1 2018-02-08 David Malcolm <dmalcolm@redhat.com>
3 PR tree-optimization/84136
4 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
5 that the result of find_edge is non-NULL.
7 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
10 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
11 storing integer register in SImode. Fix cost of 256 and 512
12 byte aligned SSE register store.
14 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
16 * config/i386/i386.c (ix86_multiplication_cost): Fix
17 multiplication cost for TARGET_AVX512DQ.
19 2018-02-08 Marek Polacek <polacek@redhat.com>
21 PR tree-optimization/84238
22 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
25 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
27 PR tree-optimization/84265
28 * tree-vect-stmts.c (vectorizable_store): Don't treat
29 VMAT_CONTIGUOUS accesses as grouped.
30 (vectorizable_load): Likewise.
32 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
34 PR tree-optimization/81635
35 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
36 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
37 (test_round_for_mask): New functions.
38 (wide_int_cc_tests): Call test_round_for_mask.
39 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
40 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
41 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
42 range returned by get_range_info.
44 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
47 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
48 * symtab.c: Include builtins.h
49 (symtab_node::output_to_lto_symbol_table_p): Move here
50 from lto-streamer-out.c:output_symbol_p.
51 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
52 (output_symbol_p): Move all logic to symtab.c
53 (produce_symtab): Update.
55 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
57 * config/s390/s390-opts.h (enum indirect_branch): Define.
58 * config/s390/s390-protos.h (s390_return_addr_from_memory)
59 (s390_indirect_branch_via_thunk)
60 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
61 (enum s390_indirect_branch_type): Define.
62 * config/s390/s390.c (struct s390_frame_layout, struct
63 machine_function): Remove.
64 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
65 (indirect_branch_table_label_no, indirect_branch_table_name):
67 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
68 (enum s390_indirect_branch_option): Define.
69 (s390_return_addr_from_memory): New function.
70 (s390_handle_string_attribute): New function.
71 (s390_attribute_table): Add new attribute handler.
72 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
73 (s390_indirect_branch_via_thunk): New function.
74 (s390_indirect_branch_via_inline_thunk): New function.
75 (s390_function_ok_for_sibcall): When jumping via thunk disallow
76 sibling call optimization for non z10 compiles.
77 (s390_emit_call): Force indirect branch target to be a single
78 register. Add r1 clobber for non-z10 compiles.
79 (s390_emit_epilogue): Emit return jump via return_use expander.
80 (s390_reorg): Handle JUMP_INSNs as execute targets.
81 (s390_option_override_internal): Perform validity checks for the
82 new command line options.
83 (s390_indirect_branch_attrvalue): New function.
84 (s390_indirect_branch_settings): New function.
85 (s390_set_current_function): Invoke s390_indirect_branch_settings.
86 (s390_output_indirect_thunk_function): New function.
87 (s390_code_end): Implement target hook.
88 (s390_case_values_threshold): Implement target hook.
89 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
91 * config/s390/s390.h (struct s390_frame_layout)
92 (struct machine_function): Move here from s390.c.
93 (TARGET_INDIRECT_BRANCH_NOBP_RET)
94 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
95 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
96 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
97 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
98 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
99 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
100 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
101 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
102 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
103 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
104 (mnemonic attribute): Add values which aren't recognized
106 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
107 pattern for branch conversion. Fix mnemonic attribute.
108 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
109 indirect branch via thunk if requested.
110 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
111 ("*indirect_jump"): Disable for branch conversion using out of
113 ("indirect_jump_via_thunk<mode>_z10")
114 ("indirect_jump_via_thunk<mode>")
115 ("indirect_jump_via_inlinethunk<mode>_z10")
116 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
117 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
118 ("casesi_jump_via_inlinethunk<mode>_z10")
119 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
120 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
121 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
122 ("*indirect2_jump"): Disable for branch conversion.
123 ("casesi_jump"): Turn into expander and expand patterns for branch
125 ("return_use"): New expander.
126 ("*return"): Emit return via thunk and rename it to ...
127 ("*return<mode>"): ... this one.
128 * config/s390/s390.opt: Add new options and and enum for the
131 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
133 * lra-constraints.c (match_reload): Unconditionally use
134 gen_lowpart_SUBREG, rather than selecting between that
135 and equivalent gen_rtx_SUBREG code.
137 2018-02-08 Richard Biener <rguenther@suse.de>
139 PR tree-optimization/84233
140 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
141 changed flag instead of boguously re-using phi_inserted.
143 2018-02-08 Martin Jambor <mjambor@suse.cz>
145 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
146 static local variables.
148 2018-02-08 Richard Biener <rguenther@suse.de>
150 PR tree-optimization/84278
151 * tree-vect-stmts.c (vectorizable_store): When looking for
152 smaller vector types to perform grouped strided loads/stores
153 make sure the mode is supported by the target.
154 (vectorizable_load): Likewise.
156 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
158 * config/aarch64/aarch64.c (aarch64_components_for_bb):
159 Increase LDP/STP opportunities by adding adjacent callee-saves.
161 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
163 PR rtl-optimization/84068
164 PR rtl-optimization/83459
165 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
167 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
169 PR tree-optimization/84224
170 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
171 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
174 2018-02-07 Iain Sandoe <iain@codesourcery.com>
177 * config/rs6000/altivec.md (*restore_world): Remove LR use.
178 * config/rs6000/predicates.md (restore_world_operation): Adjust op
179 count, remove one USE.
181 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
183 * doc/install.texi (Configuration): Document the
184 --with-long-double-format={ibm,ieee} PowerPC configuration
188 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
189 Convert from define_expand to be define_insn_and_split. Rework
190 float/double/_Float128 conversions to QI/HI/SImode to work with
191 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
192 conversions to QI/HImode types did a store and then a load to
193 truncate the value. For conversions to VSX registers, don't split
194 the insn, instead emit the code directly. Use the code iterator
195 any_fix to combine signed and unsigned conversions.
196 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
197 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
198 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
199 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
200 (fix_<mode>di2_hw): Likewise.
201 (fixuns_<mode>di2_hw): Likewise.
202 (fix_<mode>si2_hw): Likewise.
203 (fixuns_<mode>si2_hw): Likewise.
204 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
205 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
206 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
207 fix<uns>_trunc<SFDF:mode>si2_p8.
208 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
210 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
211 (fix<uns>_<mode>_mem): Likewise.
212 (fctiw<u>z_<mode>_mem): Likewise.
213 (fix<uns>_<mode>_mem): Likewise.
214 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
215 the register allocator from doing a direct move to the GPRs to do
216 a store, and instead use the ISA 3.0 store byte/half-word from
217 vector register instruction. For IEEE 128-bit floating point,
218 also optimize stores of 32-bit ints.
219 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
221 2018-02-07 Alan Hayward <alan.hayward@arm.com>
223 * genextract.c (push_pathstr_operand): New function to support
225 (walk_rtx): Call push_pathstr_operand.
226 (print_path): Support [a-zA-Z].
228 2018-02-07 Richard Biener <rguenther@suse.de>
230 PR tree-optimization/84037
231 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
232 (cse_and_gimplify_to_preheader): Declare.
233 (vect_get_place_in_interleaving_chain): Likewise.
234 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
236 (_loop_vec_info::~_loop_vec_info): Delete it.
237 (cse_and_gimplify_to_preheader): New function.
238 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
239 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
240 (vectorizable_load): Likewise. For grouped stores always base
241 the IV on the first element.
242 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
243 condition before gimplifying.
245 2018-02-07 Jakub Jelinek <jakub@redhat.com>
247 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
248 *DIV_EXPR and *MOD_EXPR.
250 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
253 * config/i386/i386.c (ix86_option_override_internal): Mask out
254 the CF_SET bit when checking -fcf-protection.
256 2018-02-07 Tom de Vries <tom@codesourcery.com>
259 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
262 2018-02-07 Richard Biener <rguenther@suse.de>
264 PR tree-optimization/84204
265 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
268 PR tree-optimization/84205
269 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
270 special-case isl_ast_op_zdiv_r.
272 PR tree-optimization/84223
273 * graphite-scop-detection.c (gather_bbs::before_dom_children):
274 Only add conditions from within the region.
275 (gather_bbs::after_dom_children): Adjust.
277 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
280 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
281 * config/avr/avr.md: Only post-reload split REG-REG moves if
282 either register is REGERAL_REG_P.
284 2018-02-07 Jakub Jelinek <jakub@redhat.com>
286 PR tree-optimization/84235
287 * tree-ssa-scopedtables.c
288 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
289 if the subtraction is performed in floating point type where NaNs are
290 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
291 build 1. Formatting fix.
293 2018-02-06 Jakub Jelinek <jakub@redhat.com>
296 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
297 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
298 and skip it regardless of bb boundaries. Use CALL_P macro,
299 don't test INSN_P (insn) together with CALL_P or JUMP_P check
300 unnecessarily, formatting fix.
302 2018-02-06 Michael Collison <michael.collison@arm.com>
304 * config/arm/thumb2.md:
305 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
306 (*thumb_mov_notscc): Ditto.
308 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
311 * config/rs6000/rs6000.md (su code attribute): Use "u" for
312 unsigned_fix, not "s".
314 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
316 * configure.ac (gcc_fn_eh_frame_ro): New function.
317 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
318 correct .eh_frame permissions.
319 * configure: Regenerate.
321 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
323 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
326 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
328 * config/rs6000/rs6000.c (rs6000_option_override_internal):
329 Display warning message for -mno-speculate-indirect-jumps.
331 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
333 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
335 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
337 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
339 PR tree-optimization/84225
340 * tree-eh.c (find_trapping_overflow): Only call
341 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
343 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
346 * config/i386/i386.c: Reimplement the check of possible options
347 -mibt/-mshstk conbination. Change error messages.
348 * doc/invoke.texi: Fix a typo: remove extra '='.
350 2018-02-06 Marek Polacek <polacek@redhat.com>
352 PR tree-optimization/84228
353 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
355 2018-02-06 Tamar Christina <tamar.christina@arm.com>
358 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
359 emitted arch directives.
360 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
361 __ARM_FEATURE_COPROC before changing architectures.
363 2018-02-06 Richard Biener <rguenther@suse.de>
365 * config/i386/i386.c (print_reg): Fix typo.
366 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
368 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
370 * configure: Regenerate.
372 2018-02-05 Martin Sebor <msebor@redhat.com>
374 PR tree-optimization/83369
375 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
378 2018-02-05 Martin Liska <mliska@suse.cz>
380 * doc/invoke.texi: Cherry-pick upstream r323995.
382 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
384 * ira.c (ira_init_register_move_cost): Adjust comment.
386 2018-02-05 Martin Liska <mliska@suse.cz>
388 PR gcov-profile/84137
389 * doc/gcov.texi: Fix typo in documentation.
391 2018-02-05 Martin Liska <mliska@suse.cz>
393 PR gcov-profile/83879
394 * doc/gcov.texi: Document necessity of --dynamic-list-data when
395 using dlopen functionality.
397 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
399 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
400 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
401 _mm_maskz_range_ss, _mm_mask_range_round_ss,
402 _mm_maskz_range_round_ss): New intrinsics.
403 (__builtin_ia32_rangesd128_round)
404 (__builtin_ia32_rangess128_round): Remove.
405 (__builtin_ia32_rangesd128_mask_round,
406 __builtin_ia32_rangess128_mask_round): New builtins.
407 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
408 __builtin_ia32_rangess128_round): Remove.
409 (__builtin_ia32_rangesd128_mask_round,
410 __builtin_ia32_rangess128_mask_round): New builtins.
411 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
412 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
413 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
414 "<round_saeonly_constraint>")): Changed to ...
415 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
416 "<round_saeonly_scalar_constraint>")): ... this.
417 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
418 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
419 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
420 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
421 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
423 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
425 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
427 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
428 Remove all values except native, 8540 and 8548.
430 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
432 * config/i386/i386.c (ix86_output_function_return): Pass
433 INVALID_REGNUM, instead of -1, as invalid register number to
434 indirect_thunk_name and output_indirect_thunk.
436 2018-02-02 Julia Koval <julia.koval@intel.com>
438 * config.gcc: Add -march=icelake.
439 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
440 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
441 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
442 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
443 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
444 (processor_target_table): Add icelake.
445 (ix86_option_override_internal): Handle new PTAs.
446 (get_builtin_code_for_version): Handle icelake.
447 (M_INTEL_COREI7_ICELAKE): New.
448 (fold_builtin_cpu): Handle icelake.
449 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
450 * doc/invoke.texi: Add -march=icelake.
452 2018-02-02 Julia Koval <julia.koval@intel.com>
454 * config/i386/i386.c (ix86_option_override_internal): Change flags type
456 * wide-int-bitmask.h: New.
458 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
461 * config/i386/i386.md: Replace Pmode with word_mode in
462 builtin_setjmp_setup and builtin_longjmp to support x32.
464 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
468 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
470 (rs6000_supported_cpu_names): New static variable.
471 (linux_cpu_translation_table): Likewise.
472 (elf_platform) <cpu>: Define new static variable and use it.
473 Translate kernel AT_PLATFORM name to canonical name if needed.
474 Error if platform name is unknown.
476 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
479 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
481 2018-02-01 Jeff Law <law@redhat.com>
484 * config/i386/i386.c (release_scratch_register_on_entry): Add new
485 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
486 the scratch if RELEASE_VIA_POP is false.
487 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
488 If we have to save a temporary register, decrement SIZE appropriately.
489 Pass new arguments to release_scratch_register_on_entry.
490 (ix86_adjust_stack_and_probe): Likewise.
491 (ix86_emit_probe_stack_range): Pass new arguments to
492 release_scratch_register_on_entry.
494 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
496 PR rtl-optimization/84157
497 * combine.c (change_zero_ext): Use REG_P predicate in
498 front of HARD_REGISTER_P predicate.
500 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
502 * config/avr/avr.c (avr_option_override): Move disabling of
503 -fdelete-null-pointer-checks to...
504 * common/config/avr/avr-common.c (avr_option_optimization_table):
507 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
509 PR tree-optimization/81635
510 * tree-data-ref.c (split_constant_offset_1): For types that
511 wrap on overflow, try to use range info to prove that wrapping
514 2018-02-01 Renlin Li <renlin.li@arm.com>
517 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
519 (aarch64_register_move_cost): Likewise.
520 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
522 (REG_CLASS_NAMES): Likewise.
523 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
524 TAILCALL_ADDR_REGS. Remove IP registers.
525 * config/aarch64/aarch64.md (Ucs): Update register constraint.
527 2018-02-01 Richard Biener <rguenther@suse.de>
529 * domwalk.h (dom_walker::dom_walker): Add additional constructor
530 for specifying RPO order and allow NULL for that.
531 * domwalk.c (dom_walker::dom_walker): Likewise.
532 (dom_walker::walk): Handle NULL RPO order.
533 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
535 (rewrite_update_dom_walker): Likewise.
536 (mark_def_dom_walker): Likewise.
538 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
540 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
541 (aarch64_maybe_expand_sve_subreg_move): Declare.
542 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
543 * config/aarch64/predicates.md (aarch64_any_register_operand): New
545 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
546 that are semantically a reverse operation.
547 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
548 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
549 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
551 (aarch64_can_change_mode_class): For big-endian, forbid changes
552 between two SVE modes if they have different element sizes.
554 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
556 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
557 the TImode handling for big-endian targets.
559 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
561 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
562 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
564 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
565 Remove BSWAP handing for big-endian targets and use the form of
566 LD1RQ appropariate for the mode.
568 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
570 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
571 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
574 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
577 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
578 check for operands that need to go through aarch64_sve_reload_be.
580 2018-02-01 Jakub Jelinek <jakub@redhat.com>
582 PR tree-optimization/81661
583 PR tree-optimization/84117
584 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
585 * tree-eh.c: Include gimplify.h.
586 (find_trapping_overflow, replace_trapping_overflow,
587 rewrite_to_non_trapping_overflow): New functions.
588 * tree-vect-loop.c: Include tree-eh.h.
589 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
590 * tree-data-ref.c: Include tree-eh.h.
591 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
593 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
595 PR rtl-optimization/84123
596 * combine.c (change_zero_ext): Check if hard register satisfies
597 can_change_dest_mode before calling gen_lowpart_SUBREG.
599 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
602 * ira.c (ira_init_register_move_cost): Remove assert.
604 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
606 PR rtl-optimization/84071
607 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
608 * doc/tm.texi: Regenerate.
610 2018-01-31 Richard Biener <rguenther@suse.de>
612 PR tree-optimization/84132
613 * tree-data-ref.c (analyze_miv_subscript): Properly
614 check whether evolution_function_is_affine_multivariate_p
615 before calling gcd_of_steps_may_divide_p.
617 2018-01-31 Julia Koval <julia.koval@intel.com>
620 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
621 * config/i386/i386.md (rdpid_rex64) New.
622 (rdpid): Make 32bit only.
624 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
627 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
628 an IDENTIFIER_NODE for FUNCTION_TYPE's.
630 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
633 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
635 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
637 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
639 PR rtl-optimization/84071
640 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
641 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
643 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
645 * config/arc/arc.c (arc_handle_aux_attribute): New function.
646 (arc_attribute_table): Add 'aux' attribute.
647 (arc_in_small_data_p): Consider aux like variables.
648 (arc_is_aux_reg_p): New function.
649 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
650 (arc_get_aux_arg): New function.
651 (prepare_move_operands): Handle aux-register access.
652 (arc_handle_aux_attribute): New function.
653 * doc/extend.texi (ARC Variable attributes): Add subsection.
655 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
657 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
658 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
659 (arc_attribute_table): Add 'uncached' attribute.
