can_change_dest_mode before calling gen_lowpart_SUBREG.
567 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
570 * ira.c (ira_init_register_move_cost): Remove assert.
572 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
574 PR rtl-optimization/84071
575 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
576 * doc/tm.texi: Regenerate.
578 2018-01-31 Richard Biener <rguenther@suse.de>
580 PR tree-optimization/84132
581 * tree-data-ref.c (analyze_miv_subscript): Properly
582 check whether evolution_function_is_affine_multivariate_p
583 before calling gcd_of_steps_may_divide_p.
585 2018-01-31 Julia Koval <julia.koval@intel.com>
588 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
589 * config/i386/i386.md (rdpid_rex64) New.
590 (rdpid): Make 32bit only.
592 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
595 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
596 an IDENTIFIER_NODE for FUNCTION_TYPE's.
598 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
601 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
603 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
605 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
607 PR rtl-optimization/84071
608 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
609 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
611 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
613 * config/arc/arc.c (arc_handle_aux_attribute): New function.
614 (arc_attribute_table): Add 'aux' attribute.
615 (arc_in_small_data_p): Consider aux like variables.
616 (arc_is_aux_reg_p): New function.
617 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
618 (arc_get_aux_arg): New function.
619 (prepare_move_operands): Handle aux-register access.
620 (arc_handle_aux_attribute): New function.
621 * doc/extend.texi (ARC Variable attributes): Add subsection.
623 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
625 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
626 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
627 (arc_attribute_table): Add 'uncached' attribute.
628 (arc_print_operand): Print '.di' flag for uncached memory
630 (arc_in_small_data_p): Do not consider for small data the uncached
632 (arc_is_uncached_mem_p): New function.
633 * config/arc/predicates.md (compact_store_memory_operand): Check
634 for uncached memory accesses.
635 (nonvol_nonimm_operand): Likewise.
636 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
638 2018-01-31 Jakub Jelinek <jakub@redhat.com>
641 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
642 falign-loops=): Add Optimization flag.
644 2018-01-30 Jeff Law <law@redhat.com>
647 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
648 INT_REGISTERS_SAVED. Check it prior to calling
649 get_scratch_register_on_entry.
650 (ix86_adjust_stack_and_probe): Similarly.
651 (ix86_emit_probe_stack_range): Similarly.
652 (ix86_expand_prologue): Corresponding changes.
654 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
657 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
658 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
660 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
663 * lra-constraints.c (curr_insn_transform): Process AND in the
666 2018-01-30 Jakub Jelinek <jakub@redhat.com>
668 PR rtl-optimization/83986
669 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
670 dependence against last_pending_memory_flush in addition to
673 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
675 PR tree-optimization/81611
676 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
679 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
681 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
684 2018-01-30 Richard Biener <rguenther@suse.de>
685 Jakub Jelinek <jakub@redhat.com>
687 PR tree-optimization/84111
688 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
689 inner loops added during recursion, as they don't have up-to-date
692 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
695 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
696 (can_inline_edge_by_limits_p): ... here.
697 (can_early_inline_edge_p, check_callers,
698 update_caller_keys, update_callee_keys, recursive_inlining,
699 add_new_edges_to_heap, speculation_useful_p,
700 inline_small_functions,
701 inline_small_functions, flatten_function,
702 inline_to_all_callers_1): Update.
704 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
706 * profile-count.c (profile_count::combine_with_ipa_count): Handle
709 2018-01-30 Richard Biener <rguenther@suse.de>
711 PR tree-optimization/83008
712 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
713 invariant and constant vector uses in stmts when they need
716 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
719 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
720 * configure: Regenerate.
722 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
724 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
726 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
727 Use gen_rtx_REG rather than gen_lowpart.
729 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
731 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
732 rather than 0 when creating partial subregs.
734 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
736 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
739 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
742 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
743 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
744 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
745 flags. This restores the settings used before the 2017-07-24.
746 Turning off pre increment/decrement/modify allows IVOPTS to
747 optimize DF/SF loops where the index is an int.
749 2018-01-29 Richard Biener <rguenther@suse.de>
750 Kelvin Nilsen <kelvin@gcc.gnu.org>
753 * tree-vect-stmts.c (vectorizable_call): Don't call
754 targetm.vectorize_builtin_md_vectorized_function if callee is
757 2018-01-22 Carl Love <cel@us.ibm.com>
759 * doc/extend.tex: Fix typo in second arg in
760 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
762 2018-01-29 Richard Biener <rguenther@suse.de>
764 PR tree-optimization/84086
765 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
766 (flush_ssaname_freelist): When SSA names were released reset
769 2018-01-29 Richard Biener <rguenther@suse.de>
771 PR tree-optimization/84057
772 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
773 removed paths when removing edges.
775 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
777 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
778 -mfunction-return=@var{choice}.
780 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
783 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
785 (layout::print_source_line): Likewise.
786 (test_get_line_width_without_trailing_whitespace): Add test cases.
788 2018-01-27 Jakub Jelinek <jakub@redhat.com>
791 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
794 2018-01-26 Jim Wilson <jimw@sifive.com>
796 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
798 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
801 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
803 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
804 and CMP + SUB-immediate -> SUBS.
806 2018-01-26 Martin Sebor <msebor@redhat.com>
808 PR tree-optimization/83896
809 * tree-ssa-strlen.c (get_string_len): Rename...
810 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
811 Avoid assuming length is constant.
812 (handle_char_store): Use HOST_WIDE_INT for string length.
814 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
817 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
818 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
820 2018-01-26 Richard Biener <rguenther@suse.de>
822 PR rtl-optimization/84003
823 * dse.c (record_store): Only record redundant stores when
824 the earlier store aliases at least all accesses the later one does.
826 2018-01-26 Jakub Jelinek <jakub@redhat.com>
828 PR rtl-optimization/83985
829 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
830 REG_CFA_RESTORE insns.
831 (delete_unmarked_insns): Don't ignore separate shrink wrapping
832 REG_CFA_RESTORE insns here.
835 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
836 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
838 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
840 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
841 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
842 (arc_init): Likewise.
843 (arc_override_options): Likewise.
844 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
846 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
848 * config/arc/arc.h (TARGET_DBNZ): Define.
849 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
850 properly set the tune attribute.
851 (dbnz): Use TARGET_DBNZ guard.
852 * config/arc/arc.opt (mtune): Add core3 option.
854 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
856 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
857 recognize new pic like addresses.
858 (arc_delegitimize_address): Clean up.
860 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
862 * config/arc/arc-arches.def: Option mrf16 valid for all
864 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
865 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
866 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
867 * config/arc/arc-tables.opt: Regenerate.
868 * config/arc/arc.c (arc_conditional_register_usage): Handle
869 reduced register file case.
870 (arc_file_start): Set must have build attributes.
871 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
873 * config/arc/arc.opt (mrf16): Add new option.
874 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
875 * config/arc/genmultilib.awk: Handle new mrf16 option.
876 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
877 * config/arc/t-multilib: Regenerate.
878 * doc/invoke.texi (ARC Options): Document mrf16 option.
880 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
882 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
883 * config/arc/arc.c (arc_handle_secure_attribute): New function.
884 (arc_attribute_table): Add 'secure_call' attribute.
885 (arc_print_operand): Print secure call operand.
886 (arc_function_ok_for_sibcall): Don't optimize tail calls when
888 (arc_is_secure_call_p): New function. * config/arc/arc.md
889 (call_i): Add support for sjli instruction.
890 (call_value_i): Likewise.
891 * config/arc/constraints.md (Csc): New constraint.
893 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
894 John Eric Martin <John.Martin@emmicro-us.com>
896 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
897 * config/arc/arc.c (_arc_jli_section): New struct.
898 (arc_jli_section): New type.
899 (rc_jli_sections): New static variable.
900 (arc_handle_jli_attribute): New function.
901 (arc_attribute_table): Add jli_always and jli_fixed attribute.
902 (arc_file_end): New function.
903 (TARGET_ASM_FILE_END): Define.
904 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
905 (arc_add_jli_section): New function.
906 (jli_call_scan): Likewise.
907 (arc_reorg): Call jli_call_scan.
908 (arc_output_addsi): Remove 'S' from printing asm operand.
909 (arc_is_jli_call_p): New function.
910 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
912 (movhi_insn): Likewise.
913 (movsi_insn): Likewise.
914 (movsi_set_cc_insn): Likewise.
915 (loadqi_update): Likewise.
916 (load_zeroextendqisi_update): Likewise.
917 (load_signextendqisi_update): Likewise.
918 (loadhi_update): Likewise.
919 (load_zeroextendhisi_update): Likewise.
920 (load_signextendhisi_update): Likewise.
921 (loadsi_update): Likewise.
922 (loadsf_update): Likewise.
923 (movsicc_insn): Likewise.
924 (bset_insn): Likewise.
925 (bxor_insn): Likewise.
926 (bclr_insn): Likewise.
927 (bmsk_insn): Likewise.
928 (bicsi3_insn): Likewise.
929 (cmpsi_cc_c_insn): Likewise.
930 (movsi_ne): Likewise.
931 (movsi_cond_exec): Likewise.
932 (clrsbsi2): Likewise.
943 (call_i): Remove 'S' asm letter, add jli instruction.
944 (call_value_i): Likewise.
945 * config/arc/arc.op (mjli-always): New option.
946 * config/arc/constraints.md (Cji): New constraint.
947 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
949 (subsf3_fpx): Likewise.
950 (mulsf3_fpx): Likewise.
951 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
953 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
955 * doc/invoke.texi (ARC): Document mjli-always option.
957 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
959 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
960 and use incw and decw where possible
961 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
963 2018-01-26 Richard Biener <rguenther@suse.de>
965 PR tree-optimization/81082
966 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
967 association if it requires casting to unsigned.
968 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
969 from fold_plusminus_mult_expr to catch important cases late when
970 range info is available.
972 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
974 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
975 * configure.ac (hidden_linkonce): New test.
976 * configure: Regenerate.
977 * config.in: Regenerate.
979 2018-01-26 Julia Koval <julia.koval@intel.com>
981 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
982 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
983 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
984 _mm_mask_bitshuffle_epi64_mask): Fix type.
985 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
986 USI_FTYPE_V4DI_V4DI_USI): Remove.
987 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
988 __builtin_ia32_vpshufbitqmb256_mask,
989 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
990 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
991 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
993 2018-01-26 Alan Modra <amodra@gmail.com>
996 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
997 UNSPEC_VBPERMQ. Sort other unspecs.
999 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1001 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1003 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1006 * predict.c (drop_profile): Do not push/pop cfun; update also
1008 (handle_missing_profiles): Fix logic looking for zero profiles.
1010 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1013 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1014 on functions with #pragma omp declare simd or functions with simd
1016 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1017 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1018 Remove trailing \n from warning_at calls.
1020 2018-01-25 Tom de Vries <tom@codesourcery.com>
1023 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1024 for neutered workers.
1026 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1029 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1030 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1032 2018-01-24 Jeff Law <law@redhat.com>
1035 * i386.c (get_probe_interval): Move to earlier point.
1036 (ix86_compute_frame_layout): If -fstack-clash-protection and
1037 the frame is larger than the probe interval, then use pushes
1038 to save registers rather than reg->mem moves.
1039 (ix86_expand_prologue): Remove conditional for int_registers_saved
1042 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1045 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1046 min/max for never referenced object.
1048 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1051 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1053 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1054 attributes from DECL_ATTRIBUTES (decl) without affecting
1055 DECL_ATTRIBUTES (current_function_decl).
1056 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1057 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1059 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1061 PR tree-optimization/83979
1062 * fold-const.c (fold_comparison): Use constant_boolean_node
1063 instead of boolean_{true,false}_node.
1065 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1067 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1070 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1072 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1073 Simplify the clause that sets the length attribute.
1074 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1075 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1076 clause that sets the length attribute.
1077 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1079 2018-01-24 Tom de Vries <tom@codesourcery.com>
1082 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1083 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1084 Add strict parameter.
1085 (prevent_branch_around_nothing): Insert dummy insn between branch to
1086 label and label with no ptx insn inbetween.
1087 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1089 2018-01-24 Tom de Vries <tom@codesourcery.com>
1092 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1093 for neutered threads in warp.
1094 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1096 2018-01-24 Richard Biener <rguenther@suse.de>
1098 PR tree-optimization/83176
1099 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1102 2018-01-24 Richard Biener <rguenther@suse.de>
1104 PR tree-optimization/82819
1105 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1106 code generating pluses that are no-ops in the target precision.
1108 2018-01-24 Richard Biener <rguenther@suse.de>
1111 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1113 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1115 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1116 to merge probabilities.
1117 * predict.c (probably_never_executed): Also mark as cold functions
1118 with global 0 profile and guessed local profile.
1119 * profile-count.c (profile_probability::combine_with_count): New
1121 * profile-count.h (profile_probability::operator*,
1122 profile_probability::operator*=, profile_probability::operator/,
1123 profile_probability::operator/=): Reduce precision to adjusted
1124 and set value to guessed on contradictory divisions.
1125 (profile_probability::combine_with_freq): Remove.
1126 (profile_probability::combine_wiht_count): Declare.
1127 (profile_count::force_nonzero):: Set to adjusted.
1128 (profile_count::probability_in):: Set quality to adjusted.
1129 * tree-ssa-tail-merge.c (replace_block_by): Use
1132 2018-01-23 Andrew Waterman <andrew@sifive.com>
1133 Jim Wilson <jimw@sifive.com>
1135 * config/riscv/riscv.c (riscv_stack_boundary): New.
1136 (riscv_option_override): Set riscv_stack_boundary. Handle
1137 riscv_preferred_stack_boundary_arg.
1138 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1139 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1140 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1141 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1142 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1143 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1145 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1148 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1149 of struct ix86_frame.
1150 (ix86_expand_epilogue): Likewise. Add a local variable for
1151 the reg_save_offset field in struct ix86_frame.
1153 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1155 PR tree-optimization/82604
1156 * tree-loop-distribution.c (enum partition_kind): New enum item
1157 PKIND_PARTIAL_MEMSET.
1158 (partition_builtin_p): Support above new enum item.
1159 (generate_code_for_partition): Ditto.
1160 (compute_access_range): Differentiate cases that equality can be
1161 proven at all loops, the innermost loops or no loops.
1162 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1163 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1164 (finalize_partitions, distribute_loop): Don't fuse partition of
1165 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1166 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1169 2018-01-23 Martin Liska <mliska@suse.cz>
1171 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1172 order to ignore the predictor.
1173 (PRED_POLYMORPHIC_CALL): Likewise.
1174 (PRED_RECURSIVE_CALL): Likewise.
1176 2018-01-23 Martin Liska <mliska@suse.cz>
1178 * tree-profile.c (tree_profiling): Print function header to
1179 aware reader which function we are working on.
1180 * value-prof.c (gimple_find_values_to_profile): Do not print
1181 not interesting value histograms.
1183 2018-01-23 Martin Liska <mliska@suse.cz>
1185 * profile-count.h (enum profile_quality): Add
1186 profile_uninitialized as the first value. Do not number values
1187 as they are zero based.
1188 (profile_count::verify): Update sanity check.
1189 (profile_probability::verify): Likewise.
1191 2018-01-23 Nathan Sidwell <nathan@acm.org>
1193 * doc/invoke.texi (ffor-scope): Deprecate.
1195 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1197 PR tree-optimization/83510
1198 * domwalk.c (set_all_edges_as_executable): New function.
1199 (dom_walker::dom_walker): Convert bool param
1200 "skip_unreachable_blocks" to enum reachability. Move setup of
1201 edge flags to set_all_edges_as_executable and only do it when
1202 reachability is REACHABLE_BLOCKS.
1203 * domwalk.h (enum dom_walker::reachability): New enum.
1204 (dom_walker::dom_walker): Convert bool param
1205 "skip_unreachable_blocks" to enum reachability.
1206 (set_all_edges_as_executable): New decl.
1207 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1208 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1210 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1211 but converting true to REACHABLE_BLOCKS.
1212 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1214 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1215 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1216 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1218 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1219 if check_all_array_refs will be called.
1221 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1223 * tree.c (selftest::test_location_wrappers): Add more test
1226 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1228 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1229 (selftest::test_bit_in_range): Likewise.
1231 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1234 * doc/sourcebuild.texi (vect_float): Say that the selector
1235 only describes the situation when -funsafe-math-optimizations is on.
1236 (vect_float_strict): Document.
1238 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1240 PR tree-optimization/83965
1241 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1242 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1243 instead of checking only for a reduction.
1244 (vect_recog_widen_sum_pattern): Likewise.
1246 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1248 * predict.c (probably_never_executed): Only use precise profile info.
1249 (compute_function_frequency): Skip after inlining hack since we now
1250 have quality checking.
1252 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1254 * profile-count.h (profile_probability::very_unlikely,
1255 profile_probability::unlikely, profile_probability::even): Set
1256 precision to guessed.
1258 2018-01-23 Richard Biener <rguenther@suse.de>
1260 PR tree-optimization/83963
1261 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1262 Properly terminate dominator walk when crossing the exit edge not
1263 when visiting its source block.
1265 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1268 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1269 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1271 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1273 PR tree-optimization/83957
1274 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1275 semicolon after for body surrounded by braces.
1277 PR tree-optimization/83081
1278 * profile-count.h (profile_probability::split): New method.
1279 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1280 Use profile_probability::split.
1281 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1282 when splitting a single conditional jump into 2.
1284 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1286 PR tree-optimization/69452
1287 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1290 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1292 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1293 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1294 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1296 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1298 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1299 * config/rl78/rl78.md: New define_expand "movdi"
1300 * config/rl78/rl78.c: New function definition rl78_split_movdi
1302 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1305 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1307 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1308 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1309 128-bit to produce an UNSPEC move to get the double word with the
1310 signbit and then a shift directly to do signbit.
1311 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1312 implementation with a new version that just does either a direct
1313 move or a regular move. Move memory interface to separate insns.
1314 Move insns so they are next to the expander.
1315 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1316 with signbit move. Split big and little endian case.
1317 (signbit<mode>2_dm_mem_le): Likewise.
1318 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1319 (signbit<mode>2_dm2): Likewise.
1321 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1323 * config/rl78/rl78.md: New define_expand "anddi3".
1325 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1327 * config/rl78/rl78.md: New define_expand "umindi3".
1329 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1331 * config/rl78/rl78.md: New define_expand "smindi3".
1333 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1335 * config/rl78/rl78.md: New define_expand "smaxdi3".
1337 2018-01-22 Carl Love <cel@us.ibm.com>
1339 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1340 LVX_V1TI): Add macro expansion.
1341 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1342 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1343 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1344 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1345 Change check to determine if the instruction is a byte reversing
1346 entry. Fix typo in comment.
1347 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1348 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1349 Add def_builtin calls for new builtins.
1350 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1351 Add define_insn expansion.
1353 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1355 * config/rl78/rl78.md: New define_expand "umaxdi3".
1357 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1359 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1360 for non-QImode registers
1362 2018-01-22 Richard Biener <rguenther@suse.de>
1364 PR tree-optimization/83963
1365 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1366 including the loop exit block.
1367 (scop_detection::merge_sese): Likewise.
1368 (scop_detection::add_scop): Do it here instead.
1370 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1372 * doc/sourcebuild.texi (arm_softfloat): Document.
1374 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1377 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1378 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1379 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1381 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1382 David Edelsohn <dje.gcc@gmail.com>
1385 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1386 Change "crset eq" to "crset 2".
1387 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1388 (*call_indirect_aix<mode>_nospec): Likewise.
1389 (*call_value_indirect_aix<mode>_nospec): Likewise.
1390 (*call_indirect_elfv2<mode>_nospec): Likewise.
1391 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1392 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1393 change assembly output from . to $.
1394 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1395 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1396 (*tablejump<mode>_internal1_nospec): Likewise.
1398 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1401 * config/sh/sh_optimize_sett_clrt.cc:
1402 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1404 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1406 PR tree-optimization/83940
1407 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1408 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1409 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1410 use it to pass back the definition type.
1411 (vect_check_store_rhs): Likewise rhs_dt_out.
1412 (vect_build_gather_load_calls): Add a mask_dt argument and use
1413 it instead of a call to vect_is_simple_use.
1414 (vectorizable_store): Update calls to vect_check_load_store_mask
1415 and vect_check_store_rhs. Use the dt returned by the latter instead
1416 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1417 instead of calls to vect_is_simple_use. Pass the scalar rather
1418 than the vector operand to vect_is_simple_use when handling
1419 second and subsequent copies of an rhs value.
1420 (vectorizable_load): Update calls to vect_check_load_store_mask
1421 and vect_build_gather_load_calls. Use the cached mask_dt and
1422 gs_info.offset_dt instead of calls to vect_is_simple_use.
1424 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1427 * tree-emutls.c: Include gimplify.h.
1428 (lower_emutls_2): New function.
1429 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1430 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1431 it before further processing.