660 (arc_print_operand): Print '.di' flag for uncached memory
662 (arc_in_small_data_p): Do not consider for small data the uncached
664 (arc_is_uncached_mem_p): New function.
665 * config/arc/predicates.md (compact_store_memory_operand): Check
666 for uncached memory accesses.
667 (nonvol_nonimm_operand): Likewise.
668 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
670 2018-01-31 Jakub Jelinek <jakub@redhat.com>
673 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
674 falign-loops=): Add Optimization flag.
676 2018-01-30 Jeff Law <law@redhat.com>
679 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
680 INT_REGISTERS_SAVED. Check it prior to calling
681 get_scratch_register_on_entry.
682 (ix86_adjust_stack_and_probe): Similarly.
683 (ix86_emit_probe_stack_range): Similarly.
684 (ix86_expand_prologue): Corresponding changes.
686 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
689 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
690 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
692 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
695 * lra-constraints.c (curr_insn_transform): Process AND in the
698 2018-01-30 Jakub Jelinek <jakub@redhat.com>
700 PR rtl-optimization/83986
701 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
702 dependence against last_pending_memory_flush in addition to
705 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
707 PR tree-optimization/81611
708 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
711 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
713 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
716 2018-01-30 Richard Biener <rguenther@suse.de>
717 Jakub Jelinek <jakub@redhat.com>
719 PR tree-optimization/84111
720 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
721 inner loops added during recursion, as they don't have up-to-date
724 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
727 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
728 (can_inline_edge_by_limits_p): ... here.
729 (can_early_inline_edge_p, check_callers,
730 update_caller_keys, update_callee_keys, recursive_inlining,
731 add_new_edges_to_heap, speculation_useful_p,
732 inline_small_functions,
733 inline_small_functions, flatten_function,
734 inline_to_all_callers_1): Update.
736 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
738 * profile-count.c (profile_count::combine_with_ipa_count): Handle
741 2018-01-30 Richard Biener <rguenther@suse.de>
743 PR tree-optimization/83008
744 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
745 invariant and constant vector uses in stmts when they need
748 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
751 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
752 * configure: Regenerate.
754 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
756 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
758 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
759 Use gen_rtx_REG rather than gen_lowpart.
761 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
763 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
764 rather than 0 when creating partial subregs.
766 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
768 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
771 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
774 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
775 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
776 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
777 flags. This restores the settings used before the 2017-07-24.
778 Turning off pre increment/decrement/modify allows IVOPTS to
779 optimize DF/SF loops where the index is an int.
781 2018-01-29 Richard Biener <rguenther@suse.de>
782 Kelvin Nilsen <kelvin@gcc.gnu.org>
785 * tree-vect-stmts.c (vectorizable_call): Don't call
786 targetm.vectorize_builtin_md_vectorized_function if callee is
789 2018-01-22 Carl Love <cel@us.ibm.com>
791 * doc/extend.tex: Fix typo in second arg in
792 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
794 2018-01-29 Richard Biener <rguenther@suse.de>
796 PR tree-optimization/84086
797 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
798 (flush_ssaname_freelist): When SSA names were released reset
801 2018-01-29 Richard Biener <rguenther@suse.de>
803 PR tree-optimization/84057
804 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
805 removed paths when removing edges.
807 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
809 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
810 -mfunction-return=@var{choice}.
812 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
815 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
817 (layout::print_source_line): Likewise.
818 (test_get_line_width_without_trailing_whitespace): Add test cases.
820 2018-01-27 Jakub Jelinek <jakub@redhat.com>
823 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
826 2018-01-26 Jim Wilson <jimw@sifive.com>
828 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
830 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
833 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
835 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
836 and CMP + SUB-immediate -> SUBS.
838 2018-01-26 Martin Sebor <msebor@redhat.com>
840 PR tree-optimization/83896
841 * tree-ssa-strlen.c (get_string_len): Rename...
842 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
843 Avoid assuming length is constant.
844 (handle_char_store): Use HOST_WIDE_INT for string length.
846 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
849 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
850 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
852 2018-01-26 Richard Biener <rguenther@suse.de>
854 PR rtl-optimization/84003
855 * dse.c (record_store): Only record redundant stores when
856 the earlier store aliases at least all accesses the later one does.
858 2018-01-26 Jakub Jelinek <jakub@redhat.com>
860 PR rtl-optimization/83985
861 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
862 REG_CFA_RESTORE insns.
863 (delete_unmarked_insns): Don't ignore separate shrink wrapping
864 REG_CFA_RESTORE insns here.
867 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
868 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
870 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
872 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
873 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
874 (arc_init): Likewise.
875 (arc_override_options): Likewise.
876 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
878 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
880 * config/arc/arc.h (TARGET_DBNZ): Define.
881 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
882 properly set the tune attribute.
883 (dbnz): Use TARGET_DBNZ guard.
884 * config/arc/arc.opt (mtune): Add core3 option.
886 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
888 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
889 recognize new pic like addresses.
890 (arc_delegitimize_address): Clean up.
892 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
894 * config/arc/arc-arches.def: Option mrf16 valid for all
896 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
897 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
898 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
899 * config/arc/arc-tables.opt: Regenerate.
900 * config/arc/arc.c (arc_conditional_register_usage): Handle
901 reduced register file case.
902 (arc_file_start): Set must have build attributes.
903 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
905 * config/arc/arc.opt (mrf16): Add new option.
906 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
907 * config/arc/genmultilib.awk: Handle new mrf16 option.
908 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
909 * config/arc/t-multilib: Regenerate.
910 * doc/invoke.texi (ARC Options): Document mrf16 option.
912 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
914 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
915 * config/arc/arc.c (arc_handle_secure_attribute): New function.
916 (arc_attribute_table): Add 'secure_call' attribute.
917 (arc_print_operand): Print secure call operand.
918 (arc_function_ok_for_sibcall): Don't optimize tail calls when
920 (arc_is_secure_call_p): New function. * config/arc/arc.md
921 (call_i): Add support for sjli instruction.
922 (call_value_i): Likewise.
923 * config/arc/constraints.md (Csc): New constraint.
925 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
926 John Eric Martin <John.Martin@emmicro-us.com>
928 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
929 * config/arc/arc.c (_arc_jli_section): New struct.
930 (arc_jli_section): New type.
931 (rc_jli_sections): New static variable.
932 (arc_handle_jli_attribute): New function.
933 (arc_attribute_table): Add jli_always and jli_fixed attribute.
934 (arc_file_end): New function.
935 (TARGET_ASM_FILE_END): Define.
936 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
937 (arc_add_jli_section): New function.
938 (jli_call_scan): Likewise.
939 (arc_reorg): Call jli_call_scan.
940 (arc_output_addsi): Remove 'S' from printing asm operand.
941 (arc_is_jli_call_p): New function.
942 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
944 (movhi_insn): Likewise.
945 (movsi_insn): Likewise.
946 (movsi_set_cc_insn): Likewise.
947 (loadqi_update): Likewise.
948 (load_zeroextendqisi_update): Likewise.
949 (load_signextendqisi_update): Likewise.
950 (loadhi_update): Likewise.
951 (load_zeroextendhisi_update): Likewise.
952 (load_signextendhisi_update): Likewise.
953 (loadsi_update): Likewise.
954 (loadsf_update): Likewise.
955 (movsicc_insn): Likewise.
956 (bset_insn): Likewise.
957 (bxor_insn): Likewise.
958 (bclr_insn): Likewise.
959 (bmsk_insn): Likewise.
960 (bicsi3_insn): Likewise.
961 (cmpsi_cc_c_insn): Likewise.
962 (movsi_ne): Likewise.
963 (movsi_cond_exec): Likewise.
964 (clrsbsi2): Likewise.
975 (call_i): Remove 'S' asm letter, add jli instruction.
976 (call_value_i): Likewise.
977 * config/arc/arc.op (mjli-always): New option.
978 * config/arc/constraints.md (Cji): New constraint.
979 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
981 (subsf3_fpx): Likewise.
982 (mulsf3_fpx): Likewise.
983 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
985 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
987 * doc/invoke.texi (ARC): Document mjli-always option.
989 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
991 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
992 and use incw and decw where possible
993 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
995 2018-01-26 Richard Biener <rguenther@suse.de>
997 PR tree-optimization/81082
998 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
999 association if it requires casting to unsigned.
1000 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1001 from fold_plusminus_mult_expr to catch important cases late when
1002 range info is available.
1004 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1006 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1007 * configure.ac (hidden_linkonce): New test.
1008 * configure: Regenerate.
1009 * config.in: Regenerate.
1011 2018-01-26 Julia Koval <julia.koval@intel.com>
1013 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1014 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1015 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1016 _mm_mask_bitshuffle_epi64_mask): Fix type.
1017 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1018 USI_FTYPE_V4DI_V4DI_USI): Remove.
1019 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1020 __builtin_ia32_vpshufbitqmb256_mask,
1021 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1022 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1023 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1025 2018-01-26 Alan Modra <amodra@gmail.com>
1028 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1029 UNSPEC_VBPERMQ. Sort other unspecs.
1031 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1033 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1035 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1038 * predict.c (drop_profile): Do not push/pop cfun; update also
1040 (handle_missing_profiles): Fix logic looking for zero profiles.
1042 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1045 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1046 on functions with #pragma omp declare simd or functions with simd
1048 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1049 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1050 Remove trailing \n from warning_at calls.
1052 2018-01-25 Tom de Vries <tom@codesourcery.com>
1055 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1056 for neutered workers.
1058 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1061 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1062 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1064 2018-01-24 Jeff Law <law@redhat.com>
1067 * i386.c (get_probe_interval): Move to earlier point.
1068 (ix86_compute_frame_layout): If -fstack-clash-protection and
1069 the frame is larger than the probe interval, then use pushes
1070 to save registers rather than reg->mem moves.
1071 (ix86_expand_prologue): Remove conditional for int_registers_saved
1074 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1077 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1078 min/max for never referenced object.
1080 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1083 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1085 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1086 attributes from DECL_ATTRIBUTES (decl) without affecting
1087 DECL_ATTRIBUTES (current_function_decl).
1088 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1089 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1091 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1093 PR tree-optimization/83979
1094 * fold-const.c (fold_comparison): Use constant_boolean_node
1095 instead of boolean_{true,false}_node.
1097 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1099 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1102 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1104 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1105 Simplify the clause that sets the length attribute.
1106 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1107 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1108 clause that sets the length attribute.
1109 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1111 2018-01-24 Tom de Vries <tom@codesourcery.com>
1114 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1115 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1116 Add strict parameter.
1117 (prevent_branch_around_nothing): Insert dummy insn between branch to
1118 label and label with no ptx insn inbetween.
1119 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1121 2018-01-24 Tom de Vries <tom@codesourcery.com>
1124 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1125 for neutered threads in warp.
1126 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1128 2018-01-24 Richard Biener <rguenther@suse.de>
1130 PR tree-optimization/83176
1131 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1134 2018-01-24 Richard Biener <rguenther@suse.de>
1136 PR tree-optimization/82819
1137 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1138 code generating pluses that are no-ops in the target precision.
1140 2018-01-24 Richard Biener <rguenther@suse.de>
1143 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1145 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1147 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1148 to merge probabilities.
1149 * predict.c (probably_never_executed): Also mark as cold functions
1150 with global 0 profile and guessed local profile.
1151 * profile-count.c (profile_probability::combine_with_count): New
1153 * profile-count.h (profile_probability::operator*,
1154 profile_probability::operator*=, profile_probability::operator/,
1155 profile_probability::operator/=): Reduce precision to adjusted
1156 and set value to guessed on contradictory divisions.
1157 (profile_probability::combine_with_freq): Remove.
1158 (profile_probability::combine_wiht_count): Declare.
1159 (profile_count::force_nonzero):: Set to adjusted.
1160 (profile_count::probability_in):: Set quality to adjusted.
1161 * tree-ssa-tail-merge.c (replace_block_by): Use
1164 2018-01-23 Andrew Waterman <andrew@sifive.com>
1165 Jim Wilson <jimw@sifive.com>
1167 * config/riscv/riscv.c (riscv_stack_boundary): New.
1168 (riscv_option_override): Set riscv_stack_boundary. Handle
1169 riscv_preferred_stack_boundary_arg.
1170 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1171 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1172 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1173 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1174 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1175 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1177 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1180 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1181 of struct ix86_frame.
1182 (ix86_expand_epilogue): Likewise. Add a local variable for
1183 the reg_save_offset field in struct ix86_frame.
1185 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1187 PR tree-optimization/82604
1188 * tree-loop-distribution.c (enum partition_kind): New enum item
1189 PKIND_PARTIAL_MEMSET.
1190 (partition_builtin_p): Support above new enum item.
1191 (generate_code_for_partition): Ditto.
1192 (compute_access_range): Differentiate cases that equality can be
1193 proven at all loops, the innermost loops or no loops.
1194 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1195 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1196 (finalize_partitions, distribute_loop): Don't fuse partition of
1197 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1198 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1201 2018-01-23 Martin Liska <mliska@suse.cz>
1203 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1204 order to ignore the predictor.
1205 (PRED_POLYMORPHIC_CALL): Likewise.
1206 (PRED_RECURSIVE_CALL): Likewise.
1208 2018-01-23 Martin Liska <mliska@suse.cz>
1210 * tree-profile.c (tree_profiling): Print function header to
1211 aware reader which function we are working on.
1212 * value-prof.c (gimple_find_values_to_profile): Do not print
1213 not interesting value histograms.
1215 2018-01-23 Martin Liska <mliska@suse.cz>
1217 * profile-count.h (enum profile_quality): Add
1218 profile_uninitialized as the first value. Do not number values
1219 as they are zero based.
1220 (profile_count::verify): Update sanity check.
1221 (profile_probability::verify): Likewise.
1223 2018-01-23 Nathan Sidwell <nathan@acm.org>
1225 * doc/invoke.texi (ffor-scope): Deprecate.
1227 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1229 PR tree-optimization/83510
1230 * domwalk.c (set_all_edges_as_executable): New function.
1231 (dom_walker::dom_walker): Convert bool param
1232 "skip_unreachable_blocks" to enum reachability. Move setup of
1233 edge flags to set_all_edges_as_executable and only do it when
1234 reachability is REACHABLE_BLOCKS.
1235 * domwalk.h (enum dom_walker::reachability): New enum.
1236 (dom_walker::dom_walker): Convert bool param
1237 "skip_unreachable_blocks" to enum reachability.
1238 (set_all_edges_as_executable): New decl.
1239 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1240 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1242 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1243 but converting true to REACHABLE_BLOCKS.
1244 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1246 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1247 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1248 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1250 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1251 if check_all_array_refs will be called.
1253 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1255 * tree.c (selftest::test_location_wrappers): Add more test
1258 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1260 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1261 (selftest::test_bit_in_range): Likewise.
1263 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1266 * doc/sourcebuild.texi (vect_float): Say that the selector
1267 only describes the situation when -funsafe-math-optimizations is on.
1268 (vect_float_strict): Document.
1270 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1272 PR tree-optimization/83965
1273 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1274 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1275 instead of checking only for a reduction.
1276 (vect_recog_widen_sum_pattern): Likewise.
1278 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1280 * predict.c (probably_never_executed): Only use precise profile info.
1281 (compute_function_frequency): Skip after inlining hack since we now
1282 have quality checking.
1284 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1286 * profile-count.h (profile_probability::very_unlikely,
1287 profile_probability::unlikely, profile_probability::even): Set
1288 precision to guessed.
1290 2018-01-23 Richard Biener <rguenther@suse.de>
1292 PR tree-optimization/83963
1293 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1294 Properly terminate dominator walk when crossing the exit edge not
1295 when visiting its source block.
1297 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1300 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1301 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1303 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1305 PR tree-optimization/83957
1306 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1307 semicolon after for body surrounded by braces.
1309 PR tree-optimization/83081
1310 * profile-count.h (profile_probability::split): New method.
1311 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1312 Use profile_probability::split.
1313 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1314 when splitting a single conditional jump into 2.
1316 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1318 PR tree-optimization/69452
1319 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1322 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1324 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1325 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1326 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1328 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1330 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1331 * config/rl78/rl78.md: New define_expand "movdi"
1332 * config/rl78/rl78.c: New function definition rl78_split_movdi
1334 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1337 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1339 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1340 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1341 128-bit to produce an UNSPEC move to get the double word with the
1342 signbit and then a shift directly to do signbit.
1343 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1344 implementation with a new version that just does either a direct
1345 move or a regular move. Move memory interface to separate insns.
1346 Move insns so they are next to the expander.
1347 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1348 with signbit move. Split big and little endian case.
1349 (signbit<mode>2_dm_mem_le): Likewise.
1350 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1351 (signbit<mode>2_dm2): Likewise.
1353 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1355 * config/rl78/rl78.md: New define_expand "anddi3".
1357 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1359 * config/rl78/rl78.md: New define_expand "umindi3".
1361 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1363 * config/rl78/rl78.md: New define_expand "smindi3".
1365 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1367 * config/rl78/rl78.md: New define_expand "smaxdi3".
1369 2018-01-22 Carl Love <cel@us.ibm.com>
1371 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1372 LVX_V1TI): Add macro expansion.
1373 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1374 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1375 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1376 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1377 Change check to determine if the instruction is a byte reversing
1378 entry. Fix typo in comment.
1379 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1380 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1381 Add def_builtin calls for new builtins.
1382 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1383 Add define_insn expansion.
1385 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1387 * config/rl78/rl78.md: New define_expand "umaxdi3".
1389 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1391 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1392 for non-QImode registers
1394 2018-01-22 Richard Biener <rguenther@suse.de>
1396 PR tree-optimization/83963
1397 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1398 including the loop exit block.
1399 (scop_detection::merge_sese): Likewise.
1400 (scop_detection::add_scop): Do it here instead.
1402 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1404 * doc/sourcebuild.texi (arm_softfloat): Document.
1406 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1409 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1410 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1411 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1413 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1414 David Edelsohn <dje.gcc@gmail.com>
1417 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1418 Change "crset eq" to "crset 2".
1419 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1420 (*call_indirect_aix<mode>_nospec): Likewise.
1421 (*call_value_indirect_aix<mode>_nospec): Likewise.
1422 (*call_indirect_elfv2<mode>_nospec): Likewise.
1423 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1424 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1425 change assembly output from . to $.
1426 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1427 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1428 (*tablejump<mode>_internal1_nospec): Likewise.
1430 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1433 * config/sh/sh_optimize_sett_clrt.cc:
1434 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1436 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1438 PR tree-optimization/83940
1439 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1440 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1441 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1442 use it to pass back the definition type.
1443 (vect_check_store_rhs): Likewise rhs_dt_out.
1444 (vect_build_gather_load_calls): Add a mask_dt argument and use
1445 it instead of a call to vect_is_simple_use.
1446 (vectorizable_store): Update calls to vect_check_load_store_mask
1447 and vect_check_store_rhs. Use the dt returned by the latter instead
1448 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1449 instead of calls to vect_is_simple_use. Pass the scalar rather
1450 than the vector operand to vect_is_simple_use when handling
1451 second and subsequent copies of an rhs value.
1452 (vectorizable_load): Update calls to vect_check_load_store_mask
1453 and vect_build_gather_load_calls. Use the cached mask_dt and
1454 gs_info.offset_dt instead of calls to vect_is_simple_use.
1456 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1459 * tree-emutls.c: Include gimplify.h.
1460 (lower_emutls_2): New function.
1461 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1462 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1463 it before further processing.
1466 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1467 UINTVAL (trueop1) instead of INTVAL (op1).
1469 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1473 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1474 INCOMING_FRAME_SP_OFFSET if not defined.
1475 (scan_trace): Add ENTRY argument. If true and
1476 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1477 emit a note to adjust the CFA offset.
1478 (create_cfi_notes): Adjust scan_trace callers.
1479 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1480 INCOMING_FRAME_SP_OFFSET in the CIE.
1481 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1482 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1484 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1485 * doc/tm.texi: Regenerated.
1487 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1489 PR rtl-optimization/83147
1490 * lra-constraints.c (remove_inheritance_pseudos): Use
1491 lra_substitute_pseudo_within_insn.
1493 2018-01-19 Tom de Vries <tom@codesourcery.com>
1494 Cesar Philippidis <cesar@codesourcery.com>
1497 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1499 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1502 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1503 spaces for function labels.
1505 2018-01-19 Martin Liska <mliska@suse.cz>
1507 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1508 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1509 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1510 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1511 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1512 (PRED_CONST_RETURN): Change from 69 to 65.
1513 (PRED_NULL_RETURN): Change from 91 to 71.
1514 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1515 (PRED_LOOP_GUARD): Change from 66 to 73.
1517 2018-01-19 Martin Liska <mliska@suse.cz>
1519 * predict.c (predict_insn_def): Add new assert.
1520 (struct branch_predictor): Change type to signed integer.
1521 (test_prediction_value_range): Amend test to cover
1523 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1524 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1525 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1526 (PRED_LOOP_IV_COMPARE): Likewise.
1527 * predict.h (PROB_UNINITIALIZED): Define new constant.
1529 2018-01-19 Martin Liska <mliska@suse.cz>
1531 * predict.c (dump_prediction): Add new format for
1532 analyze_brprob.py script which is enabled with -details
1534 * profile-count.h (precise_p): New function.
1536 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1538 PR tree-optimization/83922
1539 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1540 there are no statements that need masking.
1541 (vect_active_double_reduction_p): New function.
1542 (vect_analyze_loop_operations): Use it when handling phis that
1543 are not in the loop header.
1545 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1547 PR tree-optimization/83914
1548 * tree-vect-loop.c (vectorizable_induction): Don't convert
1549 init_expr or apply the peeling adjustment for inductions
1550 that are nested within the vectorized loop.
1552 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1554 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1557 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1561 * function.h (gimplify_parameters): Add gimple_seq * argument.
1562 * function.c: Include gimple.h and options.h.
1563 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1564 for the added local temporaries if needed.
1565 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1566 if there are any parameter cleanups, wrap whole body into a
1567 try/finally with the cleanups.
1569 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1572 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1573 Use GET_MODE_CLASS for scalar floating point.
1575 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1579 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1580 Fix call of call_cgraph_insertion_hooks.
1582 2018-01-18 Martin Sebor <msebor@redhat.com>
1584 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1586 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1589 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1592 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1595 * common.opt: (-ffile-prefix-map): New option.
1596 * opts.c (common_handle_option): Defer it.
1597 * opts-global.c (handle_common_deferred_options): Handle it.
1598 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1599 * file-prefix-map.h: New file.
1600 (remap_debug_filename, add_debug_prefix_map): ...here.
1601 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1602 * final.c (debug_prefix_map, add_debug_prefix_map
1603 remap_debug_filename): Move to...
1604 * file-prefix-map.c: New file.
1605 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1606 generalize, get rid of alloca(), use strrchr() instead of strchr().
1607 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1608 Implement in terms of add_prefix_map().
1609 (remap_macro_filename, remap_debug_filename): Implement in term of
1611 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1612 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1613 * dbxout.c: Include file-prefix-map.h.
1614 * varasm.c: Likewise.
1615 * vmsdbgout.c: Likewise.
1616 * xcoffout.c: Likewise.
1617 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1618 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1619 * doc/invoke.texi (-ffile-prefix-map): Document.
1620 (-fdebug-prefix-map): Update description.
1622 2018-01-18 Martin Liska <mliska@suse.cz>
1624 * config/i386/i386.c (indirect_thunk_name): Document that also
1626 (output_indirect_thunk): Document why both instructions
1627 (pause and lfence) are generated.
1629 2018-01-18 Richard Biener <rguenther@suse.de>
1631 PR tree-optimization/83887
1632 * graphite-scop-detection.c
1633 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1634 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1635 (scop_detection::merge_sese): Re-implement with a flood-fill
1636 algorithm that properly finds a SESE region if it exists.
1638 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1641 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1642 pointer_diff optimizations use view_convert instead of convert.
1644 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1646 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1647 Generate different code for -mno-speculate-indirect-jumps.
1648 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1649 (*call_indirect_aix<mode>): Disable for
1650 -mno-speculate-indirect-jumps.
1651 (*call_indirect_aix<mode>_nospec): New define_insn.
1652 (*call_value_indirect_aix<mode>): Disable for
1653 -mno-speculate-indirect-jumps.
1654 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1655 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1656 -mno-speculate-indirect-jumps.
1657 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1659 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1661 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1662 long double type, set the flags for noting the default long double
1663 type, even if we don't pass or return a long double type.
1665 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1668 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1671 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1673 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1674 support for merge[hl].
1675 (fold_mergehl_helper): New helper function.
1676 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1677 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1678 (altivec_vmrglw_direct): Add xxmrglw insn.
1680 2018-01-17 Andrew Waterman <andrew@sifive.com>
1682 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1683 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1685 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1688 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1689 call the lto_location_cache before reading the
1690 DECL_SOURCE_LOCATION of the types.
1692 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1693 Richard Sandiford <richard.sandiford@linaro.org>
1695 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1696 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1697 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1698 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1699 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1701 * config/aarch64/constraints.md (aarch64_movti_operand):
1703 * config/aarch64/predicates.md (Uti): Add new constraint.
1705 2018-01-17 Carl Love <cel@us.ibm.com>
1706 * config/rs6000/vsx.md (define_expand xl_len_r,
1707 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1708 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1710 (define_expand, define_insn): Move the shift left from the
1711 define_insn to the define_expand for lxvl and stxvl instructions.
1712 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1713 and XL_LEN_R definitions to PURE.
1715 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1717 * config/i386/i386.c (indirect_thunk_name): Declare regno
1718 as unsigned int. Compare regno with INVALID_REGNUM.
1719 (output_indirect_thunk): Ditto.
1720 (output_indirect_thunk_function): Ditto.
1721 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1722 in the call to output_indirect_thunk_function.
1724 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1727 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1728 rather than the size of inner_type to determine the stack slot size
1729 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1731 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1734 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1737 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1739 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1740 endian Linux systems to optionally enable multilibs for selecting
1741 the long double type if the user configured an explicit type.
1742 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1743 have no long double multilibs if not defined.
1744 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1745 warn if the user used -mabi={ieee,ibm}longdouble and we built
1746 multilibs for long double.
1747 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1748 appropriate multilib option.
1749 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1751 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1752 for building long double multilibs.
1753 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1755 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1757 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1760 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1762 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1765 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1768 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1771 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1773 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1774 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1776 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1778 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1779 different rtl trees depending on TARGET_64BIT.
1780 (rs6000_gen_lvx): Likewise.
1782 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1784 * config/visium/visium.md (nop): Tweak comment.
1785 (hazard_nop): Likewise.
1787 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1789 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1790 -mspeculate-indirect-jumps.
1791 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1792 for -mno-speculate-indirect-jumps.
1793 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1794 (*call_value_indirect_elfv2<mode>): Disable for
1795 -mno-speculate-indirect-jumps.
1796 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1797 (indirect_jump): Emit different RTL for
1798 -mno-speculate-indirect-jumps.
1799 (*indirect_jump<mode>): Disable for
1800 -mno-speculate-indirect-jumps.
1801 (*indirect_jump<mode>_nospec): New define_insn.
1802 (tablejump): Emit different RTL for
1803 -mno-speculate-indirect-jumps.
1804 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1805 (tablejumpsi_nospec): New define_expand.
1806 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1807 (tablejumpdi_nospec): New define_expand.
1808 (*tablejump<mode>_internal1): Disable for
1809 -mno-speculate-indirect-jumps.
1810 (*tablejump<mode>_internal1_nospec): New define_insn.
1811 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1814 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1816 * caller-save.c (insert_save): Drop unnecessary parameter. All
1819 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1820 Richard Biener <rguenth@suse.de>
1823 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1824 return early, inline manually is_gimple_sizepos. Make sure if we
1825 call gimplify_expr we don't end up with a gimple constant.
1826 * tree.c (variably_modified_type_p): Don't return true for
1827 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1828 * gimplify.h (is_gimple_sizepos): Remove.
1830 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1832 PR tree-optimization/83857
1833 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1834 vectorizable_live_operation for pure SLP statements.
1835 (vectorizable_live_operation): Handle PHIs.
1837 2018-01-16 Richard Biener <rguenther@suse.de>
1839 PR tree-optimization/83867
1840 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1841 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1843 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1846 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1847 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1848 If off is not INTEGER_CST, issue a may not be aligned warning
1849 rather than isn't aligned. Use isn%'t rather than isn't.
1850 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1852 <case MULT_EXPR>: Improve the case when bottom and one of the
1853 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1854 operand, in that case check if the other operand is multiple of
1855 bottom divided by the INTEGER_CST operand.
1857 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1860 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1861 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1862 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1863 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1864 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1865 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1866 (pa_function_arg_size): New function.
1868 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1870 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1871 in a separate statement.
1873 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1875 PR tree-optimization/83847
1876 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1877 group gathers and scatters.
1879 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1881 PR rtl-optimization/86620
1882 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1884 PR rtl-optimization/83213
1885 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1886 to last if both are JUMP_INSNs.
1888 PR tree-optimization/83843
1889 * gimple-ssa-store-merging.c
1890 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1891 store_immediate_info for bswap/nop orig_stores.
1893 2018-01-15 Andrew Waterman <andrew@sifive.com>
1895 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1897 <UDIV>: Increase cost if !TARGET_DIV.
1899 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1901 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1902 (define_attr "cr_logical_3op"): New.
1903 (cceq_ior_compare): Adjust.
1904 (cceq_ior_compare_complement): Adjust.
1905 (*cceq_rev_compare): Adjust.
1906 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1907 (is_cracked_insn): Adjust.
1908 (insn_must_be_first_in_group): Adjust.
1909 * config/rs6000/40x.md: Adjust.
1910 * config/rs6000/440.md: Adjust.
1911 * config/rs6000/476.md: Adjust.
1912 * config/rs6000/601.md: Adjust.
1913 * config/rs6000/603.md: Adjust.
1914 * config/rs6000/6xx.md: Adjust.
1915 * config/rs6000/7450.md: Adjust.
1916 * config/rs6000/7xx.md: Adjust.
1917 * config/rs6000/8540.md: Adjust.
1918 * config/rs6000/cell.md: Adjust.
1919 * config/rs6000/e300c2c3.md: Adjust.
1920 * config/rs6000/e500mc.md: Adjust.
1921 * config/rs6000/e500mc64.md: Adjust.
1922 * config/rs6000/e5500.md: Adjust.
1923 * config/rs6000/e6500.md: Adjust.
1924 * config/rs6000/mpc.md: Adjust.
1925 * config/rs6000/power4.md: Adjust.
1926 * config/rs6000/power5.md: Adjust.
1927 * config/rs6000/power6.md: Adjust.
1928 * config/rs6000/power7.md: Adjust.
1929 * config/rs6000/power8.md: Adjust.
1930 * config/rs6000/power9.md: Adjust.
1931 * config/rs6000/rs64.md: Adjust.
1932 * config/rs6000/titan.md: Adjust.
1934 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1936 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1937 ix86_indirect_branch_register logic.
1939 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1941 * config/i386/constraints.md (Bs): Update
1942 ix86_indirect_branch_register check. Don't check
1943 ix86_indirect_branch_register with GOT_memory_operand.
1945 * config/i386/predicates.md (GOT_memory_operand): Don't check
1946 ix86_indirect_branch_register here.
1947 (GOT32_symbol_operand): Likewise.
1949 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1951 * config/i386/predicates.md (constant_call_address_operand):
1952 Rewrite ix86_indirect_branch_register logic.
1953 (sibcall_insn_operand): Likewise.
1955 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1957 * config/i386/constraints.md (Bs): Replace
1958 ix86_indirect_branch_thunk_register with
1959 ix86_indirect_branch_register.
1961 * config/i386/i386.md (indirect_jump): Likewise.
1962 (tablejump): Likewise.
1963 (*sibcall_memory): Likewise.
1964 (*sibcall_value_memory): Likewise.
1965 Peepholes of indirect call and jump via memory: Likewise.
1966 * config/i386/i386.opt: Likewise.
1967 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1968 (GOT_memory_operand): Likewise.
1969 (call_insn_operand): Likewise.
1970 (sibcall_insn_operand): Likewise.
1971 (GOT32_symbol_operand): Likewise.
1973 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1976 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1977 type rather than type addr's type points to.
1978 (expand_omp_atomic_mutex): Likewise.
1979 (expand_omp_atomic): Likewise.
1981 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1984 * config/i386/i386.c (output_indirect_thunk_function): Use
1985 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1986 for __x86_return_thunk.
1988 2018-01-15 Richard Biener <rguenther@suse.de>
1991 * expmed.c (extract_bit_field_1): Fix typo.
1993 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1996 * config/arm/iterators.md (VF): New mode iterator.
1997 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1998 Remove integer-related logic from pattern.
1999 (neon_vabd<mode>_3): Likewise.
2001 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2004 * common.opt (fstrict-overflow): No longer an alias.
2005 (fwrapv-pointer): New option.
2006 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2007 also for pointer types based on flag_wrapv_pointer.
2008 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2009 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2010 opts->x_flag_wrapv got set.
2011 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2012 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2013 POINTER_TYPE_OVERFLOW_UNDEFINED.
2014 * match.pd: Likewise in address comparison pattern.
2015 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2017 2018-01-15 Richard Biener <rguenther@suse.de>
2020 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2021 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2022 Reset type names to their identifier if their TYPE_DECL doesn't
2023 have linkage (and thus is used for ODR and devirt).
2024 (save_debug_info_for_decl): Remove.
2025 (save_debug_info_for_type): Likewise.
2026 (add_tree_to_fld_list): Adjust.
2027 * tree-pretty-print.c (dump_generic_node): Make dumping of
2028 type names more robust.
2030 2018-01-15 Richard Biener <rguenther@suse.de>
2032 * BASE-VER: Bump to 8.0.1.
2034 2018-01-14 Martin Sebor <msebor@redhat.com>
2037 * builtins.c (check_access): Avoid warning when the no-warning bit
2040 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2042 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2043 * ira-color (allocno_hard_regs_compare): Likewise.
2045 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2048 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2049 Use .pushsection/.popsection.
2051 2018-01-14 Martin Sebor <msebor@redhat.com>
2054 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2056 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2058 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2059 entry from extra_headers.
2060 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2061 extra_headers, make the list bitwise identical to the i?86-*-* one.
2063 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2065 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2066 -mcmodel=large with -mindirect-branch=thunk,
2067 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2068 -mfunction-return=thunk-extern.
2069 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2070 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2071 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2073 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2075 * config/i386/i386.c (print_reg): Print the name of the full
2076 integer register without '%'.
2077 (ix86_print_operand): Handle 'V'.
2078 * doc/extend.texi: Document 'V' modifier.
2080 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2082 * config/i386/constraints.md (Bs): Disallow memory operand for
2083 -mindirect-branch-register.
2085 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2086 (GOT_memory_operand): Likewise.
2087 (call_insn_operand): Likewise.
2088 (sibcall_insn_operand): Likewise.
2089 (GOT32_symbol_operand): Likewise.
2090 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2091 for -mindirect-branch-register.
2092 (tablejump): Likewise.
2093 (*sibcall_memory): Likewise.
2094 (*sibcall_value_memory): Likewise.