1434 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1435 UINTVAL (trueop1) instead of INTVAL (op1).
1437 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1441 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1442 INCOMING_FRAME_SP_OFFSET if not defined.
1443 (scan_trace): Add ENTRY argument. If true and
1444 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1445 emit a note to adjust the CFA offset.
1446 (create_cfi_notes): Adjust scan_trace callers.
1447 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1448 INCOMING_FRAME_SP_OFFSET in the CIE.
1449 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1450 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1452 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1453 * doc/tm.texi: Regenerated.
1455 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1457 PR rtl-optimization/83147
1458 * lra-constraints.c (remove_inheritance_pseudos): Use
1459 lra_substitute_pseudo_within_insn.
1461 2018-01-19 Tom de Vries <tom@codesourcery.com>
1462 Cesar Philippidis <cesar@codesourcery.com>
1465 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1467 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1470 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1471 spaces for function labels.
1473 2018-01-19 Martin Liska <mliska@suse.cz>
1475 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1476 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1477 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1478 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1479 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1480 (PRED_CONST_RETURN): Change from 69 to 65.
1481 (PRED_NULL_RETURN): Change from 91 to 71.
1482 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1483 (PRED_LOOP_GUARD): Change from 66 to 73.
1485 2018-01-19 Martin Liska <mliska@suse.cz>
1487 * predict.c (predict_insn_def): Add new assert.
1488 (struct branch_predictor): Change type to signed integer.
1489 (test_prediction_value_range): Amend test to cover
1491 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1492 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1493 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1494 (PRED_LOOP_IV_COMPARE): Likewise.
1495 * predict.h (PROB_UNINITIALIZED): Define new constant.
1497 2018-01-19 Martin Liska <mliska@suse.cz>
1499 * predict.c (dump_prediction): Add new format for
1500 analyze_brprob.py script which is enabled with -details
1502 * profile-count.h (precise_p): New function.
1504 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1506 PR tree-optimization/83922
1507 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1508 there are no statements that need masking.
1509 (vect_active_double_reduction_p): New function.
1510 (vect_analyze_loop_operations): Use it when handling phis that
1511 are not in the loop header.
1513 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1515 PR tree-optimization/83914
1516 * tree-vect-loop.c (vectorizable_induction): Don't convert
1517 init_expr or apply the peeling adjustment for inductions
1518 that are nested within the vectorized loop.
1520 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1522 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1525 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1529 * function.h (gimplify_parameters): Add gimple_seq * argument.
1530 * function.c: Include gimple.h and options.h.
1531 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1532 for the added local temporaries if needed.
1533 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1534 if there are any parameter cleanups, wrap whole body into a
1535 try/finally with the cleanups.
1537 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1540 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1541 Use GET_MODE_CLASS for scalar floating point.
1543 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1547 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1548 Fix call of call_cgraph_insertion_hooks.
1550 2018-01-18 Martin Sebor <msebor@redhat.com>
1552 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1554 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1557 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1560 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1563 * common.opt: (-ffile-prefix-map): New option.
1564 * opts.c (common_handle_option): Defer it.
1565 * opts-global.c (handle_common_deferred_options): Handle it.
1566 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1567 * file-prefix-map.h: New file.
1568 (remap_debug_filename, add_debug_prefix_map): ...here.
1569 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1570 * final.c (debug_prefix_map, add_debug_prefix_map
1571 remap_debug_filename): Move to...
1572 * file-prefix-map.c: New file.
1573 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1574 generalize, get rid of alloca(), use strrchr() instead of strchr().
1575 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1576 Implement in terms of add_prefix_map().
1577 (remap_macro_filename, remap_debug_filename): Implement in term of
1579 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1580 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1581 * dbxout.c: Include file-prefix-map.h.
1582 * varasm.c: Likewise.
1583 * vmsdbgout.c: Likewise.
1584 * xcoffout.c: Likewise.
1585 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1586 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1587 * doc/invoke.texi (-ffile-prefix-map): Document.
1588 (-fdebug-prefix-map): Update description.
1590 2018-01-18 Martin Liska <mliska@suse.cz>
1592 * config/i386/i386.c (indirect_thunk_name): Document that also
1594 (output_indirect_thunk): Document why both instructions
1595 (pause and lfence) are generated.
1597 2018-01-18 Richard Biener <rguenther@suse.de>
1599 PR tree-optimization/83887
1600 * graphite-scop-detection.c
1601 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1602 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1603 (scop_detection::merge_sese): Re-implement with a flood-fill
1604 algorithm that properly finds a SESE region if it exists.
1606 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1609 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1610 pointer_diff optimizations use view_convert instead of convert.
1612 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1614 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1615 Generate different code for -mno-speculate-indirect-jumps.
1616 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1617 (*call_indirect_aix<mode>): Disable for
1618 -mno-speculate-indirect-jumps.
1619 (*call_indirect_aix<mode>_nospec): New define_insn.
1620 (*call_value_indirect_aix<mode>): Disable for
1621 -mno-speculate-indirect-jumps.
1622 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1623 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1624 -mno-speculate-indirect-jumps.
1625 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1627 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1629 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1630 long double type, set the flags for noting the default long double
1631 type, even if we don't pass or return a long double type.
1633 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1636 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1639 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1641 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1642 support for merge[hl].
1643 (fold_mergehl_helper): New helper function.
1644 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1645 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1646 (altivec_vmrglw_direct): Add xxmrglw insn.
1648 2018-01-17 Andrew Waterman <andrew@sifive.com>
1650 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1651 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1653 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1656 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1657 call the lto_location_cache before reading the
1658 DECL_SOURCE_LOCATION of the types.
1660 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1661 Richard Sandiford <richard.sandiford@linaro.org>
1663 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1664 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1665 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1666 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1667 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1669 * config/aarch64/constraints.md (aarch64_movti_operand):
1671 * config/aarch64/predicates.md (Uti): Add new constraint.
1673 2018-01-17 Carl Love <cel@us.ibm.com>
1674 * config/rs6000/vsx.md (define_expand xl_len_r,
1675 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1676 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1678 (define_expand, define_insn): Move the shift left from the
1679 define_insn to the define_expand for lxvl and stxvl instructions.
1680 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1681 and XL_LEN_R definitions to PURE.
1683 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1685 * config/i386/i386.c (indirect_thunk_name): Declare regno
1686 as unsigned int. Compare regno with INVALID_REGNUM.
1687 (output_indirect_thunk): Ditto.
1688 (output_indirect_thunk_function): Ditto.
1689 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1690 in the call to output_indirect_thunk_function.
1692 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1695 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1696 rather than the size of inner_type to determine the stack slot size
1697 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1699 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1702 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1705 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1707 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1708 endian Linux systems to optionally enable multilibs for selecting
1709 the long double type if the user configured an explicit type.
1710 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1711 have no long double multilibs if not defined.
1712 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1713 warn if the user used -mabi={ieee,ibm}longdouble and we built
1714 multilibs for long double.
1715 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1716 appropriate multilib option.
1717 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1719 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1720 for building long double multilibs.
1721 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1723 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1725 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1728 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1730 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1733 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1736 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1739 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1741 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1742 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1744 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1746 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1747 different rtl trees depending on TARGET_64BIT.
1748 (rs6000_gen_lvx): Likewise.
1750 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1752 * config/visium/visium.md (nop): Tweak comment.
1753 (hazard_nop): Likewise.
1755 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1757 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1758 -mspeculate-indirect-jumps.
1759 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1760 for -mno-speculate-indirect-jumps.
1761 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1762 (*call_value_indirect_elfv2<mode>): Disable for
1763 -mno-speculate-indirect-jumps.
1764 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1765 (indirect_jump): Emit different RTL for
1766 -mno-speculate-indirect-jumps.
1767 (*indirect_jump<mode>): Disable for
1768 -mno-speculate-indirect-jumps.
1769 (*indirect_jump<mode>_nospec): New define_insn.
1770 (tablejump): Emit different RTL for
1771 -mno-speculate-indirect-jumps.
1772 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1773 (tablejumpsi_nospec): New define_expand.
1774 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1775 (tablejumpdi_nospec): New define_expand.
1776 (*tablejump<mode>_internal1): Disable for
1777 -mno-speculate-indirect-jumps.
1778 (*tablejump<mode>_internal1_nospec): New define_insn.
1779 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1782 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1784 * caller-save.c (insert_save): Drop unnecessary parameter. All
1787 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1788 Richard Biener <rguenth@suse.de>
1791 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1792 return early, inline manually is_gimple_sizepos. Make sure if we
1793 call gimplify_expr we don't end up with a gimple constant.
1794 * tree.c (variably_modified_type_p): Don't return true for
1795 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1796 * gimplify.h (is_gimple_sizepos): Remove.
1798 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1800 PR tree-optimization/83857
1801 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1802 vectorizable_live_operation for pure SLP statements.
1803 (vectorizable_live_operation): Handle PHIs.
1805 2018-01-16 Richard Biener <rguenther@suse.de>
1807 PR tree-optimization/83867
1808 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1809 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1811 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1814 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1815 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1816 If off is not INTEGER_CST, issue a may not be aligned warning
1817 rather than isn't aligned. Use isn%'t rather than isn't.
1818 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1820 <case MULT_EXPR>: Improve the case when bottom and one of the
1821 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1822 operand, in that case check if the other operand is multiple of
1823 bottom divided by the INTEGER_CST operand.
1825 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1828 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1829 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1830 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1831 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1832 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1833 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1834 (pa_function_arg_size): New function.
1836 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1838 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1839 in a separate statement.
1841 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1843 PR tree-optimization/83847
1844 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1845 group gathers and scatters.
1847 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1849 PR rtl-optimization/86620
1850 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1852 PR rtl-optimization/83213
1853 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1854 to last if both are JUMP_INSNs.
1856 PR tree-optimization/83843
1857 * gimple-ssa-store-merging.c
1858 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1859 store_immediate_info for bswap/nop orig_stores.
1861 2018-01-15 Andrew Waterman <andrew@sifive.com>
1863 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1865 <UDIV>: Increase cost if !TARGET_DIV.
1867 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1869 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1870 (define_attr "cr_logical_3op"): New.
1871 (cceq_ior_compare): Adjust.
1872 (cceq_ior_compare_complement): Adjust.
1873 (*cceq_rev_compare): Adjust.
1874 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1875 (is_cracked_insn): Adjust.
1876 (insn_must_be_first_in_group): Adjust.
1877 * config/rs6000/40x.md: Adjust.
1878 * config/rs6000/440.md: Adjust.
1879 * config/rs6000/476.md: Adjust.
1880 * config/rs6000/601.md: Adjust.
1881 * config/rs6000/603.md: Adjust.
1882 * config/rs6000/6xx.md: Adjust.
1883 * config/rs6000/7450.md: Adjust.
1884 * config/rs6000/7xx.md: Adjust.
1885 * config/rs6000/8540.md: Adjust.
1886 * config/rs6000/cell.md: Adjust.
1887 * config/rs6000/e300c2c3.md: Adjust.
1888 * config/rs6000/e500mc.md: Adjust.
1889 * config/rs6000/e500mc64.md: Adjust.
1890 * config/rs6000/e5500.md: Adjust.
1891 * config/rs6000/e6500.md: Adjust.
1892 * config/rs6000/mpc.md: Adjust.
1893 * config/rs6000/power4.md: Adjust.
1894 * config/rs6000/power5.md: Adjust.
1895 * config/rs6000/power6.md: Adjust.
1896 * config/rs6000/power7.md: Adjust.
1897 * config/rs6000/power8.md: Adjust.
1898 * config/rs6000/power9.md: Adjust.
1899 * config/rs6000/rs64.md: Adjust.
1900 * config/rs6000/titan.md: Adjust.
1902 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1904 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1905 ix86_indirect_branch_register logic.
1907 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1909 * config/i386/constraints.md (Bs): Update
1910 ix86_indirect_branch_register check. Don't check
1911 ix86_indirect_branch_register with GOT_memory_operand.
1913 * config/i386/predicates.md (GOT_memory_operand): Don't check
1914 ix86_indirect_branch_register here.
1915 (GOT32_symbol_operand): Likewise.
1917 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1919 * config/i386/predicates.md (constant_call_address_operand):
1920 Rewrite ix86_indirect_branch_register logic.
1921 (sibcall_insn_operand): Likewise.
1923 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1925 * config/i386/constraints.md (Bs): Replace
1926 ix86_indirect_branch_thunk_register with
1927 ix86_indirect_branch_register.
1929 * config/i386/i386.md (indirect_jump): Likewise.
1930 (tablejump): Likewise.
1931 (*sibcall_memory): Likewise.
1932 (*sibcall_value_memory): Likewise.
1933 Peepholes of indirect call and jump via memory: Likewise.
1934 * config/i386/i386.opt: Likewise.
1935 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1936 (GOT_memory_operand): Likewise.
1937 (call_insn_operand): Likewise.
1938 (sibcall_insn_operand): Likewise.
1939 (GOT32_symbol_operand): Likewise.
1941 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1944 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1945 type rather than type addr's type points to.
1946 (expand_omp_atomic_mutex): Likewise.
1947 (expand_omp_atomic): Likewise.
1949 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1952 * config/i386/i386.c (output_indirect_thunk_function): Use
1953 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1954 for __x86_return_thunk.
1956 2018-01-15 Richard Biener <rguenther@suse.de>
1959 * expmed.c (extract_bit_field_1): Fix typo.
1961 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1964 * config/arm/iterators.md (VF): New mode iterator.
1965 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1966 Remove integer-related logic from pattern.
1967 (neon_vabd<mode>_3): Likewise.
1969 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1972 * common.opt (fstrict-overflow): No longer an alias.
1973 (fwrapv-pointer): New option.
1974 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1975 also for pointer types based on flag_wrapv_pointer.
1976 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1977 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1978 opts->x_flag_wrapv got set.
1979 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1980 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1981 POINTER_TYPE_OVERFLOW_UNDEFINED.
1982 * match.pd: Likewise in address comparison pattern.
1983 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1985 2018-01-15 Richard Biener <rguenther@suse.de>
1988 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1989 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1990 Reset type names to their identifier if their TYPE_DECL doesn't
1991 have linkage (and thus is used for ODR and devirt).
1992 (save_debug_info_for_decl): Remove.
1993 (save_debug_info_for_type): Likewise.
1994 (add_tree_to_fld_list): Adjust.
1995 * tree-pretty-print.c (dump_generic_node): Make dumping of
1996 type names more robust.
1998 2018-01-15 Richard Biener <rguenther@suse.de>
2000 * BASE-VER: Bump to 8.0.1.
2002 2018-01-14 Martin Sebor <msebor@redhat.com>
2005 * builtins.c (check_access): Avoid warning when the no-warning bit
2008 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2010 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2011 * ira-color (allocno_hard_regs_compare): Likewise.
2013 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2016 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2017 Use .pushsection/.popsection.
2019 2018-01-14 Martin Sebor <msebor@redhat.com>
2022 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2024 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2026 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2027 entry from extra_headers.
2028 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2029 extra_headers, make the list bitwise identical to the i?86-*-* one.
2031 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2033 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2034 -mcmodel=large with -mindirect-branch=thunk,
2035 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2036 -mfunction-return=thunk-extern.
2037 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2038 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2039 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2041 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2043 * config/i386/i386.c (print_reg): Print the name of the full
2044 integer register without '%'.
2045 (ix86_print_operand): Handle 'V'.
2046 * doc/extend.texi: Document 'V' modifier.
2048 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2050 * config/i386/constraints.md (Bs): Disallow memory operand for
2051 -mindirect-branch-register.
2053 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2054 (GOT_memory_operand): Likewise.
2055 (call_insn_operand): Likewise.
2056 (sibcall_insn_operand): Likewise.
2057 (GOT32_symbol_operand): Likewise.
2058 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2059 for -mindirect-branch-register.
2060 (tablejump): Likewise.
2061 (*sibcall_memory): Likewise.
2062 (*sibcall_value_memory): Likewise.
2063 Disallow peepholes of indirect call and jump via memory for
2064 -mindirect-branch-register.
2065 (*call_pop): Replace m with Bw.
2066 (*call_value_pop): Likewise.
2067 (*sibcall_pop_memory): Replace m with Bs.
2068 * config/i386/i386.opt (mindirect-branch-register): New option.
2069 * doc/invoke.texi: Document -mindirect-branch-register option.
2071 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2073 * config/i386/i386-protos.h (ix86_output_function_return): New.
2074 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2075 set function_return_type.
2076 (indirect_thunk_name): Add ret_p to indicate thunk for function
2078 (output_indirect_thunk_function): Pass false to
2079 indirect_thunk_name.
2080 (ix86_output_indirect_branch_via_reg): Likewise.
2081 (ix86_output_indirect_branch_via_push): Likewise.
2082 (output_indirect_thunk_function): Create alias for function
2083 return thunk if regno < 0.
2084 (ix86_output_function_return): New function.
2085 (ix86_handle_fndecl_attribute): Handle function_return.
2086 (ix86_attribute_table): Add function_return.
2087 * config/i386/i386.h (machine_function): Add
2088 function_return_type.
2089 * config/i386/i386.md (simple_return_internal): Use
2090 ix86_output_function_return.
2091 (simple_return_internal_long): Likewise.
2092 * config/i386/i386.opt (mfunction-return=): New option.
2093 (indirect_branch): Mention -mfunction-return=.
2094 * doc/extend.texi: Document function_return function attribute.
2095 * doc/invoke.texi: Document -mfunction-return= option.
2097 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2099 * config/i386/i386-opts.h (indirect_branch): New.
2100 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2101 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2102 with local indirect jump when converting indirect call and jump.
2103 (ix86_set_indirect_branch_type): New.
2104 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2105 (indirectlabelno): New.
2106 (indirect_thunk_needed): Likewise.
2107 (indirect_thunk_bnd_needed): Likewise.
2108 (indirect_thunks_used): Likewise.
2109 (indirect_thunks_bnd_used): Likewise.
2110 (INDIRECT_LABEL): Likewise.
2111 (indirect_thunk_name): Likewise.
2112 (output_indirect_thunk): Likewise.
2113 (output_indirect_thunk_function): Likewise.
2114 (ix86_output_indirect_branch_via_reg): Likewise.
2115 (ix86_output_indirect_branch_via_push): Likewise.
2116 (ix86_output_indirect_branch): Likewise.
2117 (ix86_output_indirect_jmp): Likewise.
2118 (ix86_code_end): Call output_indirect_thunk_function if needed.
2119 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2121 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2122 (ix86_attribute_table): Add indirect_branch.
2123 * config/i386/i386.h (machine_function): Add indirect_branch_type
2124 and has_local_indirect_jump.
2125 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2127 (tablejump): Likewise.
2128 (*indirect_jump): Use ix86_output_indirect_jmp.
2129 (*tablejump_1): Likewise.
2130 (simple_return_indirect_internal): Likewise.
2131 * config/i386/i386.opt (mindirect-branch=): New option.
2132 (indirect_branch): New.
2135 (thunk-inline): Likewise.
2136 (thunk-extern): Likewise.
2137 * doc/extend.texi: Document indirect_branch function attribute.
2138 * doc/invoke.texi: Document -mindirect-branch= option.
2140 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2143 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2145 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2147 * ipa-inline.c (want_inline_small_function_p): Return false if
2148 inlining has already failed with CIF_FINAL_ERROR.
2149 (update_caller_keys): Call want_inline_small_function_p before
2151 (update_callee_keys): Likewise.
2153 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2155 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2157 (rs6000_quadword_masked_address_p): Likewise.
2158 (quad_aligned_load_p): Likewise.
2159 (quad_aligned_store_p): Likewise.
2160 (const_load_sequence_p): Add comment to describe the outer-most loop.
2161 (mimic_memory_attributes_and_flags): New function.
2162 (rs6000_gen_stvx): Likewise.
2163 (replace_swapped_aligned_store): Likewise.
2164 (rs6000_gen_lvx): Likewise.
2165 (replace_swapped_aligned_load): Likewise.
2166 (replace_swapped_load_constant): Capitalize argument name in
2167 comment describing this function.
2168 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2169 and stores that access quad-word aligned addresses and replace
2170 with stvx or lvx instructions when appropriate.
2171 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2172 New function prototype.
2173 (rs6000_quadword_masked_address_p): Likewise.
2174 (rs6000_gen_lvx): Likewise.
2175 (rs6000_gen_stvx): Likewise.
2176 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2177 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2178 when memory address is aligned.
2179 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2180 this split to select lvx instruction when memory address is aligned.
2181 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2182 instruction when memory address is aligned.
2183 (*vsx_le_perm_load_v16qi): Likewise.
2184 (four unnamed splitters): Modify to select the stvx instruction
2185 when memory is aligned.
2187 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2189 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2190 which appears in the queue multiple times.
2192 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2193 Alan Hayward <alan.hayward@arm.com>
2194 David Sherwood <david.sherwood@arm.com>
2196 * tree-vectorizer.h (vec_lower_bound): New structure.