2095 Disallow peepholes of indirect call and jump via memory for
2096 -mindirect-branch-register.
2097 (*call_pop): Replace m with Bw.
2098 (*call_value_pop): Likewise.
2099 (*sibcall_pop_memory): Replace m with Bs.
2100 * config/i386/i386.opt (mindirect-branch-register): New option.
2101 * doc/invoke.texi: Document -mindirect-branch-register option.
2103 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2105 * config/i386/i386-protos.h (ix86_output_function_return): New.
2106 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2107 set function_return_type.
2108 (indirect_thunk_name): Add ret_p to indicate thunk for function
2110 (output_indirect_thunk_function): Pass false to
2111 indirect_thunk_name.
2112 (ix86_output_indirect_branch_via_reg): Likewise.
2113 (ix86_output_indirect_branch_via_push): Likewise.
2114 (output_indirect_thunk_function): Create alias for function
2115 return thunk if regno < 0.
2116 (ix86_output_function_return): New function.
2117 (ix86_handle_fndecl_attribute): Handle function_return.
2118 (ix86_attribute_table): Add function_return.
2119 * config/i386/i386.h (machine_function): Add
2120 function_return_type.
2121 * config/i386/i386.md (simple_return_internal): Use
2122 ix86_output_function_return.
2123 (simple_return_internal_long): Likewise.
2124 * config/i386/i386.opt (mfunction-return=): New option.
2125 (indirect_branch): Mention -mfunction-return=.
2126 * doc/extend.texi: Document function_return function attribute.
2127 * doc/invoke.texi: Document -mfunction-return= option.
2129 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2131 * config/i386/i386-opts.h (indirect_branch): New.
2132 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2133 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2134 with local indirect jump when converting indirect call and jump.
2135 (ix86_set_indirect_branch_type): New.
2136 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2137 (indirectlabelno): New.
2138 (indirect_thunk_needed): Likewise.
2139 (indirect_thunk_bnd_needed): Likewise.
2140 (indirect_thunks_used): Likewise.
2141 (indirect_thunks_bnd_used): Likewise.
2142 (INDIRECT_LABEL): Likewise.
2143 (indirect_thunk_name): Likewise.
2144 (output_indirect_thunk): Likewise.
2145 (output_indirect_thunk_function): Likewise.
2146 (ix86_output_indirect_branch_via_reg): Likewise.
2147 (ix86_output_indirect_branch_via_push): Likewise.
2148 (ix86_output_indirect_branch): Likewise.
2149 (ix86_output_indirect_jmp): Likewise.
2150 (ix86_code_end): Call output_indirect_thunk_function if needed.
2151 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2153 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2154 (ix86_attribute_table): Add indirect_branch.
2155 * config/i386/i386.h (machine_function): Add indirect_branch_type
2156 and has_local_indirect_jump.
2157 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2159 (tablejump): Likewise.
2160 (*indirect_jump): Use ix86_output_indirect_jmp.
2161 (*tablejump_1): Likewise.
2162 (simple_return_indirect_internal): Likewise.
2163 * config/i386/i386.opt (mindirect-branch=): New option.
2164 (indirect_branch): New.
2167 (thunk-inline): Likewise.
2168 (thunk-extern): Likewise.
2169 * doc/extend.texi: Document indirect_branch function attribute.
2170 * doc/invoke.texi: Document -mindirect-branch= option.
2172 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2175 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2177 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2179 * ipa-inline.c (want_inline_small_function_p): Return false if
2180 inlining has already failed with CIF_FINAL_ERROR.
2181 (update_caller_keys): Call want_inline_small_function_p before
2183 (update_callee_keys): Likewise.
2185 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2187 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2189 (rs6000_quadword_masked_address_p): Likewise.
2190 (quad_aligned_load_p): Likewise.
2191 (quad_aligned_store_p): Likewise.
2192 (const_load_sequence_p): Add comment to describe the outer-most loop.
2193 (mimic_memory_attributes_and_flags): New function.
2194 (rs6000_gen_stvx): Likewise.
2195 (replace_swapped_aligned_store): Likewise.
2196 (rs6000_gen_lvx): Likewise.
2197 (replace_swapped_aligned_load): Likewise.
2198 (replace_swapped_load_constant): Capitalize argument name in
2199 comment describing this function.
2200 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2201 and stores that access quad-word aligned addresses and replace
2202 with stvx or lvx instructions when appropriate.
2203 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2204 New function prototype.
2205 (rs6000_quadword_masked_address_p): Likewise.
2206 (rs6000_gen_lvx): Likewise.
2207 (rs6000_gen_stvx): Likewise.
2208 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2209 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2210 when memory address is aligned.
2211 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2212 this split to select lvx instruction when memory address is aligned.
2213 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2214 instruction when memory address is aligned.
2215 (*vsx_le_perm_load_v16qi): Likewise.
2216 (four unnamed splitters): Modify to select the stvx instruction
2217 when memory is aligned.
2219 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2221 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2222 which appears in the queue multiple times.
2224 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2225 Alan Hayward <alan.hayward@arm.com>
2226 David Sherwood <david.sherwood@arm.com>
2228 * tree-vectorizer.h (vec_lower_bound): New structure.
2229 (_loop_vec_info): Add check_nonzero and lower_bounds.
2230 (LOOP_VINFO_CHECK_NONZERO): New macro.
2231 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2232 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2233 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2234 fields. Make seg_len the distance travelled, not including the
2236 (dr_direction_indicator): Declare.
2237 (dr_zero_step_indicator): Likewise.
2238 (dr_known_forward_stride_p): Likewise.
2239 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2241 (runtime_alias_check_p): Allow runtime alias checks with
2243 (operator ==): Compare access_size and align.
2244 (prune_runtime_alias_test_list): Rework for new distinction between
2245 the access_size and seg_len.
2246 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2248 (get_segment_min_max): New function.
2249 (create_intersect_range_checks): Use it.
2250 (dr_step_indicator): New function.
2251 (dr_direction_indicator): Likewise.
2252 (dr_zero_step_indicator): Likewise.
2253 (dr_known_forward_stride_p): Likewise.
2254 * tree-loop-distribution.c (data_ref_segment_size): Return
2255 DR_STEP * (niters - 1).
2256 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2258 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2259 (vect_preserves_scalar_order_p): New function, split out from...
2260 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2261 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2262 (vect_vfa_access_size): New function.
2263 (vect_vfa_align): Likewise.
2264 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2265 (dump_lower_bound): New function.
2266 (vect_check_lower_bound): Likewise.
2267 (vect_small_gap_p): Likewise.
2268 (vectorizable_with_step_bound_p): Likewise.
2269 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2270 depencies if the vectorization factor is 1. Convert the checks
2271 for nonzero steps into checks on the bounds of DR_STEP. Try using
2272 a bunds check for variable steps if the minimum required step is
2273 relatively small. Update calls to the dr_with_seg_len
2274 constructor and to vect_compile_time_alias.
2275 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2277 (vect_loop_versioning): Call it.
2278 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2280 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2282 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2283 Alan Hayward <alan.hayward@arm.com>
2284 David Sherwood <david.sherwood@arm.com>
2286 * doc/sourcebuild.texi (vect_scatter_store): Document.
2287 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2289 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2291 * genopinit.c (main): Add supports_vec_scatter_store and
2292 supports_vec_scatter_store_cached to target_optabs.
2293 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2294 IFN_MASK_SCATTER_STORE.
2295 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2297 * internal-fn.h (internal_store_fn_p): Declare.
2298 (internal_fn_stored_value_index): Likewise.
2299 * internal-fn.c (scatter_store_direct): New macro.
2300 (expand_scatter_store_optab_fn): New function.
2301 (direct_scatter_store_optab_supported_p): New macro.
2302 (internal_store_fn_p): New function.
2303 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2304 IFN_MASK_SCATTER_STORE.
2305 (internal_fn_mask_index): Likewise.
2306 (internal_fn_stored_value_index): New function.
2307 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2309 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2310 * optabs-query.c (supports_vec_scatter_store_p): New function.
2311 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2312 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2313 true for scatter stores.
2314 (vect_gather_scatter_fn_p): Handle scatter stores too.
2315 (vect_check_gather_scatter): Consider using scatter stores if
2316 supports_vec_scatter_store_p.
2317 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2319 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2320 internal_fn_stored_value_index.
2321 (check_load_store_masking): Handle scatter stores too.
2322 (vect_get_store_rhs): Make public.
2323 (vectorizable_call): Use internal_store_fn_p.
2324 (vectorizable_store): Handle scatter store internal functions.
2325 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2326 when deciding whether the end of the group has been reached.
2327 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2328 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2329 (mask_scatter_store<mode>): New insns.
2331 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2332 Alan Hayward <alan.hayward@arm.com>
2333 David Sherwood <david.sherwood@arm.com>
2335 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2336 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2337 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2339 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2340 Use vect_truncate_gather_scatter_offset if we can't treat the
2341 operation as a normal gather load or scatter store.
2342 (get_group_load_store_type): Take the gather_scatter_info
2343 as argument. Try using a gather load or scatter store for
2344 single-element groups.
2345 (get_load_store_type): Update calls to get_group_load_store_type
2346 and vect_use_strided_gather_scatters_p.
2348 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2349 Alan Hayward <alan.hayward@arm.com>
2350 David Sherwood <david.sherwood@arm.com>
2352 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2353 optional tree argument.
2354 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2356 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2357 but continue to use the current value as a fallback.
2358 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2359 to compare the updates.
2360 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2361 (get_load_store_type): Use it when handling a strided access.
2362 (vect_get_strided_load_store_ops): New function.
2363 (vect_get_data_ptr_increment): Likewise.
2364 (vectorizable_load): Handle strided gather loads. Always pass
2365 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2367 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2368 Alan Hayward <alan.hayward@arm.com>
2369 David Sherwood <david.sherwood@arm.com>
2371 * doc/md.texi (gather_load@var{m}): Document.
2372 (mask_gather_load@var{m}): Likewise.
2373 * genopinit.c (main): Add supports_vec_gather_load and
2374 supports_vec_gather_load_cached to target_optabs.
2375 * optabs-tree.c (init_tree_optimization_optabs): Use
2376 ggc_cleared_alloc to allocate target_optabs.
2377 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2378 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2380 * internal-fn.h (internal_load_fn_p): Declare.
2381 (internal_gather_scatter_fn_p): Likewise.
2382 (internal_fn_mask_index): Likewise.
2383 (internal_gather_scatter_fn_supported_p): Likewise.
2384 * internal-fn.c (gather_load_direct): New macro.
2385 (expand_gather_load_optab_fn): New function.
2386 (direct_gather_load_optab_supported_p): New macro.
2387 (direct_internal_fn_optab): New function.
2388 (internal_load_fn_p): Likewise.
2389 (internal_gather_scatter_fn_p): Likewise.
2390 (internal_fn_mask_index): Likewise.
2391 (internal_gather_scatter_fn_supported_p): Likewise.
2392 * optabs-query.c (supports_at_least_one_mode_p): New function.
2393 (supports_vec_gather_load_p): Likewise.
2394 * optabs-query.h (supports_vec_gather_load_p): Declare.
2395 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2396 and memory_type field.
2397 (NUM_PATTERNS): Bump to 15.
2398 * tree-vect-data-refs.c: Include internal-fn.h.
2399 (vect_gather_scatter_fn_p): New function.
2400 (vect_describe_gather_scatter_call): Likewise.
2401 (vect_check_gather_scatter): Try using internal functions for
2402 gather loads. Recognize existing calls to a gather load function.
2403 (vect_analyze_data_refs): Consider using gather loads if
2404 supports_vec_gather_load_p.
2405 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2406 (vect_get_gather_scatter_offset_type): Likewise.
2407 (vect_convert_mask_for_vectype): Likewise.
2408 (vect_add_conversion_to_patterm): Likewise.
2409 (vect_try_gather_scatter_pattern): Likewise.
2410 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2411 (vect_vect_recog_func_ptrs): Add it.
2412 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2413 internal_fn_mask_index and internal_gather_scatter_fn_p.
2414 (check_load_store_masking): Take the gather_scatter_info as an
2415 argument and handle gather loads.
2416 (vect_get_gather_scatter_ops): New function.
2417 (vectorizable_call): Check internal_load_fn_p.
2418 (vectorizable_load): Likewise. Handle gather load internal
2420 (vectorizable_store): Update call to check_load_store_masking.
2421 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2422 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2423 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2424 (aarch64_gather_scale_operand_d): New predicates.
2425 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2426 (mask_gather_load<mode>): New insns.
2428 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2429 Alan Hayward <alan.hayward@arm.com>
2430 David Sherwood <david.sherwood@arm.com>
2432 * optabs.def (fold_left_plus_optab): New optab.
2433 * doc/md.texi (fold_left_plus_@var{m}): Document.
2434 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2435 * internal-fn.c (fold_left_direct): Define.
2436 (expand_fold_left_optab_fn): Likewise.
2437 (direct_fold_left_optab_supported_p): Likewise.
2438 * fold-const-call.c (fold_const_fold_left): New function.
2439 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2440 * tree-parloops.c (valid_reduction_p): New function.
2441 (gather_scalar_reductions): Use it.
2442 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2443 (vect_finish_replace_stmt): Declare.
2444 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2445 (needs_fold_left_reduction_p): New function, split out from...
2446 (vect_is_simple_reduction): ...here. Accept reductions that
2447 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2448 (vect_force_simple_reduction): Also store the reduction type in
2449 the assignment's STMT_VINFO_REDUC_TYPE.
2450 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2451 (merge_with_identity): New function.
2452 (vect_expand_fold_left): Likewise.
2453 (vectorize_fold_left_reduction): Likewise.
2454 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2455 scalar phi in place for it. Check for target support and reject
2456 cases that would reassociate the operation. Defer the transform
2457 phase to vectorize_fold_left_reduction.
2458 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2459 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2460 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2462 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2464 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2465 call to ifc_temp_var.
2467 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2468 Alan Hayward <alan.hayward@arm.com>
2469 David Sherwood <david.sherwood@arm.com>
2471 * target.def (legitimize_address_displacement): Take the original
2472 offset as a poly_int.
2473 * targhooks.h (default_legitimize_address_displacement): Update
2475 * targhooks.c (default_legitimize_address_displacement): Likewise.
2476 * doc/tm.texi: Regenerate.
2477 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2478 as an argument, moving assert of ad->disp == ad->disp_term to...
2479 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2480 Try calling targetm.legitimize_address_displacement before expanding
2481 the address rather than afterwards, and adjust for the new interface.
2482 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2483 Match the new hook interface. Handle SVE addresses.
2484 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2487 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2489 * Makefile.in (OBJS): Add early-remat.o.
2490 * target.def (select_early_remat_modes): New hook.
2491 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2492 * doc/tm.texi: Regenerate.
2493 * targhooks.h (default_select_early_remat_modes): Declare.
2494 * targhooks.c (default_select_early_remat_modes): New function.
2495 * timevar.def (TV_EARLY_REMAT): New timevar.
2496 * passes.def (pass_early_remat): New pass.
2497 * tree-pass.h (make_pass_early_remat): Declare.
2498 * early-remat.c: New file.
2499 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2501 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2503 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2504 Alan Hayward <alan.hayward@arm.com>
2505 David Sherwood <david.sherwood@arm.com>
2507 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2508 vfm1 with a bound_epilog parameter.
2509 (vect_do_peeling): Update calls accordingly, and move the prologue
2510 call earlier in the function. Treat the base bound_epilog as 0 for
2511 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2512 this base when peeling for gaps.
2513 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2514 with fully-masked loops.
2515 (vect_estimate_min_profitable_iters): Handle the single peeled
2516 iteration in that case.
2518 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2519 Alan Hayward <alan.hayward@arm.com>
2520 David Sherwood <david.sherwood@arm.com>
2522 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2523 single-element interleaving even if the size is not a power of 2.
2524 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2525 accesses for single-element interleaving if the group size is
2528 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2529 Alan Hayward <alan.hayward@arm.com>
2530 David Sherwood <david.sherwood@arm.com>
2532 * doc/md.texi (fold_extract_last_@var{m}): Document.
2533 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2534 * optabs.def (fold_extract_last_optab): New optab.
2535 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2536 * internal-fn.c (fold_extract_direct): New macro.
2537 (expand_fold_extract_optab_fn): Likewise.
2538 (direct_fold_extract_optab_supported_p): Likewise.
2539 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2540 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2541 EXTRACT_LAST_REDUCTION.
2542 (get_initial_def_for_reduction): Do not create an initial vector
2543 for EXTRACT_LAST_REDUCTION reductions.
2544 (vectorizable_reduction): Leave the scalar phi in place for
2545 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2546 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2547 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2548 transform phase to vectorizable_condition.
2549 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2551 (vect_finish_stmt_generation): ...here.
2552 (vect_finish_replace_stmt): New function.
2553 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2554 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2556 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2558 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2559 Alan Hayward <alan.hayward@arm.com>
2560 David Sherwood <david.sherwood@arm.com>
2562 * doc/md.texi (extract_last_@var{m}): Document.
2563 * optabs.def (extract_last_optab): New optab.
2564 * internal-fn.def (EXTRACT_LAST): New internal function.
2565 * internal-fn.c (cond_unary_direct): New macro.
2566 (expand_cond_unary_optab_fn): Likewise.
2567 (direct_cond_unary_optab_supported_p): Likewise.
2568 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2569 loops using EXTRACT_LAST.
2570 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2571 (extract_last_<mode>): ...this optab.
2572 (vec_extract<mode><Vel>): Update accordingly.
2574 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2575 Alan Hayward <alan.hayward@arm.com>
2576 David Sherwood <david.sherwood@arm.com>
2578 * target.def (empty_mask_is_expensive): New hook.
2579 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2580 * doc/tm.texi: Regenerate.
2581 * targhooks.h (default_empty_mask_is_expensive): Declare.
2582 * targhooks.c (default_empty_mask_is_expensive): New function.
2583 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2584 if the target says that empty masks are expensive.
2585 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2587 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2589 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2590 Alan Hayward <alan.hayward@arm.com>
2591 David Sherwood <david.sherwood@arm.com>
2593 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2594 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2595 (vect_use_loop_mask_for_alignment_p): New function.
2596 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2597 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2598 niters_skip argument. Make sure that the first niters_skip elements
2599 of the first iteration are inactive.
2600 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2601 Update call to vect_set_loop_masks_directly.
2602 (get_misalign_in_elems): New function, split out from...
2603 (vect_gen_prolog_loop_niters): ...here.
2604 (vect_update_init_of_dr): Take a code argument that specifies whether
2605 the adjustment should be added or subtracted.
2606 (vect_update_init_of_drs): Likewise.
2607 (vect_prepare_for_masked_peels): New function.
2608 (vect_do_peeling): Skip prologue peeling if we're using a mask
2609 instead. Update call to vect_update_inits_of_drs.
2610 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2612 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2613 alignment. Do not include the number of peeled iterations in
2614 the minimum threshold in that case.
2615 (vectorizable_induction): Adjust the start value down by
2616 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2617 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2618 Take the number of skipped iterations into account when calculating
2620 * tree-vect-stmts.c (vect_gen_while_not): New function.
2622 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2623 Alan Hayward <alan.hayward@arm.com>
2624 David Sherwood <david.sherwood@arm.com>
2626 * doc/sourcebuild.texi (vect_fully_masked): Document.
2627 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2629 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2631 (vect_analyze_loop_2): ...here. Don't check the vectorization
2632 factor against the number of loop iterations if the loop is
2635 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2636 Alan Hayward <alan.hayward@arm.com>
2637 David Sherwood <david.sherwood@arm.com>
2639 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2640 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2641 (dump_groups): Update accordingly.
2642 (iv_use::mem_type): New member variable.
2643 (address_p): New function.
2644 (record_use): Add a mem_type argument and initialize the new
2646 (record_group_use): Add a mem_type argument. Use address_p.
2647 Remove obsolete null checks of base_object. Update call to record_use.
2648 (find_interesting_uses_op): Update call to record_group_use.
2649 (find_interesting_uses_cond): Likewise.
2650 (find_interesting_uses_address): Likewise.
2651 (get_mem_type_for_internal_fn): New function.
2652 (find_address_like_use): Likewise.
2653 (find_interesting_uses_stmt): Try find_address_like_use before
2654 calling find_interesting_uses_op.
2655 (addr_offset_valid_p): Use the iv mem_type field as the type
2656 of the addressed memory.
2657 (add_autoinc_candidates): Likewise.
2658 (get_address_cost): Likewise.
2659 (split_small_address_groups_p): Use address_p.
2660 (split_address_groups): Likewise.
2661 (add_iv_candidate_for_use): Likewise.
2662 (autoinc_possible_for_pair): Likewise.
2663 (rewrite_groups): Likewise.
2664 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2665 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2666 (get_alias_ptr_type_for_ptr_address): New function.
2667 (rewrite_use_address): Rewrite address uses in calls that were
2668 identified by find_address_like_use.
2670 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2671 Alan Hayward <alan.hayward@arm.com>
2672 David Sherwood <david.sherwood@arm.com>
2674 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2676 * gimple-expr.h (is_gimple_addressable: Likewise.
2677 * gimple-expr.c (is_gimple_address): Likewise.
2678 * internal-fn.c (expand_call_mem_ref): New function.
2679 (expand_mask_load_optab_fn): Use it.
2680 (expand_mask_store_optab_fn): Likewise.
2682 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2683 Alan Hayward <alan.hayward@arm.com>
2684 David Sherwood <david.sherwood@arm.com>
2686 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2687 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2688 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2689 (cond_umax@var{mode}): Document.
2690 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2691 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2692 (cond_umin_optab, cond_umax_optab): New optabs.
2693 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2694 (COND_IOR, COND_XOR): New internal functions.
2695 * internal-fn.h (get_conditional_internal_fn): Declare.
2696 * internal-fn.c (cond_binary_direct): New macro.
2697 (expand_cond_binary_optab_fn): Likewise.
2698 (direct_cond_binary_optab_supported_p): Likewise.
2699 (get_conditional_internal_fn): New function.
2700 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2701 Cope with reduction statements that are vectorized as calls rather
2703 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2704 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2705 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2706 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2707 (UNSPEC_COND_EOR): New unspecs.
2708 (optab): Add mappings for them.
2709 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2710 (sve_int_op, sve_fp_op): New int attributes.
2712 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2713 Alan Hayward <alan.hayward@arm.com>
2714 David Sherwood <david.sherwood@arm.com>
2716 * optabs.def (while_ult_optab): New optab.
2717 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2718 * internal-fn.def (WHILE_ULT): New internal function.
2719 * internal-fn.h (direct_internal_fn_supported_p): New override
2720 that takes two types as argument.
2721 * internal-fn.c (while_direct): New macro.
2722 (expand_while_optab_fn): New function.
2723 (convert_optab_supported_p): Likewise.
2724 (direct_while_optab_supported_p): New macro.
2725 * wide-int.h (wi::udiv_ceil): New function.
2726 * tree-vectorizer.h (rgroup_masks): New structure.
2727 (vec_loop_masks): New typedef.
2728 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2730 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2731 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2732 (vect_max_vf): New function.
2733 (slpeel_make_loop_iterate_ntimes): Delete.
2734 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2735 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2736 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2737 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2738 internal-fn.h, stor-layout.h and optabs-query.h.
2739 (vect_set_loop_mask): New function.
2740 (add_preheader_seq): Likewise.
2741 (add_header_seq): Likewise.
2742 (interleave_supported_p): Likewise.
2743 (vect_maybe_permute_loop_masks): Likewise.
2744 (vect_set_loop_masks_directly): Likewise.
2745 (vect_set_loop_condition_masked): Likewise.
2746 (vect_set_loop_condition_unmasked): New function, split out from
2747 slpeel_make_loop_iterate_ntimes.
2748 (slpeel_make_loop_iterate_ntimes): Rename to..
2749 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2750 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2751 (vect_do_peeling): Update call accordingly.
2752 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2754 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2755 mask_compare_type, can_fully_mask_p and fully_masked_p.
2756 (release_vec_loop_masks): New function.
2757 (_loop_vec_info): Use it to free the loop masks.
2758 (can_produce_all_loop_masks_p): New function.
2759 (vect_get_max_nscalars_per_iter): Likewise.
2760 (vect_verify_full_masking): Likewise.
2761 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2762 retries, and free the mask rgroups before retrying. Check loop-wide
2763 reasons for disallowing fully-masked loops. Make the final decision
2764 about whether use a fully-masked loop or not.
2765 (vect_estimate_min_profitable_iters): Do not assume that peeling
2766 for the number of iterations will be needed for fully-masked loops.
2767 (vectorizable_reduction): Disable fully-masked loops.
2768 (vectorizable_live_operation): Likewise.
2769 (vect_halve_mask_nunits): New function.
2770 (vect_double_mask_nunits): Likewise.
2771 (vect_record_loop_mask): Likewise.
2772 (vect_get_loop_mask): Likewise.
2773 (vect_transform_loop): Handle the case in which the final loop
2774 iteration might handle a partial vector. Call vect_set_loop_condition
2775 instead of slpeel_make_loop_iterate_ntimes.
2776 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2777 (check_load_store_masking): New function.
2778 (prepare_load_store_mask): Likewise.
2779 (vectorizable_store): Handle fully-masked loops.
2780 (vectorizable_load): Likewise.
2781 (supportable_widening_operation): Use vect_halve_mask_nunits for
2783 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2784 (vect_gen_while): New function.
2785 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2786 (aarch64_uqdec<mode>): New insn.
2788 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2789 Alan Hayward <alan.hayward@arm.com>
2790 David Sherwood <david.sherwood@arm.com>
2792 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2793 (reduc_xor_scal_optab): New optabs.
2794 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2795 (reduc_xor_scal_@var{m}): Document.
2796 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2797 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2799 * fold-const-call.c (fold_const_call): Handle them.
2800 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2801 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2802 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2803 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2804 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2805 (UNSPEC_XORV): New unspecs.
2806 (optab): Add entries for them.
2807 (BITWISEV): New int iterator.
2808 (bit_reduc_op): New int attributes.
2810 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2811 Alan Hayward <alan.hayward@arm.com>
2812 David Sherwood <david.sherwood@arm.com>
2814 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2815 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2816 * optabs.def (vec_shl_insert_optab): New optab.
2817 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2818 (duplicate_and_interleave): Likewise.
2819 * tree-vect-loop.c: Include internal-fn.h.
2820 (neutral_op_for_slp_reduction): New function, split out from
2821 get_initial_defs_for_reduction.
2822 (get_initial_def_for_reduction): Handle option 2 for variable-length
2823 vectors by loading the neutral value into a vector and then shifting
2824 the initial value into element 0.
2825 (get_initial_defs_for_reduction): Replace the code argument with
2826 the neutral value calculated by neutral_op_for_slp_reduction.
2827 Use gimple_build_vector for constant-length vectors.
2828 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2829 but the first group_size elements have a neutral value.
2830 Use duplicate_and_interleave otherwise.
2831 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2832 Update call to get_initial_defs_for_reduction. Handle SLP
2833 reductions for variable-length vectors by creating one vector
2834 result for each scalar result, with the elements associated
2835 with other scalar results stubbed out with the neutral value.
2836 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2837 Require IFN_VEC_SHL_INSERT for double reductions on
2838 variable-length vectors, or SLP reductions that have
2839 a neutral value. Require can_duplicate_and_interleave_p
2840 support for variable-length unchained SLP reductions if there
2841 is no neutral value, such as for MIN/MAX reductions. Also require
2842 the number of vector elements to be a multiple of the number of
2843 SLP statements when doing variable-length unchained SLP reductions.
2844 Update call to vect_create_epilog_for_reduction.
2845 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2846 and remove initial values.
2847 (duplicate_and_interleave): Make public.
2848 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2849 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2851 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2852 Alan Hayward <alan.hayward@arm.com>
2853 David Sherwood <david.sherwood@arm.com>
2855 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2856 (can_duplicate_and_interleave_p): New function.
2857 (vect_get_and_check_slp_defs): Take the vector of statements
2858 rather than just the current one. Remove excess parentheses.
2859 Restriction rejectinon of vect_constant_def and vect_external_def
2860 for variable-length vectors to boolean types, or types for which
2861 can_duplicate_and_interleave_p is false.
2862 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2863 (duplicate_and_interleave): New function.
2864 (vect_get_constant_vectors): Use gimple_build_vector for
2865 constant-length vectors and suitable variable-length constant
2866 vectors. Use duplicate_and_interleave for other variable-length
2867 vectors. Don't defer the update when inserting new statements.
2869 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2870 Alan Hayward <alan.hayward@arm.com>
2871 David Sherwood <david.sherwood@arm.com>
2873 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2874 min_profitable_iters doesn't go negative.
2876 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2877 Alan Hayward <alan.hayward@arm.com>
2878 David Sherwood <david.sherwood@arm.com>
2880 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2881 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2882 * optabs.def (vec_mask_load_lanes_optab): New optab.
2883 (vec_mask_store_lanes_optab): Likewise.
2884 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2885 (MASK_STORE_LANES): Likewise.
2886 * internal-fn.c (mask_load_lanes_direct): New macro.
2887 (mask_store_lanes_direct): Likewise.
2888 (expand_mask_load_optab_fn): Handle masked operations.
2889 (expand_mask_load_lanes_optab_fn): New macro.
2890 (expand_mask_store_optab_fn): Handle masked operations.
2891 (expand_mask_store_lanes_optab_fn): New macro.
2892 (direct_mask_load_lanes_optab_supported_p): Likewise.
2893 (direct_mask_store_lanes_optab_supported_p): Likewise.
2894 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2896 (vect_load_lanes_supported): Likewise.
2897 * tree-vect-data-refs.c (strip_conversion): New function.
2898 (can_group_stmts_p): Likewise.
2899 (vect_analyze_data_ref_accesses): Use it instead of checking
2900 for a pair of assignments.
2901 (vect_store_lanes_supported): Take a masked_p parameter.
2902 (vect_load_lanes_supported): Likewise.
2903 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2904 vect_store_lanes_supported and vect_load_lanes_supported.
2905 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2906 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2907 parameter. Don't allow gaps for masked accesses.
2908 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2909 and vect_load_lanes_supported.
2910 (get_load_store_type): Take a masked_p parameter and update
2911 call to get_group_load_store_type.
2912 (vectorizable_store): Update call to get_load_store_type.
2913 Handle IFN_MASK_STORE_LANES.
2914 (vectorizable_load): Update call to get_load_store_type.
2915 Handle IFN_MASK_LOAD_LANES.
2917 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2918 Alan Hayward <alan.hayward@arm.com>
2919 David Sherwood <david.sherwood@arm.com>
2921 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2923 * config/aarch64/aarch64-protos.h
2924 (aarch64_sve_struct_memory_operand_p): Declare.
2925 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2926 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2927 (VPRED, vpred): Handle SVE structure modes.
2928 * config/aarch64/constraints.md (Utx): New constraint.
2929 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2930 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2931 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2932 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2933 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2934 structure modes. Split into pieces after RA.
2935 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2936 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2938 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2939 SVE structure modes.
2940 (aarch64_classify_address): Likewise.
2941 (sizetochar): Move earlier in file.
2942 (aarch64_print_operand): Handle SVE register lists.
2943 (aarch64_array_mode): New function.
2944 (aarch64_sve_struct_memory_operand_p): Likewise.
2945 (TARGET_ARRAY_MODE): Redefine.
2947 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2948 Alan Hayward <alan.hayward@arm.com>
2949 David Sherwood <david.sherwood@arm.com>
2951 * target.def (array_mode): New target hook.
2952 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2953 * doc/tm.texi: Regenerate.
2954 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2955 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2956 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2958 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2961 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2962 Alan Hayward <alan.hayward@arm.com>
2963 David Sherwood <david.sherwood@arm.com>
2965 * fold-const.c (fold_binary_loc): Check the argument types
2966 rather than the result type when testing for a vector operation.
2968 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2970 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2971 * doc/tm.texi: Regenerate.
2973 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2974 Alan Hayward <alan.hayward@arm.com>
2975 David Sherwood <david.sherwood@arm.com>
2977 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2978 (sve): Document new AArch64 extension.
2979 * doc/md.texi (w): Extend the description of the AArch64
2980 constraint to include SVE vectors.
2981 (Upl, Upa): Document new AArch64 predicate constraints.
2982 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2984 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2985 (msve-vector-bits=): New option.
2986 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2987 SVE when these are disabled.
2988 (sve): New extension.
2989 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2990 modes. Adjust their number of units based on aarch64_sve_vg.
2991 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2992 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2993 aarch64_addr_query_type.
2994 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2995 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2996 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2997 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2998 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2999 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3000 (aarch64_simd_imm_zero_p): Delete.
3001 (aarch64_check_zero_based_sve_index_immediate): Declare.
3002 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3003 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3004 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3005 (aarch64_sve_float_mul_immediate_p): Likewise.
3006 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3008 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3009 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3010 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3011 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3012 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3013 (aarch64_regmode_natural_size): Likewise.
3014 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3015 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3017 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3018 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3019 for VG and the SVE predicate registers.
3020 (V_ALIASES): Add a "z"-prefixed alias.
3021 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3022 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3023 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3024 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3025 (REG_CLASS_NAMES): Add entries for them.
3026 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3027 and the predicate registers.
3028 (aarch64_sve_vg): Declare.
3029 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3030 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3031 (REGMODE_NATURAL_SIZE): Define.
3032 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3034 * config/aarch64/aarch64.c: Include cfgrtl.h.
3035 (simd_immediate_info): Add a constructor for series vectors,
3036 and an associated step field.
3037 (aarch64_sve_vg): New variable.
3038 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3039 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3040 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3041 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3042 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3043 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3044 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3045 (aarch64_get_mask_mode): New functions.
3046 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3047 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3048 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3049 predicate modes and predicate registers. Explicitly restrict
3050 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3051 to store a vector mode if it is recognized by
3052 aarch64_classify_vector_mode.
3053 (aarch64_regmode_natural_size): New function.
3054 (aarch64_hard_regno_caller_save_mode): Return the original mode
3056 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3057 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3058 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3059 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3061 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3062 does not overlap dest if the function is frame-related. Handle
3064 (aarch64_split_add_offset): New function.
3065 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3066 them aarch64_add_offset.
3067 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3068 and update call to aarch64_sub_sp.
3069 (aarch64_add_cfa_expression): New function.
3070 (aarch64_expand_prologue): Pass extra temporary registers to the
3071 functions above. Handle the case in which we need to emit new
3072 DW_CFA_expressions for registers that were originally saved
3073 relative to the stack pointer, but now have to be expressed
3074 relative to the frame pointer.
3075 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3077 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3078 IP0 and IP1 values for SVE frames.
3079 (aarch64_expand_vec_series): New function.
3080 (aarch64_expand_sve_widened_duplicate): Likewise.
3081 (aarch64_expand_sve_const_vector): Likewise.
3082 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3083 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3084 into the register, rather than emitting a SET directly.
3085 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3086 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3087 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3088 (offset_9bit_signed_scaled_p): New functions.
3089 (aarch64_replicate_bitmask_imm): New function.
3090 (aarch64_bitmask_imm): Use it.