2197 (_loop_vec_info): Add check_nonzero and lower_bounds.
2198 (LOOP_VINFO_CHECK_NONZERO): New macro.
2199 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2200 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2201 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2202 fields. Make seg_len the distance travelled, not including the
2204 (dr_direction_indicator): Declare.
2205 (dr_zero_step_indicator): Likewise.
2206 (dr_known_forward_stride_p): Likewise.
2207 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2209 (runtime_alias_check_p): Allow runtime alias checks with
2211 (operator ==): Compare access_size and align.
2212 (prune_runtime_alias_test_list): Rework for new distinction between
2213 the access_size and seg_len.
2214 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2216 (get_segment_min_max): New function.
2217 (create_intersect_range_checks): Use it.
2218 (dr_step_indicator): New function.
2219 (dr_direction_indicator): Likewise.
2220 (dr_zero_step_indicator): Likewise.
2221 (dr_known_forward_stride_p): Likewise.
2222 * tree-loop-distribution.c (data_ref_segment_size): Return
2223 DR_STEP * (niters - 1).
2224 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2226 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2227 (vect_preserves_scalar_order_p): New function, split out from...
2228 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2229 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2230 (vect_vfa_access_size): New function.
2231 (vect_vfa_align): Likewise.
2232 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2233 (dump_lower_bound): New function.
2234 (vect_check_lower_bound): Likewise.
2235 (vect_small_gap_p): Likewise.
2236 (vectorizable_with_step_bound_p): Likewise.
2237 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2238 depencies if the vectorization factor is 1. Convert the checks
2239 for nonzero steps into checks on the bounds of DR_STEP. Try using
2240 a bunds check for variable steps if the minimum required step is
2241 relatively small. Update calls to the dr_with_seg_len
2242 constructor and to vect_compile_time_alias.
2243 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2245 (vect_loop_versioning): Call it.
2246 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2248 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2250 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2251 Alan Hayward <alan.hayward@arm.com>
2252 David Sherwood <david.sherwood@arm.com>
2254 * doc/sourcebuild.texi (vect_scatter_store): Document.
2255 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2257 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2259 * genopinit.c (main): Add supports_vec_scatter_store and
2260 supports_vec_scatter_store_cached to target_optabs.
2261 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2262 IFN_MASK_SCATTER_STORE.
2263 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2265 * internal-fn.h (internal_store_fn_p): Declare.
2266 (internal_fn_stored_value_index): Likewise.
2267 * internal-fn.c (scatter_store_direct): New macro.
2268 (expand_scatter_store_optab_fn): New function.
2269 (direct_scatter_store_optab_supported_p): New macro.
2270 (internal_store_fn_p): New function.
2271 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2272 IFN_MASK_SCATTER_STORE.
2273 (internal_fn_mask_index): Likewise.
2274 (internal_fn_stored_value_index): New function.
2275 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2277 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2278 * optabs-query.c (supports_vec_scatter_store_p): New function.
2279 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2280 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2281 true for scatter stores.
2282 (vect_gather_scatter_fn_p): Handle scatter stores too.
2283 (vect_check_gather_scatter): Consider using scatter stores if
2284 supports_vec_scatter_store_p.
2285 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2287 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2288 internal_fn_stored_value_index.
2289 (check_load_store_masking): Handle scatter stores too.
2290 (vect_get_store_rhs): Make public.
2291 (vectorizable_call): Use internal_store_fn_p.
2292 (vectorizable_store): Handle scatter store internal functions.
2293 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2294 when deciding whether the end of the group has been reached.
2295 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2296 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2297 (mask_scatter_store<mode>): New insns.
2299 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2300 Alan Hayward <alan.hayward@arm.com>
2301 David Sherwood <david.sherwood@arm.com>
2303 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2304 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2305 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2307 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2308 Use vect_truncate_gather_scatter_offset if we can't treat the
2309 operation as a normal gather load or scatter store.
2310 (get_group_load_store_type): Take the gather_scatter_info
2311 as argument. Try using a gather load or scatter store for
2312 single-element groups.
2313 (get_load_store_type): Update calls to get_group_load_store_type
2314 and vect_use_strided_gather_scatters_p.
2316 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2317 Alan Hayward <alan.hayward@arm.com>
2318 David Sherwood <david.sherwood@arm.com>
2320 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2321 optional tree argument.
2322 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2324 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2325 but continue to use the current value as a fallback.
2326 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2327 to compare the updates.
2328 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2329 (get_load_store_type): Use it when handling a strided access.
2330 (vect_get_strided_load_store_ops): New function.
2331 (vect_get_data_ptr_increment): Likewise.
2332 (vectorizable_load): Handle strided gather loads. Always pass
2333 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2335 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2336 Alan Hayward <alan.hayward@arm.com>
2337 David Sherwood <david.sherwood@arm.com>
2339 * doc/md.texi (gather_load@var{m}): Document.
2340 (mask_gather_load@var{m}): Likewise.
2341 * genopinit.c (main): Add supports_vec_gather_load and
2342 supports_vec_gather_load_cached to target_optabs.
2343 * optabs-tree.c (init_tree_optimization_optabs): Use
2344 ggc_cleared_alloc to allocate target_optabs.
2345 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2346 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2348 * internal-fn.h (internal_load_fn_p): Declare.
2349 (internal_gather_scatter_fn_p): Likewise.
2350 (internal_fn_mask_index): Likewise.
2351 (internal_gather_scatter_fn_supported_p): Likewise.
2352 * internal-fn.c (gather_load_direct): New macro.
2353 (expand_gather_load_optab_fn): New function.
2354 (direct_gather_load_optab_supported_p): New macro.
2355 (direct_internal_fn_optab): New function.
2356 (internal_load_fn_p): Likewise.
2357 (internal_gather_scatter_fn_p): Likewise.
2358 (internal_fn_mask_index): Likewise.
2359 (internal_gather_scatter_fn_supported_p): Likewise.
2360 * optabs-query.c (supports_at_least_one_mode_p): New function.
2361 (supports_vec_gather_load_p): Likewise.
2362 * optabs-query.h (supports_vec_gather_load_p): Declare.
2363 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2364 and memory_type field.
2365 (NUM_PATTERNS): Bump to 15.
2366 * tree-vect-data-refs.c: Include internal-fn.h.
2367 (vect_gather_scatter_fn_p): New function.
2368 (vect_describe_gather_scatter_call): Likewise.
2369 (vect_check_gather_scatter): Try using internal functions for
2370 gather loads. Recognize existing calls to a gather load function.
2371 (vect_analyze_data_refs): Consider using gather loads if
2372 supports_vec_gather_load_p.
2373 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2374 (vect_get_gather_scatter_offset_type): Likewise.
2375 (vect_convert_mask_for_vectype): Likewise.
2376 (vect_add_conversion_to_patterm): Likewise.
2377 (vect_try_gather_scatter_pattern): Likewise.
2378 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2379 (vect_vect_recog_func_ptrs): Add it.
2380 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2381 internal_fn_mask_index and internal_gather_scatter_fn_p.
2382 (check_load_store_masking): Take the gather_scatter_info as an
2383 argument and handle gather loads.
2384 (vect_get_gather_scatter_ops): New function.
2385 (vectorizable_call): Check internal_load_fn_p.
2386 (vectorizable_load): Likewise. Handle gather load internal
2388 (vectorizable_store): Update call to check_load_store_masking.
2389 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2390 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2391 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2392 (aarch64_gather_scale_operand_d): New predicates.
2393 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2394 (mask_gather_load<mode>): New insns.
2396 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2397 Alan Hayward <alan.hayward@arm.com>
2398 David Sherwood <david.sherwood@arm.com>
2400 * optabs.def (fold_left_plus_optab): New optab.
2401 * doc/md.texi (fold_left_plus_@var{m}): Document.
2402 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2403 * internal-fn.c (fold_left_direct): Define.
2404 (expand_fold_left_optab_fn): Likewise.
2405 (direct_fold_left_optab_supported_p): Likewise.
2406 * fold-const-call.c (fold_const_fold_left): New function.
2407 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2408 * tree-parloops.c (valid_reduction_p): New function.
2409 (gather_scalar_reductions): Use it.
2410 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2411 (vect_finish_replace_stmt): Declare.
2412 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2413 (needs_fold_left_reduction_p): New function, split out from...
2414 (vect_is_simple_reduction): ...here. Accept reductions that
2415 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2416 (vect_force_simple_reduction): Also store the reduction type in
2417 the assignment's STMT_VINFO_REDUC_TYPE.
2418 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2419 (merge_with_identity): New function.
2420 (vect_expand_fold_left): Likewise.
2421 (vectorize_fold_left_reduction): Likewise.
2422 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2423 scalar phi in place for it. Check for target support and reject
2424 cases that would reassociate the operation. Defer the transform
2425 phase to vectorize_fold_left_reduction.
2426 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2427 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2428 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2430 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2432 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2433 call to ifc_temp_var.
2435 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2436 Alan Hayward <alan.hayward@arm.com>
2437 David Sherwood <david.sherwood@arm.com>
2439 * target.def (legitimize_address_displacement): Take the original
2440 offset as a poly_int.
2441 * targhooks.h (default_legitimize_address_displacement): Update
2443 * targhooks.c (default_legitimize_address_displacement): Likewise.
2444 * doc/tm.texi: Regenerate.
2445 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2446 as an argument, moving assert of ad->disp == ad->disp_term to...
2447 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2448 Try calling targetm.legitimize_address_displacement before expanding
2449 the address rather than afterwards, and adjust for the new interface.
2450 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2451 Match the new hook interface. Handle SVE addresses.
2452 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2455 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2457 * Makefile.in (OBJS): Add early-remat.o.
2458 * target.def (select_early_remat_modes): New hook.
2459 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2460 * doc/tm.texi: Regenerate.
2461 * targhooks.h (default_select_early_remat_modes): Declare.
2462 * targhooks.c (default_select_early_remat_modes): New function.
2463 * timevar.def (TV_EARLY_REMAT): New timevar.
2464 * passes.def (pass_early_remat): New pass.
2465 * tree-pass.h (make_pass_early_remat): Declare.
2466 * early-remat.c: New file.
2467 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2469 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2471 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2472 Alan Hayward <alan.hayward@arm.com>
2473 David Sherwood <david.sherwood@arm.com>
2475 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2476 vfm1 with a bound_epilog parameter.
2477 (vect_do_peeling): Update calls accordingly, and move the prologue
2478 call earlier in the function. Treat the base bound_epilog as 0 for
2479 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2480 this base when peeling for gaps.
2481 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2482 with fully-masked loops.
2483 (vect_estimate_min_profitable_iters): Handle the single peeled
2484 iteration in that case.
2486 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2487 Alan Hayward <alan.hayward@arm.com>
2488 David Sherwood <david.sherwood@arm.com>
2490 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2491 single-element interleaving even if the size is not a power of 2.
2492 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2493 accesses for single-element interleaving if the group size is
2496 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2497 Alan Hayward <alan.hayward@arm.com>
2498 David Sherwood <david.sherwood@arm.com>
2500 * doc/md.texi (fold_extract_last_@var{m}): Document.
2501 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2502 * optabs.def (fold_extract_last_optab): New optab.
2503 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2504 * internal-fn.c (fold_extract_direct): New macro.
2505 (expand_fold_extract_optab_fn): Likewise.
2506 (direct_fold_extract_optab_supported_p): Likewise.
2507 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2508 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2509 EXTRACT_LAST_REDUCTION.
2510 (get_initial_def_for_reduction): Do not create an initial vector
2511 for EXTRACT_LAST_REDUCTION reductions.
2512 (vectorizable_reduction): Leave the scalar phi in place for
2513 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2514 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2515 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2516 transform phase to vectorizable_condition.
2517 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2519 (vect_finish_stmt_generation): ...here.
2520 (vect_finish_replace_stmt): New function.
2521 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2522 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2524 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2526 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2527 Alan Hayward <alan.hayward@arm.com>
2528 David Sherwood <david.sherwood@arm.com>
2530 * doc/md.texi (extract_last_@var{m}): Document.
2531 * optabs.def (extract_last_optab): New optab.
2532 * internal-fn.def (EXTRACT_LAST): New internal function.
2533 * internal-fn.c (cond_unary_direct): New macro.
2534 (expand_cond_unary_optab_fn): Likewise.
2535 (direct_cond_unary_optab_supported_p): Likewise.
2536 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2537 loops using EXTRACT_LAST.
2538 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2539 (extract_last_<mode>): ...this optab.
2540 (vec_extract<mode><Vel>): Update accordingly.
2542 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2543 Alan Hayward <alan.hayward@arm.com>
2544 David Sherwood <david.sherwood@arm.com>
2546 * target.def (empty_mask_is_expensive): New hook.
2547 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2548 * doc/tm.texi: Regenerate.
2549 * targhooks.h (default_empty_mask_is_expensive): Declare.
2550 * targhooks.c (default_empty_mask_is_expensive): New function.
2551 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2552 if the target says that empty masks are expensive.
2553 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2555 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2557 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2558 Alan Hayward <alan.hayward@arm.com>
2559 David Sherwood <david.sherwood@arm.com>
2561 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2562 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2563 (vect_use_loop_mask_for_alignment_p): New function.
2564 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2565 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2566 niters_skip argument. Make sure that the first niters_skip elements
2567 of the first iteration are inactive.
2568 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2569 Update call to vect_set_loop_masks_directly.
2570 (get_misalign_in_elems): New function, split out from...
2571 (vect_gen_prolog_loop_niters): ...here.
2572 (vect_update_init_of_dr): Take a code argument that specifies whether
2573 the adjustment should be added or subtracted.
2574 (vect_update_init_of_drs): Likewise.
2575 (vect_prepare_for_masked_peels): New function.
2576 (vect_do_peeling): Skip prologue peeling if we're using a mask
2577 instead. Update call to vect_update_inits_of_drs.
2578 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2580 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2581 alignment. Do not include the number of peeled iterations in
2582 the minimum threshold in that case.
2583 (vectorizable_induction): Adjust the start value down by
2584 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2585 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2586 Take the number of skipped iterations into account when calculating
2588 * tree-vect-stmts.c (vect_gen_while_not): New function.
2590 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2591 Alan Hayward <alan.hayward@arm.com>
2592 David Sherwood <david.sherwood@arm.com>
2594 * doc/sourcebuild.texi (vect_fully_masked): Document.
2595 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2597 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2599 (vect_analyze_loop_2): ...here. Don't check the vectorization
2600 factor against the number of loop iterations if the loop is
2603 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2604 Alan Hayward <alan.hayward@arm.com>
2605 David Sherwood <david.sherwood@arm.com>
2607 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2608 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2609 (dump_groups): Update accordingly.
2610 (iv_use::mem_type): New member variable.
2611 (address_p): New function.
2612 (record_use): Add a mem_type argument and initialize the new
2614 (record_group_use): Add a mem_type argument. Use address_p.
2615 Remove obsolete null checks of base_object. Update call to record_use.
2616 (find_interesting_uses_op): Update call to record_group_use.
2617 (find_interesting_uses_cond): Likewise.
2618 (find_interesting_uses_address): Likewise.
2619 (get_mem_type_for_internal_fn): New function.
2620 (find_address_like_use): Likewise.
2621 (find_interesting_uses_stmt): Try find_address_like_use before
2622 calling find_interesting_uses_op.
2623 (addr_offset_valid_p): Use the iv mem_type field as the type
2624 of the addressed memory.
2625 (add_autoinc_candidates): Likewise.
2626 (get_address_cost): Likewise.
2627 (split_small_address_groups_p): Use address_p.
2628 (split_address_groups): Likewise.
2629 (add_iv_candidate_for_use): Likewise.
2630 (autoinc_possible_for_pair): Likewise.
2631 (rewrite_groups): Likewise.
2632 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2633 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2634 (get_alias_ptr_type_for_ptr_address): New function.
2635 (rewrite_use_address): Rewrite address uses in calls that were
2636 identified by find_address_like_use.
2638 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2639 Alan Hayward <alan.hayward@arm.com>
2640 David Sherwood <david.sherwood@arm.com>
2642 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2644 * gimple-expr.h (is_gimple_addressable: Likewise.
2645 * gimple-expr.c (is_gimple_address): Likewise.
2646 * internal-fn.c (expand_call_mem_ref): New function.
2647 (expand_mask_load_optab_fn): Use it.
2648 (expand_mask_store_optab_fn): Likewise.
2650 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2651 Alan Hayward <alan.hayward@arm.com>
2652 David Sherwood <david.sherwood@arm.com>
2654 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2655 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2656 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2657 (cond_umax@var{mode}): Document.
2658 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2659 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2660 (cond_umin_optab, cond_umax_optab): New optabs.
2661 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2662 (COND_IOR, COND_XOR): New internal functions.
2663 * internal-fn.h (get_conditional_internal_fn): Declare.
2664 * internal-fn.c (cond_binary_direct): New macro.
2665 (expand_cond_binary_optab_fn): Likewise.
2666 (direct_cond_binary_optab_supported_p): Likewise.
2667 (get_conditional_internal_fn): New function.
2668 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2669 Cope with reduction statements that are vectorized as calls rather
2671 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2672 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2673 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2674 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2675 (UNSPEC_COND_EOR): New unspecs.
2676 (optab): Add mappings for them.
2677 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2678 (sve_int_op, sve_fp_op): New int attributes.
2680 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2681 Alan Hayward <alan.hayward@arm.com>
2682 David Sherwood <david.sherwood@arm.com>
2684 * optabs.def (while_ult_optab): New optab.
2685 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2686 * internal-fn.def (WHILE_ULT): New internal function.
2687 * internal-fn.h (direct_internal_fn_supported_p): New override
2688 that takes two types as argument.
2689 * internal-fn.c (while_direct): New macro.
2690 (expand_while_optab_fn): New function.
2691 (convert_optab_supported_p): Likewise.
2692 (direct_while_optab_supported_p): New macro.
2693 * wide-int.h (wi::udiv_ceil): New function.
2694 * tree-vectorizer.h (rgroup_masks): New structure.
2695 (vec_loop_masks): New typedef.
2696 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2698 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2699 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2700 (vect_max_vf): New function.
2701 (slpeel_make_loop_iterate_ntimes): Delete.
2702 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2703 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2704 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2705 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2706 internal-fn.h, stor-layout.h and optabs-query.h.
2707 (vect_set_loop_mask): New function.
2708 (add_preheader_seq): Likewise.
2709 (add_header_seq): Likewise.
2710 (interleave_supported_p): Likewise.
2711 (vect_maybe_permute_loop_masks): Likewise.
2712 (vect_set_loop_masks_directly): Likewise.
2713 (vect_set_loop_condition_masked): Likewise.
2714 (vect_set_loop_condition_unmasked): New function, split out from
2715 slpeel_make_loop_iterate_ntimes.
2716 (slpeel_make_loop_iterate_ntimes): Rename to..
2717 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2718 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2719 (vect_do_peeling): Update call accordingly.
2720 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2722 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2723 mask_compare_type, can_fully_mask_p and fully_masked_p.
2724 (release_vec_loop_masks): New function.
2725 (_loop_vec_info): Use it to free the loop masks.
2726 (can_produce_all_loop_masks_p): New function.
2727 (vect_get_max_nscalars_per_iter): Likewise.
2728 (vect_verify_full_masking): Likewise.
2729 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2730 retries, and free the mask rgroups before retrying. Check loop-wide
2731 reasons for disallowing fully-masked loops. Make the final decision
2732 about whether use a fully-masked loop or not.
2733 (vect_estimate_min_profitable_iters): Do not assume that peeling
2734 for the number of iterations will be needed for fully-masked loops.
2735 (vectorizable_reduction): Disable fully-masked loops.
2736 (vectorizable_live_operation): Likewise.
2737 (vect_halve_mask_nunits): New function.
2738 (vect_double_mask_nunits): Likewise.
2739 (vect_record_loop_mask): Likewise.
2740 (vect_get_loop_mask): Likewise.
2741 (vect_transform_loop): Handle the case in which the final loop
2742 iteration might handle a partial vector. Call vect_set_loop_condition
2743 instead of slpeel_make_loop_iterate_ntimes.
2744 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2745 (check_load_store_masking): New function.
2746 (prepare_load_store_mask): Likewise.
2747 (vectorizable_store): Handle fully-masked loops.
2748 (vectorizable_load): Likewise.
2749 (supportable_widening_operation): Use vect_halve_mask_nunits for
2751 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2752 (vect_gen_while): New function.
2753 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2754 (aarch64_uqdec<mode>): New insn.
2756 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2757 Alan Hayward <alan.hayward@arm.com>
2758 David Sherwood <david.sherwood@arm.com>
2760 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2761 (reduc_xor_scal_optab): New optabs.
2762 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2763 (reduc_xor_scal_@var{m}): Document.
2764 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2765 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2767 * fold-const-call.c (fold_const_call): Handle them.
2768 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2769 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2770 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2771 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2772 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2773 (UNSPEC_XORV): New unspecs.
2774 (optab): Add entries for them.
2775 (BITWISEV): New int iterator.
2776 (bit_reduc_op): New int attributes.
2778 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2779 Alan Hayward <alan.hayward@arm.com>
2780 David Sherwood <david.sherwood@arm.com>
2782 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2783 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2784 * optabs.def (vec_shl_insert_optab): New optab.
2785 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2786 (duplicate_and_interleave): Likewise.
2787 * tree-vect-loop.c: Include internal-fn.h.
2788 (neutral_op_for_slp_reduction): New function, split out from
2789 get_initial_defs_for_reduction.
2790 (get_initial_def_for_reduction): Handle option 2 for variable-length
2791 vectors by loading the neutral value into a vector and then shifting
2792 the initial value into element 0.
2793 (get_initial_defs_for_reduction): Replace the code argument with
2794 the neutral value calculated by neutral_op_for_slp_reduction.
2795 Use gimple_build_vector for constant-length vectors.
2796 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2797 but the first group_size elements have a neutral value.
2798 Use duplicate_and_interleave otherwise.
2799 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2800 Update call to get_initial_defs_for_reduction. Handle SLP
2801 reductions for variable-length vectors by creating one vector
2802 result for each scalar result, with the elements associated
2803 with other scalar results stubbed out with the neutral value.
2804 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2805 Require IFN_VEC_SHL_INSERT for double reductions on
2806 variable-length vectors, or SLP reductions that have
2807 a neutral value. Require can_duplicate_and_interleave_p
2808 support for variable-length unchained SLP reductions if there
2809 is no neutral value, such as for MIN/MAX reductions. Also require
2810 the number of vector elements to be a multiple of the number of
2811 SLP statements when doing variable-length unchained SLP reductions.
2812 Update call to vect_create_epilog_for_reduction.
2813 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2814 and remove initial values.
2815 (duplicate_and_interleave): Make public.
2816 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2817 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2819 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2820 Alan Hayward <alan.hayward@arm.com>
2821 David Sherwood <david.sherwood@arm.com>
2823 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2824 (can_duplicate_and_interleave_p): New function.
2825 (vect_get_and_check_slp_defs): Take the vector of statements
2826 rather than just the current one. Remove excess parentheses.
2827 Restriction rejectinon of vect_constant_def and vect_external_def
2828 for variable-length vectors to boolean types, or types for which
2829 can_duplicate_and_interleave_p is false.
2830 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2831 (duplicate_and_interleave): New function.
2832 (vect_get_constant_vectors): Use gimple_build_vector for
2833 constant-length vectors and suitable variable-length constant
2834 vectors. Use duplicate_and_interleave for other variable-length
2835 vectors. Don't defer the update when inserting new statements.
2837 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2838 Alan Hayward <alan.hayward@arm.com>
2839 David Sherwood <david.sherwood@arm.com>
2841 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2842 min_profitable_iters doesn't go negative.
2844 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2845 Alan Hayward <alan.hayward@arm.com>
2846 David Sherwood <david.sherwood@arm.com>
2848 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2849 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2850 * optabs.def (vec_mask_load_lanes_optab): New optab.
2851 (vec_mask_store_lanes_optab): Likewise.
2852 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2853 (MASK_STORE_LANES): Likewise.
2854 * internal-fn.c (mask_load_lanes_direct): New macro.
2855 (mask_store_lanes_direct): Likewise.
2856 (expand_mask_load_optab_fn): Handle masked operations.
2857 (expand_mask_load_lanes_optab_fn): New macro.
2858 (expand_mask_store_optab_fn): Handle masked operations.
2859 (expand_mask_store_lanes_optab_fn): New macro.
2860 (direct_mask_load_lanes_optab_supported_p): Likewise.
2861 (direct_mask_store_lanes_optab_supported_p): Likewise.
2862 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2864 (vect_load_lanes_supported): Likewise.
2865 * tree-vect-data-refs.c (strip_conversion): New function.
2866 (can_group_stmts_p): Likewise.
2867 (vect_analyze_data_ref_accesses): Use it instead of checking
2868 for a pair of assignments.
2869 (vect_store_lanes_supported): Take a masked_p parameter.
2870 (vect_load_lanes_supported): Likewise.
2871 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2872 vect_store_lanes_supported and vect_load_lanes_supported.
2873 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2874 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2875 parameter. Don't allow gaps for masked accesses.
2876 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2877 and vect_load_lanes_supported.
2878 (get_load_store_type): Take a masked_p parameter and update
2879 call to get_group_load_store_type.
2880 (vectorizable_store): Update call to get_load_store_type.
2881 Handle IFN_MASK_STORE_LANES.
2882 (vectorizable_load): Update call to get_load_store_type.
2883 Handle IFN_MASK_LOAD_LANES.
2885 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2886 Alan Hayward <alan.hayward@arm.com>
2887 David Sherwood <david.sherwood@arm.com>
2889 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2891 * config/aarch64/aarch64-protos.h
2892 (aarch64_sve_struct_memory_operand_p): Declare.
2893 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2894 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2895 (VPRED, vpred): Handle SVE structure modes.
2896 * config/aarch64/constraints.md (Utx): New constraint.
2897 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2898 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2899 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2900 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2901 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2902 structure modes. Split into pieces after RA.
2903 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2904 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2906 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2907 SVE structure modes.
2908 (aarch64_classify_address): Likewise.
2909 (sizetochar): Move earlier in file.
2910 (aarch64_print_operand): Handle SVE register lists.
2911 (aarch64_array_mode): New function.
2912 (aarch64_sve_struct_memory_operand_p): Likewise.
2913 (TARGET_ARRAY_MODE): Redefine.
2915 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2916 Alan Hayward <alan.hayward@arm.com>
2917 David Sherwood <david.sherwood@arm.com>
2919 * target.def (array_mode): New target hook.
2920 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2921 * doc/tm.texi: Regenerate.
2922 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2923 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2924 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2926 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2929 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2930 Alan Hayward <alan.hayward@arm.com>
2931 David Sherwood <david.sherwood@arm.com>
2933 * fold-const.c (fold_binary_loc): Check the argument types
2934 rather than the result type when testing for a vector operation.
2936 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2938 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2939 * doc/tm.texi: Regenerate.
2941 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2942 Alan Hayward <alan.hayward@arm.com>
2943 David Sherwood <david.sherwood@arm.com>
2945 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2946 (sve): Document new AArch64 extension.
2947 * doc/md.texi (w): Extend the description of the AArch64
2948 constraint to include SVE vectors.
2949 (Upl, Upa): Document new AArch64 predicate constraints.
2950 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2952 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2953 (msve-vector-bits=): New option.
2954 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2955 SVE when these are disabled.
2956 (sve): New extension.
2957 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2958 modes. Adjust their number of units based on aarch64_sve_vg.
2959 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2960 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2961 aarch64_addr_query_type.
2962 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2963 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2964 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2965 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2966 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2967 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2968 (aarch64_simd_imm_zero_p): Delete.
2969 (aarch64_check_zero_based_sve_index_immediate): Declare.
2970 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2971 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2972 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2973 (aarch64_sve_float_mul_immediate_p): Likewise.
2974 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2976 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2977 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2978 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2979 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2980 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2981 (aarch64_regmode_natural_size): Likewise.
2982 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2983 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2985 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2986 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2987 for VG and the SVE predicate registers.
2988 (V_ALIASES): Add a "z"-prefixed alias.
2989 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2990 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2991 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2992 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2993 (REG_CLASS_NAMES): Add entries for them.
2994 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2995 and the predicate registers.
2996 (aarch64_sve_vg): Declare.
2997 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2998 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2999 (REGMODE_NATURAL_SIZE): Define.
3000 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3002 * config/aarch64/aarch64.c: Include cfgrtl.h.
3003 (simd_immediate_info): Add a constructor for series vectors,
3004 and an associated step field.
3005 (aarch64_sve_vg): New variable.
3006 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3007 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3008 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3009 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3010 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3011 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3012 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3013 (aarch64_get_mask_mode): New functions.
3014 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3015 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3016 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3017 predicate modes and predicate registers. Explicitly restrict
3018 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3019 to store a vector mode if it is recognized by
3020 aarch64_classify_vector_mode.
3021 (aarch64_regmode_natural_size): New function.
3022 (aarch64_hard_regno_caller_save_mode): Return the original mode
3024 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3025 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3026 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3027 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3029 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3030 does not overlap dest if the function is frame-related. Handle
3032 (aarch64_split_add_offset): New function.
3033 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3034 them aarch64_add_offset.
3035 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3036 and update call to aarch64_sub_sp.
3037 (aarch64_add_cfa_expression): New function.
3038 (aarch64_expand_prologue): Pass extra temporary registers to the
3039 functions above. Handle the case in which we need to emit new
3040 DW_CFA_expressions for registers that were originally saved
3041 relative to the stack pointer, but now have to be expressed
3042 relative to the frame pointer.
3043 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3045 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3046 IP0 and IP1 values for SVE frames.
3047 (aarch64_expand_vec_series): New function.
3048 (aarch64_expand_sve_widened_duplicate): Likewise.
3049 (aarch64_expand_sve_const_vector): Likewise.
3050 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3051 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3052 into the register, rather than emitting a SET directly.
3053 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3054 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3055 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3056 (offset_9bit_signed_scaled_p): New functions.
3057 (aarch64_replicate_bitmask_imm): New function.
3058 (aarch64_bitmask_imm): Use it.
3059 (aarch64_cannot_force_const_mem): Reject expressions involving
3060 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3061 (aarch64_classify_index): Handle SVE indices, by requiring
3062 a plain register index with a scale that matches the element size.
3063 (aarch64_classify_address): Handle SVE addresses. Assert that
3064 the mode of the address is VOIDmode or an integer mode.
3065 Update call to aarch64_classify_symbol.
3066 (aarch64_classify_symbolic_expression): Update call to
3067 aarch64_classify_symbol.
3068 (aarch64_const_vec_all_in_range_p): New function.
3069 (aarch64_print_vector_float_operand): Likewise.
3070 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3071 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3072 and the FP immediates 1.0 and 0.5.
3073 (aarch64_print_address_internal): Handle SVE addresses.
3074 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3075 (aarch64_regno_regclass): Handle predicate registers.
3076 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3078 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3079 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3080 (aarch64_convert_sve_vector_bits): New function.
3081 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3082 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3084 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3085 Handle SVE vector and predicate modes. Accept VL-based constants
3086 that need only one temporary register, and VL offsets that require
3087 no temporary registers.
3088 (aarch64_conditional_register_usage): Mark the predicate registers
3089 as fixed if SVE isn't available.
3090 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3091 Return true for SVE vector and predicate modes.
3092 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3093 rather than an unsigned int. Handle SVE modes.
3094 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3096 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3098 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3099 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3100 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3101 (aarch64_sve_float_mul_immediate_p): New functions.
3102 (aarch64_sve_valid_immediate): New function.
3103 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3104 Explicitly reject structure modes. Check for INDEX constants.
3105 Handle PTRUE and PFALSE constants.
3106 (aarch64_check_zero_based_sve_index_immediate): New function.
3107 (aarch64_simd_imm_zero_p): Delete.
3108 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3109 vector modes. Accept constants in the range of CNT[BHWD].
3110 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3111 ask for an Advanced SIMD mode.
3112 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3113 (aarch64_simd_vector_alignment): Handle SVE predicates.
3114 (aarch64_vectorize_preferred_vector_alignment): New function.
3115 (aarch64_simd_vector_alignment_reachable): Use it instead of
3117 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3118 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3120 (MAX_VECT_LEN): Delete.
3121 (expand_vec_perm_d): Add a vec_flags field.
3122 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3123 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3124 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3126 (aarch64_evpc_rev): Rename to...
3127 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3128 (aarch64_evpc_rev_global): New function.
3129 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3130 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3132 (aarch64_evpc_sve_tbl): New function.
3133 (aarch64_expand_vec_perm_const_1): Update after rename of
3134 aarch64_evpc_rev. Handle SVE permutes too, trying
3135 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3136 than aarch64_evpc_tbl.
3137 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3138 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3139 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3140 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3141 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3142 (aarch64_expand_sve_vcond): New functions.
3143 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3144 of aarch64_vector_mode_p.
3145 (aarch64_dwarf_poly_indeterminate_value): New function.
3146 (aarch64_compute_pressure_classes): Likewise.
3147 (aarch64_can_change_mode_class): Likewise.
3148 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3149 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3150 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3151 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3152 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3153 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3154 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3155 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3157 (Dn, Dl, Dr): Accept const as well as const_vector.
3158 (Dz): Likewise. Compare against CONST0_RTX.
3159 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3160 of "vector" where appropriate.
3161 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3162 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3163 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3164 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3165 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3166 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3167 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3168 (v_int_equiv): Extend to SVE modes.
3169 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3171 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3172 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3173 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3174 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3175 (SVE_COND_FP_CMP): New int iterators.
3176 (perm_hilo): Handle the new unpack unspecs.
3177 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3179 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3180 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3181 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3182 (aarch64_equality_operator, aarch64_constant_vector_operand)
3183 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3184 (aarch64_sve_nonimmediate_operand): Likewise.
3185 (aarch64_sve_general_operand): Likewise.
3186 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3187 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3188 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3189 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3190 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3191 (aarch64_sve_float_arith_immediate): Likewise.
3192 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3193 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3194 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3195 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3196 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3197 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3198 (aarch64_sve_float_arith_operand): Likewise.
3199 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3200 (aarch64_sve_float_mul_operand): Likewise.
3201 (aarch64_sve_vec_perm_operand): Likewise.
3202 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3203 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3204 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3205 as well as const_vector.
3206 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3207 in file. Use CONST0_RTX and CONSTM1_RTX.
3208 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3209 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3210 Use aarch64_simd_imm_zero.
3211 * config/aarch64/aarch64-sve.md: New file.
3212 * config/aarch64/aarch64.md: Include it.
3213 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3214 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3215 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3216 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3217 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3218 (sve): New attribute.
3219 (enabled): Disable instructions with the sve attribute unless
3221 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3222 aarch64_expand_mov_immediate.
3223 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3224 CNT[BHSD] immediates.
3225 (movti): Split CONST_POLY_INT moves into two halves.
3226 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3227 Split additions that need a temporary here if the destination
3228 is the stack pointer.
3229 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3230 (*add<mode>3_poly_1): New instruction.
3231 (set_clobber_cc): New expander.
3233 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3235 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3236 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3237 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3238 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3239 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3240 Change innermode from fixed_mode_size to machine_mode.
3241 (simplify_subreg): Update call accordingly. Handle a constant-sized
3242 subreg of a variable-length CONST_VECTOR.
3244 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3245 Alan Hayward <alan.hayward@arm.com>
3246 David Sherwood <david.sherwood@arm.com>
3248 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3249 (add_offset_to_base): New function, split out from...
3250 (create_mem_ref): ...here. When handling a scale other than 1,
3251 check first whether the address is valid without the offset.
3252 Add it into the base if so, leaving the index and scale as-is.
3254 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3257 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3258 fold_for_warn before checking if arg2 is INTEGER_CST.
3260 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3262 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3263 (store_multiple_operation): Delete.
3264 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3265 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3266 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3267 guarded by TARGET_STRING.
3268 (rs6000_output_load_multiple): Delete.
3269 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3270 OPTION_MASK_STRING / TARGET_STRING handling.
3271 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3272 (const rs6000_opt_masks) <"string">: Change mask to 0.
3273 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3274 (MASK_STRING): Delete.
3275 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3277 (load_multiple): Delete.
3284 (store_multiple): Delete.
3291 (movmemsi_8reg): Delete.
3292 (corresponding unnamed define_insn): Delete.
3293 (movmemsi_6reg): Delete.
3294 (corresponding unnamed define_insn): Delete.
3295 (movmemsi_4reg): Delete.
3296 (corresponding unnamed define_insn): Delete.
3297 (movmemsi_2reg): Delete.
3298 (corresponding unnamed define_insn): Delete.
3299 (movmemsi_1reg): Delete.
3300 (corresponding unnamed define_insn): Delete.
3301 * config/rs6000/rs6000.opt (mno-string): New.
3302 (mstring): Replace by deprecation warning stub.
3303 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3305 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3307 * regrename.c (regrename_do_replace): If replacing the same
3308 reg multiple times, try to reuse last created gen_raw_REG.
3311 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3312 main to workaround a bug in GDB.
3314 2018-01-12 Tom de Vries <tom@codesourcery.com>
3317 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3319 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3321 PR rtl-optimization/80481
3322 * ira-color.c (get_cap_member): New function.
3323 (allocnos_conflict_by_live_ranges_p): Use it.
3324 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3325 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3327 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3330 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3331 (*saddl_se_1): Ditto.
3333 (*ssubl_se_1): Ditto.
3335 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3337 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3338 rather than wi::to_widest for DR_INITs.
3339 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3340 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3341 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3343 (vect_analyze_group_access_1): Note that here.
3345 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3347 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3348 polynomial type sizes.
3350 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3352 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3353 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3354 (gimple_add_tmp_var): Likewise.
3356 2018-01-12 Martin Liska <mliska@suse.cz>
3358 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3359 (gimple_alloc_sizes): Likewise.
3360 (dump_gimple_statistics): Use PRIu64 in printf format.
3361 * gimple.h: Change uint64_t to int.
3363 2018-01-12 Martin Liska <mliska@suse.cz>
3365 * tree-core.h: Use uint64_t instead of int.
3366 * tree.c (tree_node_counts): Likewise.
3367 (tree_node_sizes): Likewise.
3368 (dump_tree_statistics): Use PRIu64 in printf format.
3370 2018-01-12 Martin Liska <mliska@suse.cz>
3372 * Makefile.in: As qsort_chk is implemented in vec.c, add
3373 vec.o to linkage of gencfn-macros.
3374 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3375 passing the info to record_node_allocation_statistics.
3376 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3378 * ggc-common.c (struct ggc_usage): Add operator== and use
3379 it in operator< and compare function.
3380 * mem-stats.h (struct mem_usage): Likewise.
3381 * vec.c (struct vec_usage): Remove operator< and compare
3382 function. Can be simply inherited.
3384 2018-01-12 Martin Jambor <mjambor@suse.cz>
3387 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3388 * tree-ssa-math-opts.c: Include domwalk.h.
3389 (convert_mult_to_fma_1): New function.
3390 (fma_transformation_info): New type.
3391 (fma_deferring_state): Likewise.
3392 (cancel_fma_deferring): New function.
3393 (result_of_phi): Likewise.
3394 (last_fma_candidate_feeds_initial_phi): Likewise.
3395 (convert_mult_to_fma): Added deferring logic, split actual
3396 transformation to convert_mult_to_fma_1.
3397 (math_opts_dom_walker): New type.
3398 (math_opts_dom_walker::after_dom_children): New method, body moved
3399 here from pass_optimize_widening_mul::execute, added deferring logic
3401 (pass_optimize_widening_mul::execute): Moved most of code to
3402 math_opts_dom_walker::after_dom_children.
3403 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3404 * config/i386/i386.c (ix86_option_override_internal): Added
3405 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3407 2018-01-12 Richard Biener <rguenther@suse.de>
3410 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3411 inline instance vars.
3413 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3416 * config/rx/rx.c (rx_is_restricted_memory_address):
3419 2018-01-12 Richard Biener <rguenther@suse.de>
3421 PR tree-optimization/80846
3422 * target.def (split_reduction): New target hook.
3423 * targhooks.c (default_split_reduction): New function.
3424 * targhooks.h (default_split_reduction): Declare.
3425 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3426 target requests first reduce vectors by combining low and high
3428 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3429 (get_vectype_for_scalar_type_and_size): Export.
3430 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3431 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3432 * doc/tm.texi: Regenerate.
3433 * config/i386/i386.c (ix86_split_reduction): Implement
3434 TARGET_VECTORIZE_SPLIT_REDUCTION.
3436 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3439 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3440 in PIC mode except for TARGET_VXWORKS_RTP.
3441 * config/sparc/sparc.c: Include cfgrtl.h.
3442 (TARGET_INIT_PIC_REG): Define.
3443 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3444 (sparc_pic_register_p): New predicate.
3445 (sparc_legitimate_address_p): Use it.
3446 (sparc_legitimize_pic_address): Likewise.
3447 (sparc_delegitimize_address): Likewise.
3448 (sparc_mode_dependent_address_p): Likewise.
3449 (gen_load_pcrel_sym): Remove 4th parameter.
3450 (load_got_register): Adjust call to above. Remove obsolete stuff.
3451 (sparc_expand_prologue): Do not call load_got_register here.
3452 (sparc_flat_expand_prologue): Likewise.
3453 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3454 (sparc_use_pseudo_pic_reg): New function.