3091 (aarch64_cannot_force_const_mem): Reject expressions involving
3092 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3093 (aarch64_classify_index): Handle SVE indices, by requiring
3094 a plain register index with a scale that matches the element size.
3095 (aarch64_classify_address): Handle SVE addresses. Assert that
3096 the mode of the address is VOIDmode or an integer mode.
3097 Update call to aarch64_classify_symbol.
3098 (aarch64_classify_symbolic_expression): Update call to
3099 aarch64_classify_symbol.
3100 (aarch64_const_vec_all_in_range_p): New function.
3101 (aarch64_print_vector_float_operand): Likewise.
3102 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3103 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3104 and the FP immediates 1.0 and 0.5.
3105 (aarch64_print_address_internal): Handle SVE addresses.
3106 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3107 (aarch64_regno_regclass): Handle predicate registers.
3108 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3110 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3111 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3112 (aarch64_convert_sve_vector_bits): New function.
3113 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3114 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3116 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3117 Handle SVE vector and predicate modes. Accept VL-based constants
3118 that need only one temporary register, and VL offsets that require
3119 no temporary registers.
3120 (aarch64_conditional_register_usage): Mark the predicate registers
3121 as fixed if SVE isn't available.
3122 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3123 Return true for SVE vector and predicate modes.
3124 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3125 rather than an unsigned int. Handle SVE modes.
3126 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3128 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3130 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3131 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3132 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3133 (aarch64_sve_float_mul_immediate_p): New functions.
3134 (aarch64_sve_valid_immediate): New function.
3135 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3136 Explicitly reject structure modes. Check for INDEX constants.
3137 Handle PTRUE and PFALSE constants.
3138 (aarch64_check_zero_based_sve_index_immediate): New function.
3139 (aarch64_simd_imm_zero_p): Delete.
3140 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3141 vector modes. Accept constants in the range of CNT[BHWD].
3142 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3143 ask for an Advanced SIMD mode.
3144 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3145 (aarch64_simd_vector_alignment): Handle SVE predicates.
3146 (aarch64_vectorize_preferred_vector_alignment): New function.
3147 (aarch64_simd_vector_alignment_reachable): Use it instead of
3149 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3150 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3152 (MAX_VECT_LEN): Delete.
3153 (expand_vec_perm_d): Add a vec_flags field.
3154 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3155 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3156 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3158 (aarch64_evpc_rev): Rename to...
3159 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3160 (aarch64_evpc_rev_global): New function.
3161 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3162 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3164 (aarch64_evpc_sve_tbl): New function.
3165 (aarch64_expand_vec_perm_const_1): Update after rename of
3166 aarch64_evpc_rev. Handle SVE permutes too, trying
3167 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3168 than aarch64_evpc_tbl.
3169 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3170 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3171 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3172 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3173 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3174 (aarch64_expand_sve_vcond): New functions.
3175 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3176 of aarch64_vector_mode_p.
3177 (aarch64_dwarf_poly_indeterminate_value): New function.
3178 (aarch64_compute_pressure_classes): Likewise.
3179 (aarch64_can_change_mode_class): Likewise.
3180 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3181 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3182 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3183 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3184 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3185 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3186 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3187 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3189 (Dn, Dl, Dr): Accept const as well as const_vector.
3190 (Dz): Likewise. Compare against CONST0_RTX.
3191 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3192 of "vector" where appropriate.
3193 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3194 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3195 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3196 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3197 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3198 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3199 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3200 (v_int_equiv): Extend to SVE modes.
3201 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3203 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3204 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3205 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3206 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3207 (SVE_COND_FP_CMP): New int iterators.
3208 (perm_hilo): Handle the new unpack unspecs.
3209 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3211 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3212 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3213 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3214 (aarch64_equality_operator, aarch64_constant_vector_operand)
3215 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3216 (aarch64_sve_nonimmediate_operand): Likewise.
3217 (aarch64_sve_general_operand): Likewise.
3218 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3219 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3220 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3221 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3222 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3223 (aarch64_sve_float_arith_immediate): Likewise.
3224 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3225 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3226 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3227 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3228 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3229 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3230 (aarch64_sve_float_arith_operand): Likewise.
3231 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3232 (aarch64_sve_float_mul_operand): Likewise.
3233 (aarch64_sve_vec_perm_operand): Likewise.
3234 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3235 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3236 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3237 as well as const_vector.
3238 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3239 in file. Use CONST0_RTX and CONSTM1_RTX.
3240 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3241 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3242 Use aarch64_simd_imm_zero.
3243 * config/aarch64/aarch64-sve.md: New file.
3244 * config/aarch64/aarch64.md: Include it.
3245 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3246 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3247 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3248 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3249 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3250 (sve): New attribute.
3251 (enabled): Disable instructions with the sve attribute unless
3253 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3254 aarch64_expand_mov_immediate.
3255 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3256 CNT[BHSD] immediates.
3257 (movti): Split CONST_POLY_INT moves into two halves.
3258 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3259 Split additions that need a temporary here if the destination
3260 is the stack pointer.
3261 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3262 (*add<mode>3_poly_1): New instruction.
3263 (set_clobber_cc): New expander.
3265 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3267 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3268 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3269 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3270 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3271 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3272 Change innermode from fixed_mode_size to machine_mode.
3273 (simplify_subreg): Update call accordingly. Handle a constant-sized
3274 subreg of a variable-length CONST_VECTOR.
3276 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3277 Alan Hayward <alan.hayward@arm.com>
3278 David Sherwood <david.sherwood@arm.com>
3280 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3281 (add_offset_to_base): New function, split out from...
3282 (create_mem_ref): ...here. When handling a scale other than 1,
3283 check first whether the address is valid without the offset.
3284 Add it into the base if so, leaving the index and scale as-is.
3286 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3289 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3290 fold_for_warn before checking if arg2 is INTEGER_CST.
3292 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3294 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3295 (store_multiple_operation): Delete.
3296 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3297 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3298 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3299 guarded by TARGET_STRING.
3300 (rs6000_output_load_multiple): Delete.
3301 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3302 OPTION_MASK_STRING / TARGET_STRING handling.
3303 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3304 (const rs6000_opt_masks) <"string">: Change mask to 0.
3305 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3306 (MASK_STRING): Delete.
3307 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3309 (load_multiple): Delete.
3316 (store_multiple): Delete.
3323 (movmemsi_8reg): Delete.
3324 (corresponding unnamed define_insn): Delete.
3325 (movmemsi_6reg): Delete.
3326 (corresponding unnamed define_insn): Delete.
3327 (movmemsi_4reg): Delete.
3328 (corresponding unnamed define_insn): Delete.
3329 (movmemsi_2reg): Delete.
3330 (corresponding unnamed define_insn): Delete.
3331 (movmemsi_1reg): Delete.
3332 (corresponding unnamed define_insn): Delete.
3333 * config/rs6000/rs6000.opt (mno-string): New.
3334 (mstring): Replace by deprecation warning stub.
3335 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3337 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3339 * regrename.c (regrename_do_replace): If replacing the same
3340 reg multiple times, try to reuse last created gen_raw_REG.
3343 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3344 main to workaround a bug in GDB.
3346 2018-01-12 Tom de Vries <tom@codesourcery.com>
3349 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3351 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3353 PR rtl-optimization/80481
3354 * ira-color.c (get_cap_member): New function.
3355 (allocnos_conflict_by_live_ranges_p): Use it.
3356 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3357 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3359 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3362 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3363 (*saddl_se_1): Ditto.
3365 (*ssubl_se_1): Ditto.
3367 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3369 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3370 rather than wi::to_widest for DR_INITs.
3371 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3372 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3373 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3375 (vect_analyze_group_access_1): Note that here.
3377 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3379 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3380 polynomial type sizes.
3382 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3384 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3385 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3386 (gimple_add_tmp_var): Likewise.
3388 2018-01-12 Martin Liska <mliska@suse.cz>
3390 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3391 (gimple_alloc_sizes): Likewise.
3392 (dump_gimple_statistics): Use PRIu64 in printf format.
3393 * gimple.h: Change uint64_t to int.
3395 2018-01-12 Martin Liska <mliska@suse.cz>
3397 * tree-core.h: Use uint64_t instead of int.
3398 * tree.c (tree_node_counts): Likewise.
3399 (tree_node_sizes): Likewise.
3400 (dump_tree_statistics): Use PRIu64 in printf format.
3402 2018-01-12 Martin Liska <mliska@suse.cz>
3404 * Makefile.in: As qsort_chk is implemented in vec.c, add
3405 vec.o to linkage of gencfn-macros.
3406 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3407 passing the info to record_node_allocation_statistics.
3408 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3410 * ggc-common.c (struct ggc_usage): Add operator== and use
3411 it in operator< and compare function.
3412 * mem-stats.h (struct mem_usage): Likewise.
3413 * vec.c (struct vec_usage): Remove operator< and compare
3414 function. Can be simply inherited.
3416 2018-01-12 Martin Jambor <mjambor@suse.cz>
3419 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3420 * tree-ssa-math-opts.c: Include domwalk.h.
3421 (convert_mult_to_fma_1): New function.
3422 (fma_transformation_info): New type.
3423 (fma_deferring_state): Likewise.
3424 (cancel_fma_deferring): New function.
3425 (result_of_phi): Likewise.
3426 (last_fma_candidate_feeds_initial_phi): Likewise.
3427 (convert_mult_to_fma): Added deferring logic, split actual
3428 transformation to convert_mult_to_fma_1.
3429 (math_opts_dom_walker): New type.
3430 (math_opts_dom_walker::after_dom_children): New method, body moved
3431 here from pass_optimize_widening_mul::execute, added deferring logic
3433 (pass_optimize_widening_mul::execute): Moved most of code to
3434 math_opts_dom_walker::after_dom_children.
3435 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3436 * config/i386/i386.c (ix86_option_override_internal): Added
3437 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3439 2018-01-12 Richard Biener <rguenther@suse.de>
3442 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3443 inline instance vars.
3445 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3448 * config/rx/rx.c (rx_is_restricted_memory_address):
3451 2018-01-12 Richard Biener <rguenther@suse.de>
3453 PR tree-optimization/80846
3454 * target.def (split_reduction): New target hook.
3455 * targhooks.c (default_split_reduction): New function.
3456 * targhooks.h (default_split_reduction): Declare.
3457 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3458 target requests first reduce vectors by combining low and high
3460 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3461 (get_vectype_for_scalar_type_and_size): Export.
3462 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3463 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3464 * doc/tm.texi: Regenerate.
3465 * config/i386/i386.c (ix86_split_reduction): Implement
3466 TARGET_VECTORIZE_SPLIT_REDUCTION.
3468 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3471 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3472 in PIC mode except for TARGET_VXWORKS_RTP.
3473 * config/sparc/sparc.c: Include cfgrtl.h.
3474 (TARGET_INIT_PIC_REG): Define.
3475 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3476 (sparc_pic_register_p): New predicate.
3477 (sparc_legitimate_address_p): Use it.
3478 (sparc_legitimize_pic_address): Likewise.
3479 (sparc_delegitimize_address): Likewise.
3480 (sparc_mode_dependent_address_p): Likewise.
3481 (gen_load_pcrel_sym): Remove 4th parameter.
3482 (load_got_register): Adjust call to above. Remove obsolete stuff.
3483 (sparc_expand_prologue): Do not call load_got_register here.
3484 (sparc_flat_expand_prologue): Likewise.
3485 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3486 (sparc_use_pseudo_pic_reg): New function.
3487 (sparc_init_pic_reg): Likewise.
3488 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3489 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3491 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3493 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3494 Add item for branch_cost.
3496 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3498 PR rtl-optimization/83565
3499 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3500 not extend the result to a larger mode for rotate operations.
3501 (num_sign_bit_copies1): Likewise.
3503 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3506 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3508 Use values-Xc.o for -pedantic.
3509 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3511 2018-01-12 Martin Liska <mliska@suse.cz>
3514 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3516 (possible_polymorphic_call_targets): Use it.
3517 (ipa_devirt): Likewise.
3519 2018-01-12 Martin Liska <mliska@suse.cz>
3521 * profile-count.h (enum profile_quality): Use 0 as invalid
3522 enum value of profile_quality.
3524 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3526 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3527 -mext-string options.
3529 2018-01-12 Richard Biener <rguenther@suse.de>
3531 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3532 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3533 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3535 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3537 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3539 * configure.ac (--with-long-double-format): Add support for the
3540 configuration option to change the default long double format on
3542 * config.gcc (powerpc*-linux*-*): Likewise.
3543 * configure: Regenerate.
3544 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3545 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3546 used without modification.
3548 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3550 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3551 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3552 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3553 MISC_BUILTIN_SPEC_BARRIER.
3554 (rs6000_init_builtins): Likewise.
3555 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3557 (speculation_barrier): New define_insn.
3558 * doc/extend.texi: Document __builtin_speculation_barrier.
3560 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3563 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3564 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3565 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3567 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3568 integral modes instead of "ss" and "sd".
3569 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3570 vectors with 32-bit and 64-bit elements.
3571 (vecdupssescalarmodesuffix): New mode attribute.
3572 (vec_dup<mode>): Use it.
3574 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3577 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3578 frame if argument is passed on stack.
3580 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3583 * ree.c (combine_reaching_defs): Optimize also
3584 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3585 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3587 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3590 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3592 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3595 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3596 after they are computed.
3598 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3600 PR tree-optimization/83695
3601 * gimple-loop-linterchange.cc
3602 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3603 reset cached scev information after interchange.
3604 (pass_linterchange::execute): Remove call to scev_reset_htab.
3606 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3608 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3609 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3610 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3611 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3612 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3613 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3614 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3615 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3616 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3617 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3618 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3619 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3620 (V_lane_reg): Likewise.
3621 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3623 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3624 (vfmal_lane_low<mode>_intrinsic,
3625 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3626 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3627 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3628 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3629 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3630 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3632 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3634 * config/arm/arm-cpus.in (fp16fml): New feature.
3635 (ALL_SIMD): Add fp16fml.
3636 (armv8.2-a): Add fp16fml as an option.
3637 (armv8.3-a): Likewise.
3638 (armv8.4-a): Add fp16fml as part of fp16.
3639 * config/arm/arm.h (TARGET_FP16FML): Define.
3640 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3642 * config/arm/arm-modes.def (V2HF): Define.
3643 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3644 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3645 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3646 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3647 vfmsl_low, vfmsl_high): New set of builtins.
3648 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3649 (vfml_op): New code attribute.
3650 (VFMLHALVES): New int iterator.
3651 (VFML, VFMLSEL): New mode attributes.
3652 (V_reg): Define mapping for V2HF.
3653 (V_hi, V_lo): New mode attributes.
3654 (VF_constraint): Likewise.
3655 (vfml_half, vfml_half_selector): New int attributes.
3656 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3658 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3659 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3661 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3662 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3663 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3664 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3666 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3667 Document new effective target and option set.
3669 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3671 * config/arm/arm-cpus.in (armv8_4): New feature.
3672 (ARMv8_4a): New fgroup.
3673 (armv8.4-a): New arch.
3674 * config/arm/arm-tables.opt: Regenerate.
3675 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3676 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3677 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3678 Add matching rules for -march=armv8.4-a and extensions.
3679 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3681 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3684 * config/rx/rx.md (BW): New mode attribute.
3685 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3687 2018-01-11 Richard Biener <rguenther@suse.de>
3689 PR tree-optimization/83435
3690 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3691 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3692 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3694 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3695 Alan Hayward <alan.hayward@arm.com>
3696 David Sherwood <david.sherwood@arm.com>
3698 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3700 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3701 (aarch64_print_address_internal): Use it to check for a zero offset.
3703 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3704 Alan Hayward <alan.hayward@arm.com>
3705 David Sherwood <david.sherwood@arm.com>
3707 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3708 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3709 Return a poly_int64 rather than a HOST_WIDE_INT.
3710 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3711 rather than a HOST_WIDE_INT.
3712 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3713 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3714 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3715 final_offset from HOST_WIDE_INT to poly_int64.
3716 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3717 to_constant when getting the number of units in an Advanced SIMD
3719 (aarch64_builtin_vectorized_function): Check for a constant number
3721 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3723 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3724 attribute instead of GET_MODE_NUNITS.
3725 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3726 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3727 GET_MODE_SIZE for fixed-size registers.
3728 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3729 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3730 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3731 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3732 (aarch64_print_operand, aarch64_print_address_internal)
3733 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3734 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3735 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3736 Handle polynomial GET_MODE_SIZE.
3737 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3738 wider than SImode without modification.
3739 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3740 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3741 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3742 passing and returning SVE modes.
3743 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3744 rather than GEN_INT.
3745 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3746 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3747 (aarch64_allocate_and_probe_stack_space): Likewise.
3748 (aarch64_layout_frame): Cope with polynomial offsets.
3749 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3750 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3752 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3753 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3754 poly_int64 rather than a HOST_WIDE_INT.
3755 (aarch64_get_separate_components, aarch64_process_components)
3756 (aarch64_expand_prologue, aarch64_expand_epilogue)
3757 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3758 (aarch64_anchor_offset): New function, split out from...
3759 (aarch64_legitimize_address): ...here.
3760 (aarch64_builtin_vectorization_cost): Handle polynomial
3761 TYPE_VECTOR_SUBPARTS.
3762 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3764 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3765 number of elements from the PARALLEL rather than the mode.
3766 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3767 rather than GET_MODE_BITSIZE.
3768 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3769 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3770 (aarch64_expand_vec_perm_const_1): Handle polynomial
3771 d->perm.length () and d->perm elements.
3772 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3773 Apply to_constant to d->perm elements.
3774 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3775 polynomial CONST_VECTOR_NUNITS.