3455 (sparc_init_pic_reg): Likewise.
3456 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3457 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3459 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3461 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3462 Add item for branch_cost.
3464 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3466 PR rtl-optimization/83565
3467 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3468 not extend the result to a larger mode for rotate operations.
3469 (num_sign_bit_copies1): Likewise.
3471 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3474 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3476 Use values-Xc.o for -pedantic.
3477 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3479 2018-01-12 Martin Liska <mliska@suse.cz>
3482 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3484 (possible_polymorphic_call_targets): Use it.
3485 (ipa_devirt): Likewise.
3487 2018-01-12 Martin Liska <mliska@suse.cz>
3489 * profile-count.h (enum profile_quality): Use 0 as invalid
3490 enum value of profile_quality.
3492 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3494 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3495 -mext-string options.
3497 2018-01-12 Richard Biener <rguenther@suse.de>
3499 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3500 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3501 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3503 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3505 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3507 * configure.ac (--with-long-double-format): Add support for the
3508 configuration option to change the default long double format on
3510 * config.gcc (powerpc*-linux*-*): Likewise.
3511 * configure: Regenerate.
3512 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3513 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3514 used without modification.
3516 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3518 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3519 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3520 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3521 MISC_BUILTIN_SPEC_BARRIER.
3522 (rs6000_init_builtins): Likewise.
3523 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3525 (speculation_barrier): New define_insn.
3526 * doc/extend.texi: Document __builtin_speculation_barrier.
3528 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3531 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3532 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3533 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3535 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3536 integral modes instead of "ss" and "sd".
3537 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3538 vectors with 32-bit and 64-bit elements.
3539 (vecdupssescalarmodesuffix): New mode attribute.
3540 (vec_dup<mode>): Use it.
3542 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3545 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3546 frame if argument is passed on stack.
3548 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3551 * ree.c (combine_reaching_defs): Optimize also
3552 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3553 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3555 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3558 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3560 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3563 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3564 after they are computed.
3566 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3568 PR tree-optimization/83695
3569 * gimple-loop-linterchange.cc
3570 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3571 reset cached scev information after interchange.
3572 (pass_linterchange::execute): Remove call to scev_reset_htab.
3574 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3576 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3577 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3578 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3579 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3580 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3581 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3582 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3583 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3584 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3585 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3586 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3587 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3588 (V_lane_reg): Likewise.
3589 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3591 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3592 (vfmal_lane_low<mode>_intrinsic,
3593 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3594 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3595 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3596 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3597 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3598 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3600 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3602 * config/arm/arm-cpus.in (fp16fml): New feature.
3603 (ALL_SIMD): Add fp16fml.
3604 (armv8.2-a): Add fp16fml as an option.
3605 (armv8.3-a): Likewise.
3606 (armv8.4-a): Add fp16fml as part of fp16.
3607 * config/arm/arm.h (TARGET_FP16FML): Define.
3608 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3610 * config/arm/arm-modes.def (V2HF): Define.
3611 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3612 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3613 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3614 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3615 vfmsl_low, vfmsl_high): New set of builtins.
3616 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3617 (vfml_op): New code attribute.
3618 (VFMLHALVES): New int iterator.
3619 (VFML, VFMLSEL): New mode attributes.
3620 (V_reg): Define mapping for V2HF.
3621 (V_hi, V_lo): New mode attributes.
3622 (VF_constraint): Likewise.
3623 (vfml_half, vfml_half_selector): New int attributes.
3624 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3626 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3627 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3629 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3630 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3631 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3632 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3634 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3635 Document new effective target and option set.
3637 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3639 * config/arm/arm-cpus.in (armv8_4): New feature.
3640 (ARMv8_4a): New fgroup.
3641 (armv8.4-a): New arch.
3642 * config/arm/arm-tables.opt: Regenerate.
3643 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3644 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3645 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3646 Add matching rules for -march=armv8.4-a and extensions.
3647 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3649 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3652 * config/rx/rx.md (BW): New mode attribute.
3653 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3655 2018-01-11 Richard Biener <rguenther@suse.de>
3657 PR tree-optimization/83435
3658 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3659 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3660 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3662 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3663 Alan Hayward <alan.hayward@arm.com>
3664 David Sherwood <david.sherwood@arm.com>
3666 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3668 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3669 (aarch64_print_address_internal): Use it to check for a zero offset.
3671 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3672 Alan Hayward <alan.hayward@arm.com>
3673 David Sherwood <david.sherwood@arm.com>
3675 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3676 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3677 Return a poly_int64 rather than a HOST_WIDE_INT.
3678 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3679 rather than a HOST_WIDE_INT.
3680 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3681 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3682 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3683 final_offset from HOST_WIDE_INT to poly_int64.
3684 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3685 to_constant when getting the number of units in an Advanced SIMD
3687 (aarch64_builtin_vectorized_function): Check for a constant number
3689 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3691 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3692 attribute instead of GET_MODE_NUNITS.
3693 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3694 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3695 GET_MODE_SIZE for fixed-size registers.
3696 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3697 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3698 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3699 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3700 (aarch64_print_operand, aarch64_print_address_internal)
3701 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3702 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3703 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3704 Handle polynomial GET_MODE_SIZE.
3705 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3706 wider than SImode without modification.
3707 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3708 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3709 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3710 passing and returning SVE modes.
3711 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3712 rather than GEN_INT.
3713 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3714 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3715 (aarch64_allocate_and_probe_stack_space): Likewise.
3716 (aarch64_layout_frame): Cope with polynomial offsets.
3717 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3718 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3720 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3721 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3722 poly_int64 rather than a HOST_WIDE_INT.
3723 (aarch64_get_separate_components, aarch64_process_components)
3724 (aarch64_expand_prologue, aarch64_expand_epilogue)
3725 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3726 (aarch64_anchor_offset): New function, split out from...
3727 (aarch64_legitimize_address): ...here.
3728 (aarch64_builtin_vectorization_cost): Handle polynomial
3729 TYPE_VECTOR_SUBPARTS.
3730 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3732 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3733 number of elements from the PARALLEL rather than the mode.
3734 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3735 rather than GET_MODE_BITSIZE.
3736 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3737 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3738 (aarch64_expand_vec_perm_const_1): Handle polynomial
3739 d->perm.length () and d->perm elements.
3740 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3741 Apply to_constant to d->perm elements.
3742 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3743 polynomial CONST_VECTOR_NUNITS.
3744 (aarch64_move_pointer): Take amount as a poly_int64 rather
3746 (aarch64_progress_pointer): Avoid temporary variable.
3747 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3748 the mode attribute instead of GET_MODE.
3750 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3751 Alan Hayward <alan.hayward@arm.com>
3752 David Sherwood <david.sherwood@arm.com>
3754 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3755 x exists before using it.
3756 (aarch64_add_constant_internal): Rename to...
3757 (aarch64_add_offset_1): ...this. Replace regnum with separate
3758 src and dest rtxes. Handle the case in which they're different,
3759 including when the offset is zero. Replace scratchreg with an rtx.
3760 Use 2 additions if there is no spare register into which we can
3761 move a 16-bit constant.
3762 (aarch64_add_constant): Delete.
3763 (aarch64_add_offset): Replace reg with separate src and dest
3764 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3765 Use aarch64_add_offset_1.
3766 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3767 an rtx rather than an int. Take the delta as a poly_int64
3768 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3769 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3770 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3771 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3772 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3774 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3775 aarch64_add_constant.
3777 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3779 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3780 Use scalar_float_mode.
3782 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3784 * config/aarch64/aarch64-simd.md
3785 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3786 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3787 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3788 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3789 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3790 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3791 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3792 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3793 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3794 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3796 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3799 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3800 targ_options->x_arm_arch_string is non NULL.
3802 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3804 * config/aarch64/aarch64.h
3805 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3807 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3810 * expmed.c (emit_store_flag_force): Swap if const op0
3811 and change VOIDmode to mode of op0.
3813 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3815 PR rtl-optimization/83761
3816 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3817 than bytes to mode_for_size.
3819 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3822 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3823 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3826 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3829 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3830 when in layout mode.
3831 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3832 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3835 2018-01-10 Michael Collison <michael.collison@arm.com>
3837 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3838 * config/aarch64/aarch64-option-extension.def: Add
3839 AARCH64_OPT_EXTENSION of 'fp16fml'.
3840 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3841 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3842 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3843 * config/aarch64/constraints.md (Ui7): New constraint.
3844 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3845 (VFMLA_SEL_W): Ditto.
3848 (VFMLA16_LOW): New int iterator.
3849 (VFMLA16_HIGH): Ditto.
3850 (UNSPEC_FMLAL): New unspec.
3851 (UNSPEC_FMLSL): Ditto.
3852 (UNSPEC_FMLAL2): Ditto.
3853 (UNSPEC_FMLSL2): Ditto.
3854 (f16mac): New code attribute.
3855 * config/aarch64/aarch64-simd-builtins.def
3856 (aarch64_fmlal_lowv2sf): Ditto.
3857 (aarch64_fmlsl_lowv2sf): Ditto.
3858 (aarch64_fmlalq_lowv4sf): Ditto.
3859 (aarch64_fmlslq_lowv4sf): Ditto.
3860 (aarch64_fmlal_highv2sf): Ditto.
3861 (aarch64_fmlsl_highv2sf): Ditto.
3862 (aarch64_fmlalq_highv4sf): Ditto.
3863 (aarch64_fmlslq_highv4sf): Ditto.
3864 (aarch64_fmlal_lane_lowv2sf): Ditto.
3865 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3866 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3867 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3868 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3869 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3870 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3871 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3872 (aarch64_fmlal_lane_highv2sf): Ditto.
3873 (aarch64_fmlsl_lane_highv2sf): Ditto.
3874 (aarch64_fmlal_laneq_highv2sf): Ditto.
3875 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3876 (aarch64_fmlalq_lane_highv4sf): Ditto.
3877 (aarch64_fmlsl_lane_highv4sf): Ditto.
3878 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3879 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3880 * config/aarch64/aarch64-simd.md:
3881 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3882 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3883 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3884 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3885 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3886 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3887 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3888 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3889 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3890 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3891 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3892 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3893 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3894 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3895 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3896 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3897 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3898 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3899 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3900 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3901 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3902 (vfmlsl_low_u32): Ditto.
3903 (vfmlalq_low_u32): Ditto.
3904 (vfmlslq_low_u32): Ditto.
3905 (vfmlal_high_u32): Ditto.
3906 (vfmlsl_high_u32): Ditto.
3907 (vfmlalq_high_u32): Ditto.
3908 (vfmlslq_high_u32): Ditto.
3909 (vfmlal_lane_low_u32): Ditto.
3910 (vfmlsl_lane_low_u32): Ditto.
3911 (vfmlal_laneq_low_u32): Ditto.
3912 (vfmlsl_laneq_low_u32): Ditto.
3913 (vfmlalq_lane_low_u32): Ditto.
3914 (vfmlslq_lane_low_u32): Ditto.
3915 (vfmlalq_laneq_low_u32): Ditto.
3916 (vfmlslq_laneq_low_u32): Ditto.
3917 (vfmlal_lane_high_u32): Ditto.
3918 (vfmlsl_lane_high_u32): Ditto.
3919 (vfmlal_laneq_high_u32): Ditto.
3920 (vfmlsl_laneq_high_u32): Ditto.
3921 (vfmlalq_lane_high_u32): Ditto.
3922 (vfmlslq_lane_high_u32): Ditto.
3923 (vfmlalq_laneq_high_u32): Ditto.
3924 (vfmlslq_laneq_high_u32): Ditto.
3925 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3926 (AARCH64_FL_FOR_ARCH8_4): New.
3927 (AARCH64_ISA_F16FML): New ISA flag.
3928 (TARGET_F16FML): New feature flag for fp16fml.
3929 (doc/invoke.texi): Document new fp16fml option.
3931 2018-01-10 Michael Collison <michael.collison@arm.com>
3933 * config/aarch64/aarch64-builtins.c:
3934 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3935 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3936 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3937 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3938 (AARCH64_ISA_SHA3): New ISA flag.
3939 (TARGET_SHA3): New feature flag for sha3.
3940 * config/aarch64/iterators.md (sha512_op): New int attribute.
3941 (CRYPTO_SHA512): New int iterator.
3942 (UNSPEC_SHA512H): New unspec.
3943 (UNSPEC_SHA512H2): Ditto.
3944 (UNSPEC_SHA512SU0): Ditto.
3945 (UNSPEC_SHA512SU1): Ditto.
3946 * config/aarch64/aarch64-simd-builtins.def
3947 (aarch64_crypto_sha512hqv2di): New builtin.
3948 (aarch64_crypto_sha512h2qv2di): Ditto.
3949 (aarch64_crypto_sha512su0qv2di): Ditto.
3950 (aarch64_crypto_sha512su1qv2di): Ditto.
3951 (aarch64_eor3qv8hi): Ditto.
3952 (aarch64_rax1qv2di): Ditto.
3953 (aarch64_xarqv2di): Ditto.
3954 (aarch64_bcaxqv8hi): Ditto.
3955 * config/aarch64/aarch64-simd.md:
3956 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3957 (aarch64_crypto_sha512su0qv2di): Ditto.
3958 (aarch64_crypto_sha512su1qv2di): Ditto.
3959 (aarch64_eor3qv8hi): Ditto.
3960 (aarch64_rax1qv2di): Ditto.
3961 (aarch64_xarqv2di): Ditto.
3962 (aarch64_bcaxqv8hi): Ditto.
3963 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3964 (vsha512h2q_u64): Ditto.
3965 (vsha512su0q_u64): Ditto.
3966 (vsha512su1q_u64): Ditto.
3967 (veor3q_u16): Ditto.
3968 (vrax1q_u64): Ditto.
3970 (vbcaxq_u16): Ditto.
3971 * config/arm/types.md (crypto_sha512): New type attribute.
3972 (crypto_sha3): Ditto.
3973 (doc/invoke.texi): Document new sha3 option.
3975 2018-01-10 Michael Collison <michael.collison@arm.com>
3977 * config/aarch64/aarch64-builtins.c:
3978 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3979 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3980 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3981 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3982 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3983 (AARCH64_ISA_SM4): New ISA flag.
3984 (TARGET_SM4): New feature flag for sm4.
3985 * config/aarch64/aarch64-simd-builtins.def
3986 (aarch64_sm3ss1qv4si): Ditto.
3987 (aarch64_sm3tt1aq4si): Ditto.
3988 (aarch64_sm3tt1bq4si): Ditto.
3989 (aarch64_sm3tt2aq4si): Ditto.
3990 (aarch64_sm3tt2bq4si): Ditto.
3991 (aarch64_sm3partw1qv4si): Ditto.
3992 (aarch64_sm3partw2qv4si): Ditto.
3993 (aarch64_sm4eqv4si): Ditto.
3994 (aarch64_sm4ekeyqv4si): Ditto.
3995 * config/aarch64/aarch64-simd.md:
3996 (aarch64_sm3ss1qv4si): Ditto.
3997 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3998 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3999 (aarch64_sm4eqv4si): Ditto.
4000 (aarch64_sm4ekeyqv4si): Ditto.
4001 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4002 (sm3part_op): Ditto.
4003 (CRYPTO_SM3TT): Ditto.
4004 (CRYPTO_SM3PART): Ditto.
4005 (UNSPEC_SM3SS1): New unspec.
4006 (UNSPEC_SM3TT1A): Ditto.
4007 (UNSPEC_SM3TT1B): Ditto.
4008 (UNSPEC_SM3TT2A): Ditto.
4009 (UNSPEC_SM3TT2B): Ditto.
4010 (UNSPEC_SM3PARTW1): Ditto.
4011 (UNSPEC_SM3PARTW2): Ditto.
4012 (UNSPEC_SM4E): Ditto.
4013 (UNSPEC_SM4EKEY): Ditto.
4014 * config/aarch64/constraints.md (Ui2): New constraint.
4015 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4016 * config/arm/types.md (crypto_sm3): New type attribute.
4017 (crypto_sm4): Ditto.
4018 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4019 (vsm3tt1aq_u32): Ditto.
4020 (vsm3tt1bq_u32): Ditto.
4021 (vsm3tt2aq_u32): Ditto.
4022 (vsm3tt2bq_u32): Ditto.
4023 (vsm3partw1q_u32): Ditto.
4024 (vsm3partw2q_u32): Ditto.
4025 (vsm4eq_u32): Ditto.
4026 (vsm4ekeyq_u32): Ditto.
4027 (doc/invoke.texi): Document new sm4 option.
4029 2018-01-10 Michael Collison <michael.collison@arm.com>
4031 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4032 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4033 (AARCH64_FL_FOR_ARCH8_4): New.
4034 (AARCH64_FL_V8_4): New flag.
4035 (doc/invoke.texi): Document new armv8.4-a option.
4037 2018-01-10 Michael Collison <michael.collison@arm.com>
4039 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4040 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4041 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4042 * config/aarch64/aarch64-option-extension.def: Add
4043 AARCH64_OPT_EXTENSION of 'sha2'.
4044 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4045 (crypto): Disable sha2 and aes if crypto disabled.
4046 (crypto): Enable aes and sha2 if enabled.
4047 (simd): Disable sha2 and aes if simd disabled.
4048 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4050 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4051 (TARGET_SHA2): New feature flag for sha2.
4052 (TARGET_AES): New feature flag for aes.
4053 * config/aarch64/aarch64-simd.md:
4054 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4055 conditional on TARGET_AES.
4056 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4057 (aarch64_crypto_sha1hsi): Make pattern conditional
4059 (aarch64_crypto_sha1hv4si): Ditto.
4060 (aarch64_be_crypto_sha1hv4si): Ditto.
4061 (aarch64_crypto_sha1su1v4si): Ditto.
4062 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4063 (aarch64_crypto_sha1su0v4si): Ditto.
4064 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4065 (aarch64_crypto_sha256su0v4si): Ditto.
4066 (aarch64_crypto_sha256su1v4si): Ditto.
4067 (doc/invoke.texi): Document new aes and sha2 options.
4069 2018-01-10 Martin Sebor <msebor@redhat.com>
4071 PR tree-optimization/83781
4072 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4075 2018-01-11 Martin Sebor <msebor@gmail.com>
4076 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4078 PR tree-optimization/83501
4079 PR tree-optimization/81703
4081 * tree-ssa-strlen.c (get_string_cst): Rename...
4082 (get_string_len): ...to this. Handle global constants.
4083 (handle_char_store): Adjust.
4085 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4086 Jim Wilson <jimw@sifive.com>
4088 * config/riscv/riscv-protos.h (riscv_output_return): New.
4089 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4090 (riscv_attribute_table, riscv_output_return),
4091 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4092 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4093 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4095 (riscv_expand_prologue): Add early return for naked function.
4096 (riscv_expand_epilogue): Likewise.
4097 (riscv_function_ok_for_sibcall): Return false for naked function.
4098 (riscv_set_current_function): New.
4099 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4100 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4101 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4102 * doc/extend.texi (RISC-V Function Attributes): New.
4104 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4106 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4107 check for 128-bit long double before checking TCmode.
4108 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4109 128-bit long doubles before checking TFmode or TCmode.
4110 (FLOAT128_IBM_P): Likewise.
4112 2018-01-10 Martin Sebor <msebor@redhat.com>
4114 PR tree-optimization/83671
4115 * builtins.c (c_strlen): Unconditionally return zero for the empty
4117 Use -Warray-bounds for warnings.
4118 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4119 for non-constant array indices with COMPONENT_REF, arrays of
4120 arrays, and pointers to arrays.
4121 (gimple_fold_builtin_strlen): Determine and set length range for
4122 non-constant character arrays.
4124 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4127 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4130 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4132 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4134 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4137 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4138 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4139 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4140 indexed_or_indirect_operand predicate.
4141 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4142 (*vsx_le_perm_load_v8hi): Likewise.
4143 (*vsx_le_perm_load_v16qi): Likewise.
4144 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4145 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4146 (*vsx_le_perm_store_v8hi): Likewise.
4147 (*vsx_le_perm_store_v16qi): Likewise.
4148 (eight unnamed splitters): Likewise.
4150 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4152 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4153 * config/rs6000/emmintrin.h: Likewise.
4154 * config/rs6000/mmintrin.h: Likewise.
4155 * config/rs6000/xmmintrin.h: Likewise.
4157 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4160 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4162 * tree.c (tree_nop_conversion): Return true for location wrapper
4164 (maybe_wrap_with_location): New function.
4165 (selftest::check_strip_nops): New function.
4166 (selftest::test_location_wrappers): New function.
4167 (selftest::tree_c_tests): Call it.
4168 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4169 (maybe_wrap_with_location): New decl.
4170 (EXPR_LOCATION_WRAPPER_P): New macro.
4171 (location_wrapper_p): New inline function.
4172 (tree_strip_any_location_wrapper): New inline function.
4174 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4177 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4178 stack_realign_offset for the largest alignment of stack slot
4180 (ix86_find_max_used_stack_alignment): New function.
4181 (ix86_finalize_stack_frame_flags): Use it. Set
4182 max_used_stack_alignment if we don't realign stack.
4183 * config/i386/i386.h (machine_function): Add
4184 max_used_stack_alignment.
4186 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4188 * config/arm/arm.opt (-mbranch-cost): New option.
4189 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4192 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4195 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4196 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4198 2018-01-10 Richard Biener <rguenther@suse.de>
4201 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4202 early out so it also covers the case where we have a non-NULL
4205 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4207 PR tree-optimization/83753
4208 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4209 for non-strided grouped accesses if the number of elements is 1.
4211 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4214 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4215 * i386.h (TARGET_USE_GATHER): Define.
4216 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4218 2018-01-10 Martin Liska <mliska@suse.cz>
4221 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4222 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4224 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4225 CLEANUP_NO_PARTITIONING is not set.
4227 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4229 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4230 for vectors, as a partial revert of r254296.
4231 * rtl.h (const_vec_p): Delete.
4232 (const_vec_duplicate_p): Don't test for vector CONSTs.
4233 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4234 * expmed.c (make_tree): Likewise.
4237 * common.md (E, F): Use CONSTANT_P instead of checking for
4239 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4240 checking for CONST_VECTOR.
4242 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4245 * predict.c (force_edge_cold): Handle in more sane way edges
4248 2018-01-09 Carl Love <cel@us.ibm.com>
4250 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4252 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4253 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4254 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4255 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4256 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4257 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4258 * config/rs6000/rs6000-protos.h: Add extern defition for
4259 rs6000_generate_float2_double_code.
4260 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4262 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4263 (float2_v2df): Add define_expand.
4265 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4268 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4269 op_mode in the force_to_mode call.
4271 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4273 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4274 instead of checking each element individually.
4275 (aarch64_evpc_uzp): Likewise.
4276 (aarch64_evpc_zip): Likewise.
4277 (aarch64_evpc_ext): Likewise.
4278 (aarch64_evpc_rev): Likewise.
4279 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4280 instead of checking each element individually. Return true without
4282 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4283 whether all selected elements come from the same input, instead of
4284 checking each element individually. Remove calls to gen_rtx_REG,
4285 start_sequence and end_sequence and instead assert that no rtl is
4288 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4290 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4291 order of HIGH and CONST checks.
4293 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4295 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4296 if the destination isn't an SSA_NAME.
4298 2018-01-09 Richard Biener <rguenther@suse.de>
4300 PR tree-optimization/83668
4301 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4303 (canonicalize_loop_form): ... here, renamed from ...
4304 (canonicalize_loop_closed_ssa_form): ... this and amended to
4305 swap successor edges for loop exit blocks to make us use
4306 the RPO order we need for initial schedule generation.
4308 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4310 PR tree-optimization/64811
4311 * match.pd: When optimizing comparisons with Inf, avoid
4312 introducing or losing exceptions from comparisons with NaN.
4314 2018-01-09 Martin Liska <mliska@suse.cz>
4317 * asan.c (shadow_mem_size): Add gcc_assert.
4319 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4321 Don't save registers in main().
4324 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4325 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4326 * config/avr/avr.c (avr_set_current_function): Don't error if
4327 naked, OS_task or OS_main are specified at the same time.
4328 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4330 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4332 * common/config/avr/avr-common.c (avr_option_optimization_table):
4333 Switch on -mmain-is-OS_task for optimizing compilations.
4335 2018-01-09 Richard Biener <rguenther@suse.de>
4337 PR tree-optimization/83572
4338 * graphite.c: Include cfganal.h.
4339 (graphite_transform_loops): Connect infinite loops to exit
4340 and remove fake edges at the end.
4342 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4344 * ipa-inline.c (edge_badness): Revert accidental checkin.
4346 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4349 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4350 symbols; not inline clones.
4352 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4355 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4356 hard registers. Formatting fixes.
4358 PR preprocessor/83722
4359 * gcc.c (try_generate_repro): Pass
4360 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4361 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4364 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4365 Kito Cheng <kito.cheng@gmail.com>
4367 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4368 (riscv_leaf_function_p): Delete.
4369 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4371 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4373 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4375 (do_ifelse): New function.
4376 (do_isel): New function.
4377 (do_sub3): New function.
4378 (do_add3): New function.
4379 (do_load_mask_compare): New function.
4380 (do_overlap_load_compare): New function.
4381 (expand_compare_loop): New function.
4382 (expand_block_compare): Call expand_compare_loop() when appropriate.
4383 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4385 (-mblock-compare-inline-loop-limit): New option.
4387 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4390 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4391 Reverse order of second and third operands in first alternative.
4392 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4393 of first and second elements in UNSPEC_VPERMR vector.
4394 (altivec_expand_vec_perm_le): Likewise.
4396 2018-01-08 Jeff Law <law@redhat.com>
4398 PR rtl-optimizatin/81308
4399 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4400 (process_switch): If group_case_labels makes a change, then set
4402 (pass_convert_switch::execute): If a switch is converted, then
4403 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4405 PR rtl-optimization/81308
4406 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4409 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4411 PR target/83663 - Revert r255946
4412 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4413 generation for cases where splatting a value is not useful.
4414 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4415 across a vec_duplicate and a paradoxical subreg forming a vector
4416 mode to a vec_concat.
4418 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4420 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4421 -march=armv8.3-a variants.
4422 * config/arm/t-multilib: Likewise.
4423 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4425 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4427 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4429 (cceq_ior_compare_complement): Give it a name so I can use it, and
4430 change boolean_or_operator predicate to boolean_operator so it can
4431 be used to generate a crand.
4432 (eqne): New code iterator.
4433 (bd/bd_neg): New code_attrs.
4434 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4435 a single define_insn.
4436 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4437 decrement (bdnzt/bdnzf/bdzt/bdzf).
4438 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4439 with the new names of the branch decrement patterns, and added the
4440 names of the branch decrement conditional patterns.
4442 2018-01-08 Richard Biener <rguenther@suse.de>
4444 PR tree-optimization/83563
4445 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4448 2018-01-08 Richard Biener <rguenther@suse.de>
4451 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4453 2018-01-08 Richard Biener <rguenther@suse.de>
4455 PR tree-optimization/83685
4456 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4457 references to abnormals.
4459 2018-01-08 Richard Biener <rguenther@suse.de>
4462 * dwarf2out.c (output_indirect_strings): Handle empty
4463 skeleton_debug_str_hash.
4464 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4466 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4468 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4469 (emit_store_direct): Likewise.
4470 (arc_trampoline_adjust_address): Likewise.
4471 (arc_asm_trampoline_template): New function.
4472 (arc_initialize_trampoline): Use asm_trampoline_template.
4473 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4474 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4475 * config/arc/arc.md (flush_icache): Delete pattern.
4477 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4479 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4480 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4483 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4486 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4487 by not USED_FOR_TARGET.
4488 (make_pass_resolve_sw_modes): Likewise.
4490 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4492 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4495 2018-01-08 Richard Biener <rguenther@suse.de>
4498 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4500 2018-01-08 Richard Biener <rguenther@suse.de>
4503 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4505 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4508 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4509 basic blocks with a small number of successors.
4510 (convert_control_dep_chain_into_preds): Improve handling of
4512 (dump_predicates): Split apart into...
4513 (dump_pred_chain): ...here...
4514 (dump_pred_info): ...and here.
4515 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4516 (can_chain_union_be_invalidated_p): Improve check for invalidation
4518 (uninit_uses_cannot_happen): Avoid unnecessary if
4519 convert_control_dep_chain_into_preds yielded nothing.
4521 2018-01-06 Martin Sebor <msebor@redhat.com>
4523 PR tree-optimization/83640
4524 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4525 subtracting negative offset from size.
4526 (builtin_access::overlap): Adjust offset bounds of the access to fall
4527 within the size of the object if possible.
4529 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4531 PR rtl-optimization/83699
4532 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4533 extract_bit_field_as_subreg to cases in which the extracted
4534 value is also a vector.
4536 * lra-constraints.c (process_alt_operands): Test for the equivalence
4537 substitutions when detecting a possible reload cycle.
4539 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4542 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4543 by default if flag_selective_schedling{,2}. Formatting fixes.
4545 PR rtl-optimization/83682
4546 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4547 if it has non-VECTOR_MODE element mode.
4548 (vec_duplicate_p): Likewise.
4551 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4552 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4554 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4557 * config/i386/i386-builtin.def
4558 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4559 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4560 Require also OPTION_MASK_ISA_AVX512F in addition to
4561 OPTION_MASK_ISA_GFNI.
4562 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4563 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4564 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4565 to OPTION_MASK_ISA_GFNI.
4566 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4567 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4568 OPTION_MASK_ISA_AVX512BW.
4569 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4570 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4571 addition to OPTION_MASK_ISA_GFNI.
4572 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4573 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4574 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4575 to OPTION_MASK_ISA_GFNI.
4576 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4577 a requirement for all ISAs rather than any of them with a few
4579 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4581 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4582 bitmasks to be enabled with 3 exceptions, instead of requiring any
4583 enabled ISA with lots of exceptions.
4584 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4585 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4586 Change avx512bw in isa attribute to avx512f.
4587 * config/i386/sgxintrin.h: Add license boilerplate.
4588 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4589 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4590 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4591 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4593 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4594 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4595 temporarily sse2 rather than sse if not enabled already.
4598 * config/i386/sse.md (VI248_VLBW): Rename to ...
4599 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4600 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4601 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4602 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4603 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4604 mode iterator instead of VI248_VLBW.
4606 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4608 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4609 (record_modified): Skip clobbers; add debug output.
4610 (param_change_prob): Use sreal frequencies.
4612 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4614 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4615 punt for user-aligned variables.
4617 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4619 * tree-chrec.c (chrec_contains_symbols): Return true for
4622 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4625 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4626 of (x|y) == x for BICS pattern.
4628 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4630 PR tree-optimization/83605
4631 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4632 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4635 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4637 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4638 * config/epiphany/rtems.h: New file.
4640 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4641 Uros Bizjak <ubizjak@gmail.com>
4644 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4645 QIreg_operand instead of register_operand predicate.
4646 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4647 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4648 comments instead of -fmitigate[-_]rop.
4650 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4653 * cgraphunit.c (symbol_table::compile): Switch to text_section
4654 before calling assembly_start debug hook.
4655 * run-rtl-passes.c (run_rtl_passes): Likewise.
4658 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4660 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4661 range_int_cst_p rather than !symbolic_range_p before calling
4662 extract_range_from_multiplicative_op_1.
4664 2018-01-04 Jeff Law <law@redhat.com>
4666 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4667 redundant test in assertion.
4669 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4671 * doc/rtl.texi: Document machine_mode wrapper classes.
4673 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4675 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4678 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4680 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4681 the VEC_PERM_EXPR fold to fail.
4683 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4686 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4687 to switched_sections.
4689 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4692 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4695 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4698 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4699 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4701 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4704 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4705 is BLKmode and bitpos not zero or mode change is needed.
4707 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4710 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4713 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4716 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4717 instead of MULT rtx. Update all corresponding splitters.
4719 (*ssub<modesuffix>): Ditto.
4721 (*cmp_sadd_di): Update split patterns.
4722 (*cmp_sadd_si): Ditto.
4723 (*cmp_sadd_sidi): Ditto.
4724 (*cmp_ssub_di): Ditto.
4725 (*cmp_ssub_si): Ditto.
4726 (*cmp_ssub_sidi): Ditto.
4727 * config/alpha/predicates.md (const23_operand): New predicate.
4728 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4729 Look for ASHIFT, not MULT inner operand.
4730 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4732 2018-01-04 Martin Liska <mliska@suse.cz>
4734 PR gcov-profile/83669
4735 * gcov.c (output_intermediate_file): Add version to intermediate
4737 * doc/gcov.texi: Document new field 'version' in intermediate
4738 file format. Fix location of '-k' option of gcov command.
4740 2018-01-04 Martin Liska <mliska@suse.cz>
4743 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4745 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4747 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4749 2018-01-03 Martin Sebor <msebor@redhat.com>
4751 PR tree-optimization/83655
4752 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4753 checking calls with invalid arguments.
4755 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4757 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4758 (vectorizable_mask_load_store): Delete.
4759 (vectorizable_call): Return false for masked loads and stores.
4760 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4761 instead of gimple_assign_rhs1.
4762 (vectorizable_load): Handle IFN_MASK_LOAD.
4763 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4765 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4767 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4769 (vectorizable_mask_load_store): ...here.
4770 (vectorizable_load): ...and here.
4772 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4774 * tree-vect-stmts.c (vect_build_all_ones_mask)
4775 (vect_build_zero_merge_argument): New functions, split out from...
4776 (vectorizable_load): ...here.
4778 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4780 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4782 (vectorizable_mask_load_store): ...here.
4783 (vectorizable_store): ...and here.
4785 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4787 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4789 (vectorizable_mask_load_store): ...here.
4791 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4793 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4794 (vect_model_store_cost): Take a vec_load_store_type instead of a
4796 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4797 (vect_model_store_cost): Take a vec_load_store_type instead of a
4799 (vectorizable_mask_load_store): Update accordingly.
4800 (vectorizable_store): Likewise.
4801 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4803 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4805 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4806 IFN_MASK_LOAD calls here rather than...
4807 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4809 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4810 Alan Hayward <alan.hayward@arm.com>
4811 David Sherwood <david.sherwood@arm.com>
4813 * expmed.c (extract_bit_field_1): For vector extracts,
4814 fall back to extract_bit_field_as_subreg if vec_extract
4817 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4818 Alan Hayward <alan.hayward@arm.com>
4819 David Sherwood <david.sherwood@arm.com>
4821 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4822 they are variable or constant sized.
4823 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4824 slots for constant-sized data.
4826 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4827 Alan Hayward <alan.hayward@arm.com>
4828 David Sherwood <david.sherwood@arm.com>
4830 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4831 handling COND_EXPRs with boolean comparisons, try to find a better
4832 basis for the mask type than the boolean itself.
4834 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4836 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4837 is calculated and how it can be overridden.
4838 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4839 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4841 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4844 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4845 Alan Hayward <alan.hayward@arm.com>
4846 David Sherwood <david.sherwood@arm.com>
4848 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4849 Remove the mode argument.
4850 (aarch64_simd_valid_immediate): Remove the mode and inverse
4852 * config/aarch64/iterators.md (bitsize): New iterator.
4853 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4854 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4855 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4856 aarch64_simd_valid_immediate.
4857 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4858 (aarch64_reg_or_bic_imm): Likewise.
4859 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4860 with an insn_type enum and msl with a modifier_type enum.
4861 Replace element_width with a scalar_mode. Change the shift
4862 to unsigned int. Add constructors for scalar_float_mode and
4863 scalar_int_mode elements.
4864 (aarch64_vect_float_const_representable_p): Delete.
4865 (aarch64_can_const_movi_rtx_p)
4866 (aarch64_simd_scalar_immediate_valid_for_move)
4867 (aarch64_simd_make_constant): Update call to
4868 aarch64_simd_valid_immediate.
4869 (aarch64_advsimd_valid_immediate_hs): New function.
4870 (aarch64_advsimd_valid_immediate): Likewise.
4871 (aarch64_simd_valid_immediate): Remove mode and inverse
4872 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4873 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4874 and aarch64_float_const_representable_p on the result.
4875 (aarch64_output_simd_mov_immediate): Remove mode argument.
4876 Update call to aarch64_simd_valid_immediate and use of
4877 simd_immediate_info.
4878 (aarch64_output_scalar_simd_mov_immediate): Update call
4881 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4882 Alan Hayward <alan.hayward@arm.com>
4883 David Sherwood <david.sherwood@arm.com>
4885 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4886 (mode_nunits): Likewise CONST_MODE_NUNITS.
4887 * machmode.def (ADJUST_NUNITS): Document.
4888 * genmodes.c (mode_data::need_nunits_adj): New field.
4889 (blank_mode): Update accordingly.
4890 (adj_nunits): New variable.
4891 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4893 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4894 listed in adj_nunits.
4895 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4896 listed in adj_nunits. Don't emit case statements for such modes.
4897 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4898 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4899 nothing if adj_nunits is nonnull.
4900 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4901 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4902 (emit_mode_fbit): Update use of print_maybe_const_decl.
4903 (emit_move_size): Likewise. Treat the array as non-const
4905 (emit_mode_adjustments): Handle adj_nunits.
4907 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4909 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4910 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4911 (VECTOR_MODES): Use it.
4912 (make_vector_modes): Take the prefix as an argument.
4914 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4915 Alan Hayward <alan.hayward@arm.com>
4916 David Sherwood <david.sherwood@arm.com>
4918 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4919 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4920 for MODE_VECTOR_BOOL.
4921 * machmode.def (VECTOR_BOOL_MODE): Document.
4922 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4923 (make_vector_bool_mode): New function.
4924 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4926 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4927 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4929 * stor-layout.c (int_mode_for_mode): Likewise.
4930 * tree.c (build_vector_type_for_mode): Likewise.
4931 * varasm.c (output_constant_pool_2): Likewise.
4932 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4933 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4934 for MODE_VECTOR_BOOL.
4935 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4936 of mode class checks.
4937 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4938 instead of a list of mode class checks.
4939 (expand_vector_scalar_condition): Likewise.
4940 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4942 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4943 Alan Hayward <alan.hayward@arm.com>
4944 David Sherwood <david.sherwood@arm.com>
4946 * machmode.h (mode_size): Change from unsigned short to
4948 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4949 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4950 or if measurement_type is not polynomial.
4951 (fixed_size_mode::includes_p): Check for constant-sized modes.
4952 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4953 return a poly_uint16 rather than an unsigned short.
4954 (emit_mode_size): Change the type of mode_size from unsigned short
4955 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4956 (emit_mode_adjustments): Cope with polynomial vector sizes.
4957 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4959 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4961 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4962 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4963 * caller-save.c (setup_save_areas): Likewise.
4964 (replace_reg_with_saved_mem): Likewise.
4965 * calls.c (emit_library_call_value_1): Likewise.
4966 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4967 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4968 (gen_lowpart_for_combine): Likewise.
4969 * convert.c (convert_to_integer_1): Likewise.
4970 * cse.c (equiv_constant, cse_insn): Likewise.
4971 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4972 (cselib_subst_to_values): Likewise.
4973 * dce.c (word_dce_process_block): Likewise.
4974 * df-problems.c (df_word_lr_mark_ref): Likewise.
4975 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4976 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4977 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4978 (rtl_for_decl_location): Likewise.
4979 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4980 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4981 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4982 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4983 (expand_expr_real_1): Likewise.
4984 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4985 (pad_below): Likewise.
4986 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4987 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4988 * ira.c (get_subreg_tracking_sizes): Likewise.
4989 * ira-build.c (ira_create_allocno_objects): Likewise.
4990 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4991 (ira_sort_regnos_for_alter_reg): Likewise.
4992 * ira-costs.c (record_operand_costs): Likewise.
4993 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4994 (resolve_simple_move): Likewise.
4995 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4996 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4997 (lra_constraints): Likewise.
4998 (CONST_POOL_OK_P): Reject variable-sized modes.
4999 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5000 (add_pseudo_to_slot, lra_spill): Likewise.
5001 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5002 * optabs-query.c (get_best_extraction_insn): Likewise.
5003 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5004 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5005 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5006 * recog.c (offsettable_address_addr_space_p): Likewise.
5007 * regcprop.c (maybe_mode_change): Likewise.
5008 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5009 * regrename.c (build_def_use): Likewise.
5010 * regstat.c (dump_reg_info): Likewise.
5011 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5012 (find_reloads, find_reloads_subreg_address): Likewise.
5013 * reload1.c (eliminate_regs_1): Likewise.
5014 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5015 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5016 (simplify_binary_operation_1, simplify_subreg): Likewise.
5017 * targhooks.c (default_function_arg_padding): Likewise.
5018 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5019 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5020 (verify_gimple_assign_ternary): Likewise.
5021 * tree-inline.c (estimate_move_cost): Likewise.
5022 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5023 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5024 (get_address_cost_ainc): Likewise.
5025 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5026 (vect_supportable_dr_alignment): Likewise.
5027 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5028 (vectorizable_reduction): Likewise.
5029 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5030 (vectorizable_operation, vectorizable_load): Likewise.
5031 * tree.c (build_same_sized_truth_vector_type): Likewise.
5032 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5033 * var-tracking.c (emit_note_insn_var_location): Likewise.
5034 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5035 (ADDR_VEC_ALIGN): Likewise.
5037 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5038 Alan Hayward <alan.hayward@arm.com>
5039 David Sherwood <david.sherwood@arm.com>
5041 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5043 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5044 or if measurement_type is polynomial.
5045 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5046 * combine.c (make_extraction): Likewise.
5047 * dse.c (find_shift_sequence): Likewise.