3776 (aarch64_move_pointer): Take amount as a poly_int64 rather
3778 (aarch64_progress_pointer): Avoid temporary variable.
3779 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3780 the mode attribute instead of GET_MODE.
3782 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3783 Alan Hayward <alan.hayward@arm.com>
3784 David Sherwood <david.sherwood@arm.com>
3786 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3787 x exists before using it.
3788 (aarch64_add_constant_internal): Rename to...
3789 (aarch64_add_offset_1): ...this. Replace regnum with separate
3790 src and dest rtxes. Handle the case in which they're different,
3791 including when the offset is zero. Replace scratchreg with an rtx.
3792 Use 2 additions if there is no spare register into which we can
3793 move a 16-bit constant.
3794 (aarch64_add_constant): Delete.
3795 (aarch64_add_offset): Replace reg with separate src and dest
3796 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3797 Use aarch64_add_offset_1.
3798 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3799 an rtx rather than an int. Take the delta as a poly_int64
3800 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3801 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3802 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3803 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3804 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3806 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3807 aarch64_add_constant.
3809 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3811 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3812 Use scalar_float_mode.
3814 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3816 * config/aarch64/aarch64-simd.md
3817 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3818 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3819 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3820 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3821 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3822 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3823 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3824 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3825 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3826 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3828 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3831 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3832 targ_options->x_arm_arch_string is non NULL.
3834 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3836 * config/aarch64/aarch64.h
3837 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3839 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3842 * expmed.c (emit_store_flag_force): Swap if const op0
3843 and change VOIDmode to mode of op0.
3845 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3847 PR rtl-optimization/83761
3848 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3849 than bytes to mode_for_size.
3851 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3854 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3855 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3858 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3861 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3862 when in layout mode.
3863 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3864 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3867 2018-01-10 Michael Collison <michael.collison@arm.com>
3869 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3870 * config/aarch64/aarch64-option-extension.def: Add
3871 AARCH64_OPT_EXTENSION of 'fp16fml'.
3872 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3873 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3874 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3875 * config/aarch64/constraints.md (Ui7): New constraint.
3876 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3877 (VFMLA_SEL_W): Ditto.
3880 (VFMLA16_LOW): New int iterator.
3881 (VFMLA16_HIGH): Ditto.
3882 (UNSPEC_FMLAL): New unspec.
3883 (UNSPEC_FMLSL): Ditto.
3884 (UNSPEC_FMLAL2): Ditto.
3885 (UNSPEC_FMLSL2): Ditto.
3886 (f16mac): New code attribute.
3887 * config/aarch64/aarch64-simd-builtins.def
3888 (aarch64_fmlal_lowv2sf): Ditto.
3889 (aarch64_fmlsl_lowv2sf): Ditto.
3890 (aarch64_fmlalq_lowv4sf): Ditto.
3891 (aarch64_fmlslq_lowv4sf): Ditto.
3892 (aarch64_fmlal_highv2sf): Ditto.
3893 (aarch64_fmlsl_highv2sf): Ditto.
3894 (aarch64_fmlalq_highv4sf): Ditto.
3895 (aarch64_fmlslq_highv4sf): Ditto.
3896 (aarch64_fmlal_lane_lowv2sf): Ditto.
3897 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3898 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3899 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3900 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3901 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3902 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3903 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3904 (aarch64_fmlal_lane_highv2sf): Ditto.
3905 (aarch64_fmlsl_lane_highv2sf): Ditto.
3906 (aarch64_fmlal_laneq_highv2sf): Ditto.
3907 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3908 (aarch64_fmlalq_lane_highv4sf): Ditto.
3909 (aarch64_fmlsl_lane_highv4sf): Ditto.
3910 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3911 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3912 * config/aarch64/aarch64-simd.md:
3913 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3914 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3915 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3916 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3917 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3918 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3919 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3920 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3921 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3922 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3923 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3924 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3925 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3926 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3927 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3928 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3929 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3930 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3931 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3932 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3933 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3934 (vfmlsl_low_u32): Ditto.
3935 (vfmlalq_low_u32): Ditto.
3936 (vfmlslq_low_u32): Ditto.
3937 (vfmlal_high_u32): Ditto.
3938 (vfmlsl_high_u32): Ditto.
3939 (vfmlalq_high_u32): Ditto.
3940 (vfmlslq_high_u32): Ditto.
3941 (vfmlal_lane_low_u32): Ditto.
3942 (vfmlsl_lane_low_u32): Ditto.
3943 (vfmlal_laneq_low_u32): Ditto.
3944 (vfmlsl_laneq_low_u32): Ditto.
3945 (vfmlalq_lane_low_u32): Ditto.
3946 (vfmlslq_lane_low_u32): Ditto.
3947 (vfmlalq_laneq_low_u32): Ditto.
3948 (vfmlslq_laneq_low_u32): Ditto.
3949 (vfmlal_lane_high_u32): Ditto.
3950 (vfmlsl_lane_high_u32): Ditto.
3951 (vfmlal_laneq_high_u32): Ditto.
3952 (vfmlsl_laneq_high_u32): Ditto.
3953 (vfmlalq_lane_high_u32): Ditto.
3954 (vfmlslq_lane_high_u32): Ditto.
3955 (vfmlalq_laneq_high_u32): Ditto.
3956 (vfmlslq_laneq_high_u32): Ditto.
3957 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3958 (AARCH64_FL_FOR_ARCH8_4): New.
3959 (AARCH64_ISA_F16FML): New ISA flag.
3960 (TARGET_F16FML): New feature flag for fp16fml.
3961 (doc/invoke.texi): Document new fp16fml option.
3963 2018-01-10 Michael Collison <michael.collison@arm.com>
3965 * config/aarch64/aarch64-builtins.c:
3966 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3967 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3968 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3969 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3970 (AARCH64_ISA_SHA3): New ISA flag.
3971 (TARGET_SHA3): New feature flag for sha3.
3972 * config/aarch64/iterators.md (sha512_op): New int attribute.
3973 (CRYPTO_SHA512): New int iterator.
3974 (UNSPEC_SHA512H): New unspec.
3975 (UNSPEC_SHA512H2): Ditto.
3976 (UNSPEC_SHA512SU0): Ditto.
3977 (UNSPEC_SHA512SU1): Ditto.
3978 * config/aarch64/aarch64-simd-builtins.def
3979 (aarch64_crypto_sha512hqv2di): New builtin.
3980 (aarch64_crypto_sha512h2qv2di): Ditto.
3981 (aarch64_crypto_sha512su0qv2di): Ditto.
3982 (aarch64_crypto_sha512su1qv2di): Ditto.
3983 (aarch64_eor3qv8hi): Ditto.
3984 (aarch64_rax1qv2di): Ditto.
3985 (aarch64_xarqv2di): Ditto.
3986 (aarch64_bcaxqv8hi): Ditto.
3987 * config/aarch64/aarch64-simd.md:
3988 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3989 (aarch64_crypto_sha512su0qv2di): Ditto.
3990 (aarch64_crypto_sha512su1qv2di): Ditto.
3991 (aarch64_eor3qv8hi): Ditto.
3992 (aarch64_rax1qv2di): Ditto.
3993 (aarch64_xarqv2di): Ditto.
3994 (aarch64_bcaxqv8hi): Ditto.
3995 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3996 (vsha512h2q_u64): Ditto.
3997 (vsha512su0q_u64): Ditto.
3998 (vsha512su1q_u64): Ditto.
3999 (veor3q_u16): Ditto.
4000 (vrax1q_u64): Ditto.
4002 (vbcaxq_u16): Ditto.
4003 * config/arm/types.md (crypto_sha512): New type attribute.
4004 (crypto_sha3): Ditto.
4005 (doc/invoke.texi): Document new sha3 option.
4007 2018-01-10 Michael Collison <michael.collison@arm.com>
4009 * config/aarch64/aarch64-builtins.c:
4010 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4011 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4012 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4013 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4014 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4015 (AARCH64_ISA_SM4): New ISA flag.
4016 (TARGET_SM4): New feature flag for sm4.
4017 * config/aarch64/aarch64-simd-builtins.def
4018 (aarch64_sm3ss1qv4si): Ditto.
4019 (aarch64_sm3tt1aq4si): Ditto.
4020 (aarch64_sm3tt1bq4si): Ditto.
4021 (aarch64_sm3tt2aq4si): Ditto.
4022 (aarch64_sm3tt2bq4si): Ditto.
4023 (aarch64_sm3partw1qv4si): Ditto.
4024 (aarch64_sm3partw2qv4si): Ditto.
4025 (aarch64_sm4eqv4si): Ditto.
4026 (aarch64_sm4ekeyqv4si): Ditto.
4027 * config/aarch64/aarch64-simd.md:
4028 (aarch64_sm3ss1qv4si): Ditto.
4029 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4030 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4031 (aarch64_sm4eqv4si): Ditto.
4032 (aarch64_sm4ekeyqv4si): Ditto.
4033 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4034 (sm3part_op): Ditto.
4035 (CRYPTO_SM3TT): Ditto.
4036 (CRYPTO_SM3PART): Ditto.
4037 (UNSPEC_SM3SS1): New unspec.
4038 (UNSPEC_SM3TT1A): Ditto.
4039 (UNSPEC_SM3TT1B): Ditto.
4040 (UNSPEC_SM3TT2A): Ditto.
4041 (UNSPEC_SM3TT2B): Ditto.
4042 (UNSPEC_SM3PARTW1): Ditto.
4043 (UNSPEC_SM3PARTW2): Ditto.
4044 (UNSPEC_SM4E): Ditto.
4045 (UNSPEC_SM4EKEY): Ditto.
4046 * config/aarch64/constraints.md (Ui2): New constraint.
4047 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4048 * config/arm/types.md (crypto_sm3): New type attribute.
4049 (crypto_sm4): Ditto.
4050 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4051 (vsm3tt1aq_u32): Ditto.
4052 (vsm3tt1bq_u32): Ditto.
4053 (vsm3tt2aq_u32): Ditto.
4054 (vsm3tt2bq_u32): Ditto.
4055 (vsm3partw1q_u32): Ditto.
4056 (vsm3partw2q_u32): Ditto.
4057 (vsm4eq_u32): Ditto.
4058 (vsm4ekeyq_u32): Ditto.
4059 (doc/invoke.texi): Document new sm4 option.
4061 2018-01-10 Michael Collison <michael.collison@arm.com>
4063 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4064 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4065 (AARCH64_FL_FOR_ARCH8_4): New.
4066 (AARCH64_FL_V8_4): New flag.
4067 (doc/invoke.texi): Document new armv8.4-a option.
4069 2018-01-10 Michael Collison <michael.collison@arm.com>
4071 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4072 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4073 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4074 * config/aarch64/aarch64-option-extension.def: Add
4075 AARCH64_OPT_EXTENSION of 'sha2'.
4076 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4077 (crypto): Disable sha2 and aes if crypto disabled.
4078 (crypto): Enable aes and sha2 if enabled.
4079 (simd): Disable sha2 and aes if simd disabled.
4080 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4082 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4083 (TARGET_SHA2): New feature flag for sha2.
4084 (TARGET_AES): New feature flag for aes.
4085 * config/aarch64/aarch64-simd.md:
4086 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4087 conditional on TARGET_AES.
4088 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4089 (aarch64_crypto_sha1hsi): Make pattern conditional
4091 (aarch64_crypto_sha1hv4si): Ditto.
4092 (aarch64_be_crypto_sha1hv4si): Ditto.
4093 (aarch64_crypto_sha1su1v4si): Ditto.
4094 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4095 (aarch64_crypto_sha1su0v4si): Ditto.
4096 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4097 (aarch64_crypto_sha256su0v4si): Ditto.
4098 (aarch64_crypto_sha256su1v4si): Ditto.
4099 (doc/invoke.texi): Document new aes and sha2 options.
4101 2018-01-10 Martin Sebor <msebor@redhat.com>
4103 PR tree-optimization/83781
4104 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4107 2018-01-11 Martin Sebor <msebor@gmail.com>
4108 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4110 PR tree-optimization/83501
4111 PR tree-optimization/81703
4113 * tree-ssa-strlen.c (get_string_cst): Rename...
4114 (get_string_len): ...to this. Handle global constants.
4115 (handle_char_store): Adjust.
4117 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4118 Jim Wilson <jimw@sifive.com>
4120 * config/riscv/riscv-protos.h (riscv_output_return): New.
4121 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4122 (riscv_attribute_table, riscv_output_return),
4123 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4124 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4125 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4127 (riscv_expand_prologue): Add early return for naked function.
4128 (riscv_expand_epilogue): Likewise.
4129 (riscv_function_ok_for_sibcall): Return false for naked function.
4130 (riscv_set_current_function): New.
4131 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4132 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4133 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4134 * doc/extend.texi (RISC-V Function Attributes): New.
4136 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4138 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4139 check for 128-bit long double before checking TCmode.
4140 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4141 128-bit long doubles before checking TFmode or TCmode.
4142 (FLOAT128_IBM_P): Likewise.
4144 2018-01-10 Martin Sebor <msebor@redhat.com>
4146 PR tree-optimization/83671
4147 * builtins.c (c_strlen): Unconditionally return zero for the empty
4149 Use -Warray-bounds for warnings.
4150 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4151 for non-constant array indices with COMPONENT_REF, arrays of
4152 arrays, and pointers to arrays.
4153 (gimple_fold_builtin_strlen): Determine and set length range for
4154 non-constant character arrays.
4156 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4159 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4162 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4164 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4166 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4169 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4170 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4171 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4172 indexed_or_indirect_operand predicate.
4173 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4174 (*vsx_le_perm_load_v8hi): Likewise.
4175 (*vsx_le_perm_load_v16qi): Likewise.
4176 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4177 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4178 (*vsx_le_perm_store_v8hi): Likewise.
4179 (*vsx_le_perm_store_v16qi): Likewise.
4180 (eight unnamed splitters): Likewise.
4182 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4184 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4185 * config/rs6000/emmintrin.h: Likewise.
4186 * config/rs6000/mmintrin.h: Likewise.
4187 * config/rs6000/xmmintrin.h: Likewise.
4189 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4192 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4194 * tree.c (tree_nop_conversion): Return true for location wrapper
4196 (maybe_wrap_with_location): New function.
4197 (selftest::check_strip_nops): New function.
4198 (selftest::test_location_wrappers): New function.
4199 (selftest::tree_c_tests): Call it.
4200 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4201 (maybe_wrap_with_location): New decl.
4202 (EXPR_LOCATION_WRAPPER_P): New macro.
4203 (location_wrapper_p): New inline function.
4204 (tree_strip_any_location_wrapper): New inline function.
4206 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4209 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4210 stack_realign_offset for the largest alignment of stack slot
4212 (ix86_find_max_used_stack_alignment): New function.
4213 (ix86_finalize_stack_frame_flags): Use it. Set
4214 max_used_stack_alignment if we don't realign stack.
4215 * config/i386/i386.h (machine_function): Add
4216 max_used_stack_alignment.
4218 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4220 * config/arm/arm.opt (-mbranch-cost): New option.
4221 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4224 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4227 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4228 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4230 2018-01-10 Richard Biener <rguenther@suse.de>
4233 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4234 early out so it also covers the case where we have a non-NULL
4237 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4239 PR tree-optimization/83753
4240 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4241 for non-strided grouped accesses if the number of elements is 1.
4243 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4246 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4247 * i386.h (TARGET_USE_GATHER): Define.
4248 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4250 2018-01-10 Martin Liska <mliska@suse.cz>
4253 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4254 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4256 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4257 CLEANUP_NO_PARTITIONING is not set.
4259 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4261 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4262 for vectors, as a partial revert of r254296.
4263 * rtl.h (const_vec_p): Delete.
4264 (const_vec_duplicate_p): Don't test for vector CONSTs.
4265 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4266 * expmed.c (make_tree): Likewise.
4269 * common.md (E, F): Use CONSTANT_P instead of checking for
4271 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4272 checking for CONST_VECTOR.
4274 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4277 * predict.c (force_edge_cold): Handle in more sane way edges
4280 2018-01-09 Carl Love <cel@us.ibm.com>
4282 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4284 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4285 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4286 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4287 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4288 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4289 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4290 * config/rs6000/rs6000-protos.h: Add extern defition for
4291 rs6000_generate_float2_double_code.
4292 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4294 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4295 (float2_v2df): Add define_expand.
4297 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4300 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4301 op_mode in the force_to_mode call.
4303 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4305 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4306 instead of checking each element individually.
4307 (aarch64_evpc_uzp): Likewise.
4308 (aarch64_evpc_zip): Likewise.
4309 (aarch64_evpc_ext): Likewise.
4310 (aarch64_evpc_rev): Likewise.
4311 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4312 instead of checking each element individually. Return true without
4314 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4315 whether all selected elements come from the same input, instead of
4316 checking each element individually. Remove calls to gen_rtx_REG,
4317 start_sequence and end_sequence and instead assert that no rtl is
4320 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4322 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4323 order of HIGH and CONST checks.
4325 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4327 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4328 if the destination isn't an SSA_NAME.
4330 2018-01-09 Richard Biener <rguenther@suse.de>
4332 PR tree-optimization/83668
4333 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4335 (canonicalize_loop_form): ... here, renamed from ...
4336 (canonicalize_loop_closed_ssa_form): ... this and amended to
4337 swap successor edges for loop exit blocks to make us use
4338 the RPO order we need for initial schedule generation.
4340 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4342 PR tree-optimization/64811
4343 * match.pd: When optimizing comparisons with Inf, avoid
4344 introducing or losing exceptions from comparisons with NaN.
4346 2018-01-09 Martin Liska <mliska@suse.cz>
4349 * asan.c (shadow_mem_size): Add gcc_assert.