5048 * dwarf2out.c (mem_loc_descriptor): Likewise.
5049 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5050 (extract_bit_field, extract_low_bits): Likewise.
5051 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5052 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5053 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5054 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5055 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5056 * reload.c (find_reloads): Likewise.
5057 * reload1.c (alter_reg): Likewise.
5058 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5059 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5060 * tree-if-conv.c (predicate_mem_writes): Likewise.
5061 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5062 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5063 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5064 * valtrack.c (dead_debug_insert_temp): Likewise.
5065 * varasm.c (mergeable_constant_section): Likewise.
5066 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5068 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5069 Alan Hayward <alan.hayward@arm.com>
5070 David Sherwood <david.sherwood@arm.com>
5072 * expr.c (expand_assignment): Cope with polynomial mode sizes
5073 when assigning to a CONCAT.
5075 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5076 Alan Hayward <alan.hayward@arm.com>
5077 David Sherwood <david.sherwood@arm.com>
5079 * machmode.h (mode_precision): Change from unsigned short to
5081 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5083 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5084 or if measurement_type is not polynomial.
5085 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5086 in which the mode is already known to be a scalar_int_mode.
5087 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5088 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5090 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5091 for GET_MODE_PRECISION.
5092 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5093 for GET_MODE_PRECISION.
5094 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5096 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5097 (expand_field_assignment, make_extraction): Likewise.
5098 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5099 (get_last_value): Likewise.
5100 * convert.c (convert_to_integer_1): Likewise.
5101 * cse.c (cse_insn): Likewise.
5102 * expr.c (expand_expr_real_1): Likewise.
5103 * lra-constraints.c (simplify_operand_subreg): Likewise.
5104 * optabs-query.c (can_atomic_load_p): Likewise.
5105 * optabs.c (expand_atomic_load): Likewise.
5106 (expand_atomic_store): Likewise.
5107 * ree.c (combine_reaching_defs): Likewise.
5108 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5109 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5110 * tree.h (type_has_mode_precision_p): Likewise.
5111 * ubsan.c (instrument_si_overflow): Likewise.
5113 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5114 Alan Hayward <alan.hayward@arm.com>
5115 David Sherwood <david.sherwood@arm.com>
5117 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5118 polynomial numbers of units.
5119 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5120 (valid_vector_subparts_p): New function.
5121 (build_vector_type): Remove temporary shim and take the number
5122 of units as a poly_uint64 rather than an int.
5123 (build_opaque_vector_type): Take the number of units as a
5124 poly_uint64 rather than an int.
5125 * tree.c (build_vector_from_ctor): Handle polynomial
5126 TYPE_VECTOR_SUBPARTS.
5127 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5128 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5129 (build_vector_from_val): If the number of units is variable,
5130 use build_vec_duplicate_cst for constant operands and
5131 VEC_DUPLICATE_EXPR otherwise.
5132 (make_vector_type): Remove temporary is_constant ().
5133 (build_vector_type, build_opaque_vector_type): Take the number of
5134 units as a poly_uint64 rather than an int.
5135 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5137 * cfgexpand.c (expand_debug_expr): Likewise.
5138 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5139 (store_constructor, expand_expr_real_1): Likewise.
5140 (const_scalar_mask_from_tree): Likewise.
5141 * fold-const-call.c (fold_const_reduction): Likewise.
5142 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5143 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5144 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5145 (fold_relational_const): Likewise.
5146 (native_interpret_vector): Likewise. Change the size from an
5147 int to an unsigned int.
5148 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5149 TYPE_VECTOR_SUBPARTS.
5150 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5151 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5152 duplicating a non-constant operand into a variable-length vector.
5153 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5154 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5155 * ipa-icf.c (sem_variable::equals): Likewise.
5156 * match.pd: Likewise.
5157 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5158 * print-tree.c (print_node): Likewise.
5159 * stor-layout.c (layout_type): Likewise.
5160 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5161 * tree-cfg.c (verify_gimple_comparison): Likewise.
5162 (verify_gimple_assign_binary): Likewise.
5163 (verify_gimple_assign_ternary): Likewise.
5164 (verify_gimple_assign_single): Likewise.
5165 * tree-pretty-print.c (dump_generic_node): Likewise.
5166 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5167 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5168 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5169 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5170 (vect_shift_permute_load_chain): Likewise.
5171 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5172 (expand_vector_condition, optimize_vector_constructor): Likewise.
5173 (lower_vec_perm, get_compute_type): Likewise.
5174 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5175 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5176 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5177 (vect_recog_mask_conversion_pattern): Likewise.
5178 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5179 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5180 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5181 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5182 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5183 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5184 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5185 (supportable_widening_operation): Likewise.
5186 (supportable_narrowing_operation): Likewise.
5187 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5189 * varasm.c (output_constant): Likewise.
5191 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5192 Alan Hayward <alan.hayward@arm.com>
5193 David Sherwood <david.sherwood@arm.com>
5195 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5196 so that both the length == 3 and length != 3 cases set up their
5197 own permute vectors. Add comments explaining why we know the
5198 number of elements is constant.
5199 (vect_permute_load_chain): Likewise.
5201 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5202 Alan Hayward <alan.hayward@arm.com>
5203 David Sherwood <david.sherwood@arm.com>
5205 * machmode.h (mode_nunits): Change from unsigned char to
5207 (ONLY_FIXED_SIZE_MODES): New macro.
5208 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5209 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5210 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5212 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5213 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5214 or if measurement_type is not polynomial.
5215 * genmodes.c (ZERO_COEFFS): New macro.
5216 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5218 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5219 Use ZERO_COEFFS when emitting initializers.
5220 * data-streamer.h (bp_pack_poly_value): New function.
5221 (bp_unpack_poly_value): Likewise.
5222 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5223 for GET_MODE_NUNITS.
5224 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5225 for GET_MODE_NUNITS.
5226 * tree.c (make_vector_type): Remove temporary shim and make
5227 the real function take the number of units as a poly_uint64
5229 (build_vector_type_for_mode): Handle polynomial nunits.
5230 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5231 * emit-rtl.c (const_vec_series_p_1): Likewise.
5232 (gen_rtx_CONST_VECTOR): Likewise.
5233 * fold-const.c (test_vec_duplicate_folding): Likewise.
5234 * genrecog.c (validate_pattern): Likewise.
5235 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5236 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5237 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5238 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5239 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5240 * rtlanal.c (subreg_get_info): Likewise.
5241 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5242 (vect_grouped_load_supported): Likewise.
5243 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5244 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5245 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5246 (simplify_const_unary_operation, simplify_binary_operation_1)
5247 (simplify_const_binary_operation, simplify_ternary_operation)
5248 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5249 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5250 instead of CONST_VECTOR_NUNITS.
5251 * varasm.c (output_constant_pool_2): Likewise.
5252 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5253 explicit-encoded elements in the XVEC for variable-length vectors.
5255 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5257 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5259 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5260 Alan Hayward <alan.hayward@arm.com>
5261 David Sherwood <david.sherwood@arm.com>
5263 * coretypes.h (fixed_size_mode): Declare.
5264 (fixed_size_mode_pod): New typedef.
5265 * builtins.h (target_builtins::x_apply_args_mode)
5266 (target_builtins::x_apply_result_mode): Change type to
5267 fixed_size_mode_pod.
5268 * builtins.c (apply_args_size, apply_result_size, result_vector)
5269 (expand_builtin_apply_args_1, expand_builtin_apply)
5270 (expand_builtin_return): Update accordingly.
5272 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5274 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5275 * cselib.c (cselib_hash_rtx): Likewise.
5276 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5277 CONST_VECTOR encoding.
5279 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5280 Jeff Law <law@redhat.com>
5283 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5284 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5285 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5286 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5289 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5290 explicitly probe *sp in a noreturn function if there were any callee
5291 register saves or frame pointer is needed.
5293 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5296 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5297 BLKmode for ternary, binary or unary expressions.
5300 * var-tracking.c (delete_vta_debug_insn): New inline function.
5301 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5302 insns from get_insns () to NULL instead of each bb separately.
5303 Use delete_vta_debug_insn. No longer static.
5304 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5305 delete_vta_debug_insns callers.
5306 * rtl.h (delete_vta_debug_insns): Declare.
5307 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5308 instead of variable_tracking_main.
5310 2018-01-03 Martin Sebor <msebor@redhat.com>
5312 PR tree-optimization/83603
5313 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5314 arguments past the endof the argument list in functions declared
5315 without a prototype.
5316 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5317 Avoid checking when arguments are null.
5319 2018-01-03 Martin Sebor <msebor@redhat.com>
5322 * doc/extend.texi (attribute const): Fix a typo.
5323 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5324 issuing -Wsuggest-attribute for void functions.
5326 2018-01-03 Martin Sebor <msebor@redhat.com>
5328 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5329 offset_int::from instead of wide_int::to_shwi.
5330 (maybe_diag_overlap): Remove assertion.
5331 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5332 * gimple-ssa-sprintf.c (format_directive): Same.
5333 (parse_directive): Same.
5334 (sprintf_dom_walker::compute_format_length): Same.
5335 (try_substitute_return_value): Same.
5337 2018-01-03 Jeff Law <law@redhat.com>
5340 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5341 non-constant residual for zero at runtime and avoid probing in
5342 that case. Reorganize code for trailing problem to mirror handling
5345 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5347 PR tree-optimization/83501
5348 * tree-ssa-strlen.c (get_string_cst): New.
5349 (handle_char_store): Call get_string_cst.
5351 2018-01-03 Martin Liska <mliska@suse.cz>
5353 PR tree-optimization/83593
5354 * tree-ssa-strlen.c: Include tree-cfg.h.
5355 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5356 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5357 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5359 (strlen_dom_walker::before_dom_children): Call
5360 gimple_purge_dead_eh_edges. Dump tranformation with details
5362 (strlen_dom_walker::before_dom_children): Update call by adding
5363 new argument cleanup_eh.
5364 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5366 2018-01-03 Martin Liska <mliska@suse.cz>
5369 * cif-code.def (VARIADIC_THUNK): New enum value.
5370 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5373 2018-01-03 Jan Beulich <jbeulich@suse.com>
5375 * sse.md (mov<mode>_internal): Tighten condition for when to use
5376 vmovdqu<ssescalarsize> for TI and OI modes.
5378 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5380 Update copyright years.
5382 2018-01-03 Martin Liska <mliska@suse.cz>
5385 * ipa-visibility.c (function_and_variable_visibility): Skip
5386 functions with noipa attribure.
5388 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5390 * gcc.c (process_command): Update copyright notice dates.
5391 * gcov-dump.c (print_version): Ditto.
5392 * gcov.c (print_version): Ditto.
5393 * gcov-tool.c (print_version): Ditto.
5394 * gengtype.c (create_file): Ditto.
5395 * doc/cpp.texi: Bump @copying's copyright year.
5396 * doc/cppinternals.texi: Ditto.
5397 * doc/gcc.texi: Ditto.
5398 * doc/gccint.texi: Ditto.
5399 * doc/gcov.texi: Ditto.
5400 * doc/install.texi: Ditto.
5401 * doc/invoke.texi: Ditto.
5403 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5405 * vector-builder.h (vector_builder::m_full_nelts): Change from
5406 unsigned int to poly_uint64.
5407 (vector_builder::full_nelts): Update prototype accordingly.
5408 (vector_builder::new_vector): Likewise.
5409 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5410 (vector_builder::operator ==): Likewise.
5411 (vector_builder::finalize): Likewise.
5412 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5413 Take the number of elements as a poly_uint64 rather than an
5415 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5416 from unsigned int to poly_uint64.
5417 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5418 (vec_perm_indices::new_vector): Likewise.
5419 (vec_perm_indices::length): Likewise.
5420 (vec_perm_indices::nelts_per_input): Likewise.
5421 (vec_perm_indices::input_nelts): Likewise.
5422 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5423 number of elements per input as a poly_uint64 rather than an
5424 unsigned int. Use the original encoding for variable-length
5425 vectors, rather than clamping each individual element.
5426 For the second and subsequent elements in each pattern,
5427 clamp the step and base before clamping their sum.
5428 (vec_perm_indices::series_p): Handle polynomial element counts.
5429 (vec_perm_indices::all_in_range_p): Likewise.
5430 (vec_perm_indices_to_tree): Likewise.
5431 (vec_perm_indices_to_rtx): Likewise.
5432 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5433 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5434 (tree_vector_builder::new_binary_operation): Handle polynomial
5435 element counts. Return false if we need to know the number
5436 of elements at compile time.
5437 * fold-const.c (fold_vec_perm): Punt if the number of elements
5438 isn't known at compile time.
5440 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5442 * vec-perm-indices.h (vec_perm_builder): Change element type
5443 from HOST_WIDE_INT to poly_int64.
5444 (vec_perm_indices::element_type): Update accordingly.
5445 (vec_perm_indices::clamp): Handle polynomial element_types.
5446 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5447 (vec_perm_indices::all_in_range_p): Likewise.
5448 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5450 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5451 polynomial vec_perm_indices element types.
5452 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5453 * fold-const.c (fold_vec_perm): Likewise.
5454 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5455 * tree-vect-generic.c (lower_vec_perm): Likewise.
5456 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5457 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5458 element type to HOST_WIDE_INT.
5460 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5461 Alan Hayward <alan.hayward@arm.com>
5462 David Sherwood <david.sherwood@arm.com>
5464 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5465 rather than an int. Use plus_constant.
5466 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5467 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5469 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5470 Alan Hayward <alan.hayward@arm.com>
5471 David Sherwood <david.sherwood@arm.com>
5473 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5474 a HOST_WIDE_INT to a poly_int64.
5476 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5477 Alan Hayward <alan.hayward@arm.com>
5478 David Sherwood <david.sherwood@arm.com>
5480 * calls.c (load_register_parameters): Cope with polynomial
5481 mode sizes. Require a constant size for BLKmode parameters
5482 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5483 forces a parameter to be padded at the lsb end in order to
5484 fill a complete number of words, require the parameter size
5485 to be ordered wrt UNITS_PER_WORD.
5487 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5488 Alan Hayward <alan.hayward@arm.com>
5489 David Sherwood <david.sherwood@arm.com>
5491 * reload1.c (spill_stack_slot_width): Change element type
5492 from unsigned int to poly_uint64_pod.
5493 (alter_reg): Treat mode sizes as polynomial.
5495 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5496 Alan Hayward <alan.hayward@arm.com>
5497 David Sherwood <david.sherwood@arm.com>
5499 * reload.c (complex_word_subreg_p): New function.
5500 (reload_inner_reg_of_subreg, push_reload): Use it.
5502 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5503 Alan Hayward <alan.hayward@arm.com>
5504 David Sherwood <david.sherwood@arm.com>
5506 * lra-constraints.c (process_alt_operands): Reject matched
5507 operands whose sizes aren't ordered.
5508 (match_reload): Refer to this check here.
5510 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5511 Alan Hayward <alan.hayward@arm.com>
5512 David Sherwood <david.sherwood@arm.com>
5514 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5515 that the mode size is in the set {1, 2, 4, 8, 16}.
5517 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5518 Alan Hayward <alan.hayward@arm.com>
5519 David Sherwood <david.sherwood@arm.com>
5521 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5522 Use plus_constant instead of gen_rtx_PLUS.
5524 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5525 Alan Hayward <alan.hayward@arm.com>
5526 David Sherwood <david.sherwood@arm.com>
5528 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5529 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5530 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5531 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5532 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5533 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5534 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5535 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5536 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5537 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5539 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5540 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5541 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5542 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5543 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5544 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5545 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5546 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5547 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5548 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5550 * expr.c (emit_move_resolve_push): Treat the input and result
5551 of PUSH_ROUNDING as a poly_int64.
5552 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5553 (emit_push_insn): Likewise.
5554 * lra-eliminations.c (mark_not_eliminable): Likewise.
5555 * recog.c (push_operand): Likewise.
5556 * reload1.c (elimination_effects): Likewise.
5557 * rtlanal.c (nonzero_bits1): Likewise.
5558 * calls.c (store_one_arg): Likewise. Require the padding to be
5559 known at compile time.
5561 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5562 Alan Hayward <alan.hayward@arm.com>
5563 David Sherwood <david.sherwood@arm.com>
5565 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5566 Use plus_constant instead of gen_rtx_PLUS.
5568 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5569 Alan Hayward <alan.hayward@arm.com>
5570 David Sherwood <david.sherwood@arm.com>
5572 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5575 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5576 Alan Hayward <alan.hayward@arm.com>
5577 David Sherwood <david.sherwood@arm.com>
5579 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5580 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5581 via stack temporaries. Treat the mode size as polynomial too.
5583 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5584 Alan Hayward <alan.hayward@arm.com>
5585 David Sherwood <david.sherwood@arm.com>
5587 * expr.c (expand_expr_real_2): When handling conversions involving
5588 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5589 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5590 as a poly_uint64 too.
5592 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5593 Alan Hayward <alan.hayward@arm.com>
5594 David Sherwood <david.sherwood@arm.com>
5596 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5598 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5599 Alan Hayward <alan.hayward@arm.com>
5600 David Sherwood <david.sherwood@arm.com>
5602 * combine.c (can_change_dest_mode): Handle polynomial
5603 REGMODE_NATURAL_SIZE.
5604 * expmed.c (store_bit_field_1): Likewise.
5605 * expr.c (store_constructor): Likewise.
5606 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5607 and polynomial REGMODE_NATURAL_SIZE.
5608 (gen_lowpart_common): Likewise.
5609 * reginfo.c (record_subregs_of_mode): Likewise.
5610 * rtlanal.c (read_modify_subreg_p): Likewise.
5612 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5613 Alan Hayward <alan.hayward@arm.com>
5614 David Sherwood <david.sherwood@arm.com>
5616 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5617 numbers of elements.
5619 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5620 Alan Hayward <alan.hayward@arm.com>
5621 David Sherwood <david.sherwood@arm.com>
5623 * match.pd: Cope with polynomial numbers of vector elements.
5625 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5626 Alan Hayward <alan.hayward@arm.com>
5627 David Sherwood <david.sherwood@arm.com>
5629 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5630 in a POINTER_PLUS_EXPR.
5632 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5633 Alan Hayward <alan.hayward@arm.com>
5634 David Sherwood <david.sherwood@arm.com>
5636 * omp-simd-clone.c (simd_clone_subparts): New function.
5637 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5638 (ipa_simd_modify_function_body): Likewise.
5640 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5641 Alan Hayward <alan.hayward@arm.com>
5642 David Sherwood <david.sherwood@arm.com>
5644 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5645 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5646 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5647 (expand_vector_condition, vector_element): Likewise.
5648 (subparts_gt): New function.
5649 (get_compute_type): Use subparts_gt.
5650 (count_type_subparts): Delete.
5651 (expand_vector_operations_1): Use subparts_gt instead of
5652 count_type_subparts.
5654 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5655 Alan Hayward <alan.hayward@arm.com>
5656 David Sherwood <david.sherwood@arm.com>
5658 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5659 (vect_compile_time_alias): ...this new function. Do the calculation
5660 on poly_ints rather than trees.
5661 (vect_prune_runtime_alias_test_list): Update call accordingly.
5663 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5664 Alan Hayward <alan.hayward@arm.com>
5665 David Sherwood <david.sherwood@arm.com>
5667 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5669 (vect_schedule_slp_instance): Likewise.
5671 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5672 Alan Hayward <alan.hayward@arm.com>
5673 David Sherwood <david.sherwood@arm.com>
5675 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5676 constant and extern definitions for variable-length vectors.
5677 (vect_get_constant_vectors): Note that the number of units
5678 is known to be constant.
5680 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5681 Alan Hayward <alan.hayward@arm.com>
5682 David Sherwood <david.sherwood@arm.com>
5684 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5685 of units as polynomial. Choose between WIDE and NARROW based
5688 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5689 Alan Hayward <alan.hayward@arm.com>
5690 David Sherwood <david.sherwood@arm.com>
5692 * tree-vect-stmts.c (simd_clone_subparts): New function.
5693 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5695 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5696 Alan Hayward <alan.hayward@arm.com>
5697 David Sherwood <david.sherwood@arm.com>
5699 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5700 vectors as polynomial. Use build_index_vector for
5703 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5704 Alan Hayward <alan.hayward@arm.com>
5705 David Sherwood <david.sherwood@arm.com>
5707 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5708 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5709 for variable-length vectors.
5710 (vectorizable_mask_load_store): Treat the number of units as
5711 polynomial, asserting that it is constant if the condition has
5712 already been enforced.
5713 (vectorizable_store, vectorizable_load): Likewise.
5715 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5716 Alan Hayward <alan.hayward@arm.com>
5717 David Sherwood <david.sherwood@arm.com>
5719 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5720 of units as polynomial. Punt if we can't tell at compile time
5721 which vector contains the final result.
5723 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5724 Alan Hayward <alan.hayward@arm.com>
5725 David Sherwood <david.sherwood@arm.com>
5727 * tree-vect-loop.c (vectorizable_induction): Treat the number
5728 of units as polynomial. Punt on SLP inductions. Use an integer
5729 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5730 cast of such a series for variable-length floating-point
5733 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5734 Alan Hayward <alan.hayward@arm.com>
5735 David Sherwood <david.sherwood@arm.com>
5737 * tree.h (build_index_vector): Declare.
5738 * tree.c (build_index_vector): New function.
5739 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5740 of units as polynomial, forcibly converting it to a constant if
5741 vectorizable_reduction has already enforced the condition.
5742 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5743 to create a {1,2,3,...} vector.
5744 (vectorizable_reduction): Treat the number of units as polynomial.
5745 Choose vectype_in based on the largest scalar element size rather
5746 than the smallest number of units. Enforce the restrictions
5749 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5750 Alan Hayward <alan.hayward@arm.com>
5751 David Sherwood <david.sherwood@arm.com>
5753 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5754 number of units as polynomial.
5756 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5757 Alan Hayward <alan.hayward@arm.com>
5758 David Sherwood <david.sherwood@arm.com>
5760 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5761 * target.def (autovectorize_vector_sizes): Return the vector sizes
5762 by pointer, using vector_sizes rather than a bitmask.
5763 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5764 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5765 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5767 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5768 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5769 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5770 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5771 * omp-general.c (omp_max_vf): Likewise.
5772 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5773 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5774 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5775 * tree-vect-slp.c (vect_slp_bb): Likewise.
5776 * doc/tm.texi: Regenerate.
5777 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5779 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5780 the vector size as a poly_uint64 rather than an unsigned int.
5781 (current_vector_size): Change from an unsigned int to a poly_uint64.
5782 (get_vectype_for_scalar_type): Update accordingly.
5783 * tree.h (build_truth_vector_type): Take the size and number of
5784 units as a poly_uint64 rather than an unsigned int.
5785 (build_vector_type): Add a temporary overload that takes
5786 the number of units as a poly_uint64 rather than an unsigned int.
5787 * tree.c (make_vector_type): Likewise.
5788 (build_truth_vector_type): Take the number of units as a poly_uint64
5789 rather than an unsigned int.
5791 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5792 Alan Hayward <alan.hayward@arm.com>
5793 David Sherwood <david.sherwood@arm.com>
5795 * target.def (get_mask_mode): Take the number of units and length
5796 as poly_uint64s rather than unsigned ints.
5797 * targhooks.h (default_get_mask_mode): Update accordingly.
5798 * targhooks.c (default_get_mask_mode): Likewise.
5799 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5800 * doc/tm.texi: Regenerate.
5802 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5803 Alan Hayward <alan.hayward@arm.com>
5804 David Sherwood <david.sherwood@arm.com>
5806 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5807 * omp-general.c (omp_max_vf): Likewise.
5808 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5809 (expand_omp_simd): Handle polynomial safelen.
5810 * omp-low.c (omplow_simd_context): Add a default constructor.
5811 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5812 (lower_rec_simd_input_clauses): Update accordingly.
5813 (lower_rec_input_clauses): Likewise.
5815 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5816 Alan Hayward <alan.hayward@arm.com>
5817 David Sherwood <david.sherwood@arm.com>
5819 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5820 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5821 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5822 (vect_analyze_slp_cost): Likewise.
5823 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5824 (vect_model_load_cost): Likewise.
5826 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5827 Alan Hayward <alan.hayward@arm.com>
5828 David Sherwood <david.sherwood@arm.com>
5830 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5831 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5832 from an unsigned int * to a poly_uint64_pod *.
5833 (calculate_unrolling_factor): New function.
5834 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5836 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5837 Alan Hayward <alan.hayward@arm.com>
5838 David Sherwood <david.sherwood@arm.com>
5840 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5841 from an unsigned int to a poly_uint64.
5842 (_loop_vec_info::slp_unrolling_factor): Likewise.
5843 (_loop_vec_info::vectorization_factor): Change from an int
5845 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5846 (vect_get_num_vectors): New function.
5847 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5848 (vect_get_num_copies): Use vect_get_num_vectors.
5849 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5850 to an unsigned int *.
5851 (vect_analyze_data_refs): Change min_vf from an int * to a
5853 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5854 than an unsigned HOST_WIDE_INT.
5855 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5856 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5857 to an unsigned int *.
5858 (vect_analyze_data_ref_dependences): Likewise.
5859 (vect_compute_data_ref_alignment): Handle polynomial vf.
5860 (vect_enhance_data_refs_alignment): Likewise.
5861 (vect_prune_runtime_alias_test_list): Likewise.
5862 (vect_shift_permute_load_chain): Likewise.
5863 (vect_supportable_dr_alignment): Likewise.
5864 (dependence_distance_ge_vf): Take the vectorization factor as a
5865 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5866 (vect_analyze_data_refs): Change min_vf from an int * to a
5868 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5869 vfm1 as a poly_uint64 rather than an int. Make the same change
5870 for the returned bound_scalar.
5871 (vect_gen_vector_loop_niters): Handle polynomial vf.
5872 (vect_do_peeling): Likewise. Update call to
5873 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5874 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5876 * tree-vect-loop.c (vect_determine_vectorization_factor)
5877 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5878 (vect_get_known_peeling_cost): Likewise.
5879 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5880 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5881 (vect_transform_loop): Likewise. Use the lowest possible VF when
5882 updating the upper bounds of the loop.
5883 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5885 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5886 polynomial unroll factors.
5887 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5888 (vect_make_slp_decision): Likewise.
5889 (vect_supported_load_permutation_p): Likewise, and polynomial
5891 (vect_analyze_slp_cost): Handle polynomial vf.
5892 (vect_slp_analyze_node_operations): Likewise.
5893 (vect_slp_analyze_bb_1): Likewise.
5894 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5895 than an unsigned HOST_WIDE_INT.
5896 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5897 (vectorizable_load): Handle polynomial vf.
5898 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5900 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5902 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5903 Alan Hayward <alan.hayward@arm.com>
5904 David Sherwood <david.sherwood@arm.com>
5906 * match.pd: Handle bit operations involving three constants
5907 and try to fold one pair.
5909 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5911 * tree-vect-loop-manip.c: Include gimple-fold.h.
5912 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5913 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5914 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5915 Add a path that uses a step of VF instead of 1, but disable it
5917 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5918 and niters_no_overflow parameters. Update calls to
5919 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5920 Create a new SSA name if the latter choses to use a ste other
5921 than zero, and return it via niters_vector_mult_vf_var.
5922 * tree-vect-loop.c (vect_transform_loop): Update calls to
5923 vect_do_peeling, vect_gen_vector_loop_niters and
5924 slpeel_make_loop_iterate_ntimes.
5925 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5926 (vect_gen_vector_loop_niters): Update declarations after above changes.
5928 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5930 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5931 128-bit round to integer instructions.
5932 (ceil<mode>2): Likewise.
5933 (btrunc<mode>2): Likewise.
5934 (round<mode>2): Likewise.
5936 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5938 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5939 unaligned VSX load/store on P8/P9.
5940 (expand_block_clear): Allow the use of unaligned VSX
5941 load/store on P8/P9.
5943 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5945 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5947 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5948 swap associated with both a load and a store.
5950 2018-01-02 Andrew Waterman <andrew@sifive.com>
5952 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5953 * config/riscv/riscv.md (clear_cache): Use it.
5955 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5957 * web.c: Remove out-of-date comment.
5959 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5961 * expr.c (fixup_args_size_notes): Check that any existing
5962 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5963 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5964 (emit_single_push_insn): ...here.
5966 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5968 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5969 (const_vector_encoded_nelts): New function.
5970 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5971 (const_vector_int_elt, const_vector_elt): Declare.
5972 * emit-rtl.c (const_vector_int_elt_1): New function.
5973 (const_vector_elt): Likewise.
5974 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5975 of CONST_VECTOR_ELT.
5977 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5979 * expr.c: Include rtx-vector-builder.h.
5980 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5981 directly on the tree encoding.
5982 (const_vector_from_tree): Likewise.
5983 * optabs.c: Include rtx-vector-builder.h.
5984 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5985 sequence of "u" values.
5986 * vec-perm-indices.c: Include rtx-vector-builder.h.
5987 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5988 directly on the vec_perm_indices encoding.
5990 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5992 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5993 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5994 * rtx-vector-builder.h: New file.
5995 * rtx-vector-builder.c: Likewise.
5996 * rtl.h (rtx_def::u2): Add a const_vector field.
5997 (CONST_VECTOR_NPATTERNS): New macro.
5998 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5999 (CONST_VECTOR_DUPLICATE_P): Likewise.
6000 (CONST_VECTOR_STEPPED_P): Likewise.
6001 (CONST_VECTOR_ENCODED_ELT): Likewise.
6002 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6003 (unwrap_const_vec_duplicate): Likewise.
6004 (const_vec_series_p): Check for a non-duplicated vector encoding.
6005 Say that the function only returns true for integer vectors.
6006 * emit-rtl.c: Include rtx-vector-builder.h.
6007 (gen_const_vec_duplicate_1): Delete.
6008 (gen_const_vector): Call gen_const_vec_duplicate instead of
6009 gen_const_vec_duplicate_1.
6010 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6011 (gen_const_vec_duplicate): Use rtx_vector_builder.
6012 (gen_const_vec_series): Likewise.
6013 (gen_rtx_CONST_VECTOR): Likewise.
6014 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6015 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6016 Build a new vector rather than modifying a CONST_VECTOR in-place.
6017 (handle_special_swappables): Update call accordingly.
6018 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6019 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6020 Build a new vector rather than modifying a CONST_VECTOR in-place.
6021 (handle_special_swappables): Update call accordingly.
6023 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6025 * simplify-rtx.c (simplify_const_binary_operation): Use
6026 CONST_VECTOR_ELT instead of XVECEXP.
6028 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6030 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6031 the selector elements to be different from the data elements
6032 if the selector is a VECTOR_CST.
6033 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6034 ssizetype for the selector.
6036 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6038 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6039 before testing each element individually.
6040 * tree-vect-generic.c (lower_vec_perm): Likewise.
6042 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6044 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6045 * selftest-run-tests.c (selftest::run_tests): Call it.
6046 * vector-builder.h (vector_builder::operator ==): New function.
6047 (vector_builder::operator !=): Likewise.
6048 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6049 (vec_perm_indices::all_from_input_p): New function.
6050 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6051 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6052 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6053 instead of reading the VECTOR_CST directly. Detect whether both
6054 vector inputs are the same before constructing the vec_perm_indices,
6055 and update the number of inputs argument accordingly. Use the
6056 utility functions added above. Only construct sel2 if we need to.
6058 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6060 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6061 the broadcast of the low byte.
6062 (expand_mult_highpart): Use an explicit encoding for the permutes.
6063 * optabs-query.c (can_mult_highpart_p): Likewise.
6064 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6065 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6066 (vectorizable_bswap): Likewise.
6067 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6068 explicit encoding for the power-of-2 permutes.
6069 (vect_permute_store_chain): Likewise.
6070 (vect_grouped_load_supported): Likewise.
6071 (vect_permute_load_chain): Likewise.
6073 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6075 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6076 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6077 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6078 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6079 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6080 (vect_gen_perm_mask_any): Likewise.
6082 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6084 * int-vector-builder.h: New file.
6085 * vec-perm-indices.h: Include int-vector-builder.h.
6086 (vec_perm_indices): Redefine as an int_vector_builder.
6087 (auto_vec_perm_indices): Delete.
6088 (vec_perm_builder): Redefine as a stand-alone class.
6089 (vec_perm_indices::vec_perm_indices): New function.
6090 (vec_perm_indices::clamp): Likewise.
6091 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6092 (vec_perm_indices::new_vector): New function.
6093 (vec_perm_indices::new_expanded_vector): Update for new
6094 vec_perm_indices class.
6095 (vec_perm_indices::rotate_inputs): New function.
6096 (vec_perm_indices::all_in_range_p): Operate directly on the
6097 encoded form, without computing elided elements.
6098 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6099 encoding. Update for new vec_perm_indices class.
6100 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6101 the given vec_perm_builder.
6102 (expand_vec_perm_var): Update vec_perm_builder constructor.
6103 (expand_mult_highpart): Use vec_perm_builder instead of
6104 auto_vec_perm_indices.
6105 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6106 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6107 or double series encoding as appropriate.
6108 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6109 vec_perm_indices instead of auto_vec_perm_indices.
6110 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6111 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6112 (vect_permute_store_chain): Likewise.
6113 (vect_grouped_load_supported): Likewise.
6114 (vect_permute_load_chain): Likewise.
6115 (vect_shift_permute_load_chain): Likewise.
6116 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6117 (vect_transform_slp_perm_load): Likewise.
6118 (vect_schedule_slp_instance): Likewise.
6119 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6120 (vectorizable_mask_load_store): Likewise.
6121 (vectorizable_bswap): Likewise.
6122 (vectorizable_store): Likewise.
6123 (vectorizable_load): Likewise.
6124 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6125 vec_perm_indices instead of auto_vec_perm_indices. Use
6126 tree_to_vec_perm_builder to read the vector from a tree.
6127 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6128 vec_perm_builder instead of a vec_perm_indices.
6129 (have_whole_vector_shift): Use vec_perm_builder and
6130 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6131 truncation to calc_vec_perm_mask_for_shift.
6132 (vect_create_epilog_for_reduction): Likewise.
6133 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6134 from auto_vec_perm_indices to vec_perm_indices.
6135 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6136 instead of changing individual elements.
6137 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6138 the vector in d.perm.
6139 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6140 from auto_vec_perm_indices to vec_perm_indices.
6141 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6142 instead of changing individual elements.
6143 (arm_vectorize_vec_perm_const): Use new_vector to install
6144 the vector in d.perm.
6145 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6146 Update vec_perm_builder constructor.
6147 (rs6000_expand_interleave): Likewise.
6148 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6149 (rs6000_expand_interleave): Likewise.
6151 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6153 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6154 to qimode could truncate the indices.
6155 * optabs.c (expand_vec_perm_var): Likewise.
6157 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6159 * Makefile.in (OBJS): Add vec-perm-indices.o.
6160 * vec-perm-indices.h: New file.
6161 * vec-perm-indices.c: Likewise.
6162 * target.h (vec_perm_indices): Replace with a forward class
6164 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6165 * optabs.h: Include vec-perm-indices.h.
6166 (expand_vec_perm): Delete.
6167 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6168 (expand_vec_perm_const): Declare.
6169 * target.def (vec_perm_const_ok): Replace with...
6170 (vec_perm_const): ...this new hook.
6171 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6172 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6173 * doc/tm.texi: Regenerate.
6174 * optabs.def (vec_perm_const): Delete.
6175 * doc/md.texi (vec_perm_const): Likewise.
6176 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6177 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6178 expand_vec_perm for constant permutation vectors. Assert that
6179 the mode of variable permutation vectors is the integer equivalent
6180 of the mode that is being permuted.
6181 * optabs-query.h (selector_fits_mode_p): Declare.
6182 * optabs-query.c: Include vec-perm-indices.h.
6183 (selector_fits_mode_p): New function.
6184 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6185 is defined, instead of checking whether the vec_perm_const_optab
6186 exists. Use targetm.vectorize.vec_perm_const instead of
6187 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6188 fit in the vector mode before using a variable permute.
6189 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6190 vec_perm_indices instead of an rtx.
6191 (expand_vec_perm): Replace with...
6192 (expand_vec_perm_const): ...this new function. Take the selector
6193 as a vec_perm_indices rather than an rtx. Also take the mode of
6194 the selector. Update call to shift_amt_for_vec_perm_mask.
6195 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6196 Use vec_perm_indices::new_expanded_vector to expand the original
6197 selector into bytes. Check whether the indices fit in the vector
6198 mode before using a variable permute.
6199 (expand_vec_perm_var): Make global.
6200 (expand_mult_highpart): Use expand_vec_perm_const.
6201 * fold-const.c: Includes vec-perm-indices.h.
6202 * tree-ssa-forwprop.c: Likewise.
6203 * tree-vect-data-refs.c: Likewise.
6204 * tree-vect-generic.c: Likewise.
6205 * tree-vect-loop.c: Likewise.
6206 * tree-vect-slp.c: Likewise.
6207 * tree-vect-stmts.c: Likewise.
6208 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6210 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6211 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6212 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6213 (aarch64_vectorize_vec_perm_const): ...this new function.
6214 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6215 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6216 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6217 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6218 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6219 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6220 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6222 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6223 check for NEON modes.
6224 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6225 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6226 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6227 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6229 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6230 the old VEC_PERM_CONST conditions.
6231 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6232 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6233 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6234 (ia64_vectorize_vec_perm_const_ok): Merge into...
6235 (ia64_vectorize_vec_perm_const): ...this new function.
6236 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6237 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6238 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6239 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6240 * config/mips/mips.c (mips_expand_vec_perm_const)
6241 (mips_vectorize_vec_perm_const_ok): Merge into...
6242 (mips_vectorize_vec_perm_const): ...this new function.
6243 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6244 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6245 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6246 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6247 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6248 (rs6000_expand_vec_perm_const): Delete.
6249 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6251 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6252 (altivec_expand_vec_perm_const_le): Take each operand individually.
6253 Operate on constant selectors rather than rtxes.
6254 (altivec_expand_vec_perm_const): Likewise. Update call to
6255 altivec_expand_vec_perm_const_le.
6256 (rs6000_expand_vec_perm_const): Delete.
6257 (rs6000_vectorize_vec_perm_const_ok): Delete.
6258 (rs6000_vectorize_vec_perm_const): New function.
6259 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6260 an element count and rtx array.
6261 (rs6000_expand_extract_even): Update call accordingly.
6262 (rs6000_expand_interleave): Likewise.
6263 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6264 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6265 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6266 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6267 (rs6000_expand_vec_perm_const): Delete.
6268 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6269 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6270 (altivec_expand_vec_perm_const_le): Take each operand individually.
6271 Operate on constant selectors rather than rtxes.
6272 (altivec_expand_vec_perm_const): Likewise. Update call to
6273 altivec_expand_vec_perm_const_le.
6274 (rs6000_expand_vec_perm_const): Delete.
6275 (rs6000_vectorize_vec_perm_const_ok): Delete.
6276 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6277 reference to the SPE evmerge intructions.
6278 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6279 an element count and rtx array.
6280 (rs6000_expand_extract_even): Update call accordingly.
6281 (rs6000_expand_interleave): Likewise.
6282 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6283 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6285 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6287 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6289 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6290 vector mode and that that mode matches the mode of the data
6292 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6293 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6294 directly using expand_vec_perm_1 when forcing selectors into
6296 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6298 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6300 * optabs-query.h (can_vec_perm_p): Delete.
6301 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6302 * optabs-query.c (can_vec_perm_p): Split into...
6303 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6304 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6305 particular selector is valid.
6306 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6307 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6308 (vect_grouped_load_supported): Likewise.
6309 (vect_shift_permute_load_chain): Likewise.
6310 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6311 (vect_transform_slp_perm_load): Likewise.
6312 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6313 (vectorizable_bswap): Likewise.
6314 (vect_gen_perm_mask_checked): Likewise.
6315 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6316 implementations of variable permutation vectors into account
6317 when deciding which selector to use.
6318 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6319 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6320 with a false third argument.
6321 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6322 to test whether the constant selector is valid and can_vec_perm_var_p
6323 to test whether a variable selector is valid.
6325 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6327 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6328 * optabs-query.c (can_vec_perm_p): Likewise.
6329 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6330 instead of vec_perm_indices.
6331 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6332 (vect_gen_perm_mask_checked): Likewise,
6333 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6334 (vect_gen_perm_mask_checked): Likewise,
6336 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6338 * optabs-query.h (qimode_for_vec_perm): Declare.
6339 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6340 (qimode_for_vec_perm): ...this new function.
6341 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6343 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6345 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6346 does not have a conditional at the top.
6348 2018-01-02 Richard Biener <rguenther@suse.de>
6350 * ipa-inline.c (big_speedup_p): Fix expression.
6352 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6355 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6358 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6362 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6363 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6364 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6365 cond_taken_branch_cost 3->4.
6367 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6369 PR tree-optimization/83581
6370 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6371 TODO_cleanup_cfg if any changes have been made.
6374 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6375 convert_modes if target mode has the right side, but different mode
6379 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6380 last argument when extracting from CONCAT. If either from_real or
6381 from_imag is NULL, use expansion through memory. If result is not
6382 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6383 the parts directly to inner mode, if even that fails, use expansion
6387 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6388 check for bswap in mode rather than HImode and use that in expand_unop
6391 Copyright (C) 2018 Free Software Foundation, Inc.
6393 Copying and distribution of this file, with or without modification,
6394 are permitted in any medium without royalty provided the copyright
6395 notice and this notice are preserved.