4351 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4353 Don't save registers in main().
4356 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4357 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4358 * config/avr/avr.c (avr_set_current_function): Don't error if
4359 naked, OS_task or OS_main are specified at the same time.
4360 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4362 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4364 * common/config/avr/avr-common.c (avr_option_optimization_table):
4365 Switch on -mmain-is-OS_task for optimizing compilations.
4367 2018-01-09 Richard Biener <rguenther@suse.de>
4369 PR tree-optimization/83572
4370 * graphite.c: Include cfganal.h.
4371 (graphite_transform_loops): Connect infinite loops to exit
4372 and remove fake edges at the end.
4374 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4376 * ipa-inline.c (edge_badness): Revert accidental checkin.
4378 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4381 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4382 symbols; not inline clones.
4384 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4387 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4388 hard registers. Formatting fixes.
4390 PR preprocessor/83722
4391 * gcc.c (try_generate_repro): Pass
4392 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4393 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4396 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4397 Kito Cheng <kito.cheng@gmail.com>
4399 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4400 (riscv_leaf_function_p): Delete.
4401 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4403 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4405 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4407 (do_ifelse): New function.
4408 (do_isel): New function.
4409 (do_sub3): New function.
4410 (do_add3): New function.
4411 (do_load_mask_compare): New function.
4412 (do_overlap_load_compare): New function.
4413 (expand_compare_loop): New function.
4414 (expand_block_compare): Call expand_compare_loop() when appropriate.
4415 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4417 (-mblock-compare-inline-loop-limit): New option.
4419 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4422 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4423 Reverse order of second and third operands in first alternative.
4424 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4425 of first and second elements in UNSPEC_VPERMR vector.
4426 (altivec_expand_vec_perm_le): Likewise.
4428 2018-01-08 Jeff Law <law@redhat.com>
4430 PR rtl-optimizatin/81308
4431 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4432 (process_switch): If group_case_labels makes a change, then set
4434 (pass_convert_switch::execute): If a switch is converted, then
4435 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4437 PR rtl-optimization/81308
4438 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4441 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4443 PR target/83663 - Revert r255946
4444 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4445 generation for cases where splatting a value is not useful.
4446 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4447 across a vec_duplicate and a paradoxical subreg forming a vector
4448 mode to a vec_concat.
4450 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4452 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4453 -march=armv8.3-a variants.
4454 * config/arm/t-multilib: Likewise.
4455 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4457 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4459 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4461 (cceq_ior_compare_complement): Give it a name so I can use it, and
4462 change boolean_or_operator predicate to boolean_operator so it can
4463 be used to generate a crand.
4464 (eqne): New code iterator.
4465 (bd/bd_neg): New code_attrs.
4466 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4467 a single define_insn.
4468 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4469 decrement (bdnzt/bdnzf/bdzt/bdzf).
4470 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4471 with the new names of the branch decrement patterns, and added the
4472 names of the branch decrement conditional patterns.
4474 2018-01-08 Richard Biener <rguenther@suse.de>
4476 PR tree-optimization/83563
4477 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4480 2018-01-08 Richard Biener <rguenther@suse.de>
4483 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4485 2018-01-08 Richard Biener <rguenther@suse.de>
4487 PR tree-optimization/83685
4488 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4489 references to abnormals.
4491 2018-01-08 Richard Biener <rguenther@suse.de>
4494 * dwarf2out.c (output_indirect_strings): Handle empty
4495 skeleton_debug_str_hash.
4496 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4498 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4500 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4501 (emit_store_direct): Likewise.
4502 (arc_trampoline_adjust_address): Likewise.
4503 (arc_asm_trampoline_template): New function.
4504 (arc_initialize_trampoline): Use asm_trampoline_template.
4505 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4506 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4507 * config/arc/arc.md (flush_icache): Delete pattern.
4509 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4511 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4512 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4515 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4518 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4519 by not USED_FOR_TARGET.
4520 (make_pass_resolve_sw_modes): Likewise.
4522 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4524 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4527 2018-01-08 Richard Biener <rguenther@suse.de>
4530 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4532 2018-01-08 Richard Biener <rguenther@suse.de>
4535 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4537 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4540 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4541 basic blocks with a small number of successors.
4542 (convert_control_dep_chain_into_preds): Improve handling of
4544 (dump_predicates): Split apart into...
4545 (dump_pred_chain): ...here...
4546 (dump_pred_info): ...and here.
4547 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4548 (can_chain_union_be_invalidated_p): Improve check for invalidation
4550 (uninit_uses_cannot_happen): Avoid unnecessary if
4551 convert_control_dep_chain_into_preds yielded nothing.
4553 2018-01-06 Martin Sebor <msebor@redhat.com>
4555 PR tree-optimization/83640
4556 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4557 subtracting negative offset from size.
4558 (builtin_access::overlap): Adjust offset bounds of the access to fall
4559 within the size of the object if possible.
4561 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4563 PR rtl-optimization/83699
4564 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4565 extract_bit_field_as_subreg to cases in which the extracted
4566 value is also a vector.
4568 * lra-constraints.c (process_alt_operands): Test for the equivalence
4569 substitutions when detecting a possible reload cycle.
4571 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4574 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4575 by default if flag_selective_schedling{,2}. Formatting fixes.
4577 PR rtl-optimization/83682
4578 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4579 if it has non-VECTOR_MODE element mode.
4580 (vec_duplicate_p): Likewise.
4583 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4584 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4586 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4589 * config/i386/i386-builtin.def
4590 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4591 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4592 Require also OPTION_MASK_ISA_AVX512F in addition to
4593 OPTION_MASK_ISA_GFNI.
4594 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4595 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4596 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4597 to OPTION_MASK_ISA_GFNI.
4598 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4599 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4600 OPTION_MASK_ISA_AVX512BW.
4601 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4602 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4603 addition to OPTION_MASK_ISA_GFNI.
4604 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4605 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4606 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4607 to OPTION_MASK_ISA_GFNI.
4608 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4609 a requirement for all ISAs rather than any of them with a few
4611 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4613 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4614 bitmasks to be enabled with 3 exceptions, instead of requiring any
4615 enabled ISA with lots of exceptions.
4616 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4617 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4618 Change avx512bw in isa attribute to avx512f.
4619 * config/i386/sgxintrin.h: Add license boilerplate.
4620 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4621 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4622 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4623 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4625 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4626 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4627 temporarily sse2 rather than sse if not enabled already.
4630 * config/i386/sse.md (VI248_VLBW): Rename to ...
4631 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4632 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4633 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4634 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4635 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4636 mode iterator instead of VI248_VLBW.
4638 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4640 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4641 (record_modified): Skip clobbers; add debug output.
4642 (param_change_prob): Use sreal frequencies.
4644 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4646 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4647 punt for user-aligned variables.
4649 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4651 * tree-chrec.c (chrec_contains_symbols): Return true for
4654 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4657 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4658 of (x|y) == x for BICS pattern.
4660 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4662 PR tree-optimization/83605
4663 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4664 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4667 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4669 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4670 * config/epiphany/rtems.h: New file.
4672 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4673 Uros Bizjak <ubizjak@gmail.com>
4676 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4677 QIreg_operand instead of register_operand predicate.
4678 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4679 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4680 comments instead of -fmitigate[-_]rop.
4682 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4685 * cgraphunit.c (symbol_table::compile): Switch to text_section
4686 before calling assembly_start debug hook.
4687 * run-rtl-passes.c (run_rtl_passes): Likewise.
4690 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4692 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4693 range_int_cst_p rather than !symbolic_range_p before calling
4694 extract_range_from_multiplicative_op_1.
4696 2018-01-04 Jeff Law <law@redhat.com>
4698 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4699 redundant test in assertion.
4701 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4703 * doc/rtl.texi: Document machine_mode wrapper classes.
4705 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4707 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4710 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4712 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4713 the VEC_PERM_EXPR fold to fail.
4715 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4718 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4719 to switched_sections.
4721 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4724 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4727 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4730 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4731 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4733 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4736 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4737 is BLKmode and bitpos not zero or mode change is needed.
4739 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4742 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4745 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4748 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4749 instead of MULT rtx. Update all corresponding splitters.
4751 (*ssub<modesuffix>): Ditto.
4753 (*cmp_sadd_di): Update split patterns.
4754 (*cmp_sadd_si): Ditto.
4755 (*cmp_sadd_sidi): Ditto.
4756 (*cmp_ssub_di): Ditto.
4757 (*cmp_ssub_si): Ditto.
4758 (*cmp_ssub_sidi): Ditto.
4759 * config/alpha/predicates.md (const23_operand): New predicate.
4760 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4761 Look for ASHIFT, not MULT inner operand.
4762 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4764 2018-01-04 Martin Liska <mliska@suse.cz>
4766 PR gcov-profile/83669
4767 * gcov.c (output_intermediate_file): Add version to intermediate
4769 * doc/gcov.texi: Document new field 'version' in intermediate
4770 file format. Fix location of '-k' option of gcov command.
4772 2018-01-04 Martin Liska <mliska@suse.cz>
4775 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4777 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4779 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4781 2018-01-03 Martin Sebor <msebor@redhat.com>
4783 PR tree-optimization/83655
4784 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4785 checking calls with invalid arguments.
4787 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4789 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4790 (vectorizable_mask_load_store): Delete.
4791 (vectorizable_call): Return false for masked loads and stores.
4792 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4793 instead of gimple_assign_rhs1.
4794 (vectorizable_load): Handle IFN_MASK_LOAD.
4795 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4797 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4799 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4801 (vectorizable_mask_load_store): ...here.
4802 (vectorizable_load): ...and here.
4804 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4806 * tree-vect-stmts.c (vect_build_all_ones_mask)
4807 (vect_build_zero_merge_argument): New functions, split out from...
4808 (vectorizable_load): ...here.
4810 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4812 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4814 (vectorizable_mask_load_store): ...here.
4815 (vectorizable_store): ...and here.
4817 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4819 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4821 (vectorizable_mask_load_store): ...here.
4823 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4825 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4826 (vect_model_store_cost): Take a vec_load_store_type instead of a
4828 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4829 (vect_model_store_cost): Take a vec_load_store_type instead of a
4831 (vectorizable_mask_load_store): Update accordingly.
4832 (vectorizable_store): Likewise.
4833 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4835 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4837 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4838 IFN_MASK_LOAD calls here rather than...
4839 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4841 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4842 Alan Hayward <alan.hayward@arm.com>
4843 David Sherwood <david.sherwood@arm.com>
4845 * expmed.c (extract_bit_field_1): For vector extracts,
4846 fall back to extract_bit_field_as_subreg if vec_extract
4849 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4850 Alan Hayward <alan.hayward@arm.com>
4851 David Sherwood <david.sherwood@arm.com>
4853 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4854 they are variable or constant sized.
4855 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4856 slots for constant-sized data.
4858 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4859 Alan Hayward <alan.hayward@arm.com>
4860 David Sherwood <david.sherwood@arm.com>
4862 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4863 handling COND_EXPRs with boolean comparisons, try to find a better
4864 basis for the mask type than the boolean itself.
4866 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4868 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4869 is calculated and how it can be overridden.
4870 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4871 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4873 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4876 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4877 Alan Hayward <alan.hayward@arm.com>
4878 David Sherwood <david.sherwood@arm.com>
4880 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4881 Remove the mode argument.
4882 (aarch64_simd_valid_immediate): Remove the mode and inverse
4884 * config/aarch64/iterators.md (bitsize): New iterator.
4885 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4886 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4887 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4888 aarch64_simd_valid_immediate.
4889 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4890 (aarch64_reg_or_bic_imm): Likewise.
4891 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4892 with an insn_type enum and msl with a modifier_type enum.
4893 Replace element_width with a scalar_mode. Change the shift
4894 to unsigned int. Add constructors for scalar_float_mode and
4895 scalar_int_mode elements.
4896 (aarch64_vect_float_const_representable_p): Delete.
4897 (aarch64_can_const_movi_rtx_p)
4898 (aarch64_simd_scalar_immediate_valid_for_move)
4899 (aarch64_simd_make_constant): Update call to
4900 aarch64_simd_valid_immediate.
4901 (aarch64_advsimd_valid_immediate_hs): New function.
4902 (aarch64_advsimd_valid_immediate): Likewise.
4903 (aarch64_simd_valid_immediate): Remove mode and inverse
4904 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4905 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4906 and aarch64_float_const_representable_p on the result.
4907 (aarch64_output_simd_mov_immediate): Remove mode argument.
4908 Update call to aarch64_simd_valid_immediate and use of
4909 simd_immediate_info.
4910 (aarch64_output_scalar_simd_mov_immediate): Update call
4913 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4914 Alan Hayward <alan.hayward@arm.com>
4915 David Sherwood <david.sherwood@arm.com>
4917 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4918 (mode_nunits): Likewise CONST_MODE_NUNITS.
4919 * machmode.def (ADJUST_NUNITS): Document.
4920 * genmodes.c (mode_data::need_nunits_adj): New field.
4921 (blank_mode): Update accordingly.
4922 (adj_nunits): New variable.
4923 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4925 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4926 listed in adj_nunits.
4927 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4928 listed in adj_nunits. Don't emit case statements for such modes.
4929 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4930 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4931 nothing if adj_nunits is nonnull.
4932 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4933 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4934 (emit_mode_fbit): Update use of print_maybe_const_decl.
4935 (emit_move_size): Likewise. Treat the array as non-const
4937 (emit_mode_adjustments): Handle adj_nunits.
4939 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4941 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4942 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4943 (VECTOR_MODES): Use it.
4944 (make_vector_modes): Take the prefix as an argument.
4946 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4947 Alan Hayward <alan.hayward@arm.com>
4948 David Sherwood <david.sherwood@arm.com>
4950 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4951 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4952 for MODE_VECTOR_BOOL.
4953 * machmode.def (VECTOR_BOOL_MODE): Document.
4954 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4955 (make_vector_bool_mode): New function.
4956 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4958 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4959 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4961 * stor-layout.c (int_mode_for_mode): Likewise.
4962 * tree.c (build_vector_type_for_mode): Likewise.
4963 * varasm.c (output_constant_pool_2): Likewise.
4964 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4965 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4966 for MODE_VECTOR_BOOL.
4967 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4968 of mode class checks.
4969 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4970 instead of a list of mode class checks.
4971 (expand_vector_scalar_condition): Likewise.
4972 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4974 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4975 Alan Hayward <alan.hayward@arm.com>
4976 David Sherwood <david.sherwood@arm.com>
4978 * machmode.h (mode_size): Change from unsigned short to
4980 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4981 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4982 or if measurement_type is not polynomial.
4983 (fixed_size_mode::includes_p): Check for constant-sized modes.
4984 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4985 return a poly_uint16 rather than an unsigned short.
4986 (emit_mode_size): Change the type of mode_size from unsigned short
4987 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4988 (emit_mode_adjustments): Cope with polynomial vector sizes.
4989 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4991 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4993 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4994 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4995 * caller-save.c (setup_save_areas): Likewise.
4996 (replace_reg_with_saved_mem): Likewise.
4997 * calls.c (emit_library_call_value_1): Likewise.
4998 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4999 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5000 (gen_lowpart_for_combine): Likewise.
5001 * convert.c (convert_to_integer_1): Likewise.
5002 * cse.c (equiv_constant, cse_insn): Likewise.
5003 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5004 (cselib_subst_to_values): Likewise.
5005 * dce.c (word_dce_process_block): Likewise.
5006 * df-problems.c (df_word_lr_mark_ref): Likewise.
5007 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5008 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5009 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5010 (rtl_for_decl_location): Likewise.
5011 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5012 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5013 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5014 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5015 (expand_expr_real_1): Likewise.
5016 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5017 (pad_below): Likewise.
5018 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5019 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5020 * ira.c (get_subreg_tracking_sizes): Likewise.
5021 * ira-build.c (ira_create_allocno_objects): Likewise.
5022 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5023 (ira_sort_regnos_for_alter_reg): Likewise.
5024 * ira-costs.c (record_operand_costs): Likewise.
5025 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5026 (resolve_simple_move): Likewise.
5027 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5028 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5029 (lra_constraints): Likewise.
5030 (CONST_POOL_OK_P): Reject variable-sized modes.
5031 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5032 (add_pseudo_to_slot, lra_spill): Likewise.
5033 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5034 * optabs-query.c (get_best_extraction_insn): Likewise.
5035 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5036 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5037 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5038 * recog.c (offsettable_address_addr_space_p): Likewise.
5039 * regcprop.c (maybe_mode_change): Likewise.
5040 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5041 * regrename.c (build_def_use): Likewise.
5042 * regstat.c (dump_reg_info): Likewise.
5043 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5044 (find_reloads, find_reloads_subreg_address): Likewise.
5045 * reload1.c (eliminate_regs_1): Likewise.
5046 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5047 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5048 (simplify_binary_operation_1, simplify_subreg): Likewise.
5049 * targhooks.c (default_function_arg_padding): Likewise.
5050 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5051 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5052 (verify_gimple_assign_ternary): Likewise.
5053 * tree-inline.c (estimate_move_cost): Likewise.
5054 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5055 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5056 (get_address_cost_ainc): Likewise.
5057 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5058 (vect_supportable_dr_alignment): Likewise.
5059 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5060 (vectorizable_reduction): Likewise.
5061 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5062 (vectorizable_operation, vectorizable_load): Likewise.
5063 * tree.c (build_same_sized_truth_vector_type): Likewise.
5064 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5065 * var-tracking.c (emit_note_insn_var_location): Likewise.
5066 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5067 (ADDR_VEC_ALIGN): Likewise.
5069 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5070 Alan Hayward <alan.hayward@arm.com>
5071 David Sherwood <david.sherwood@arm.com>
5073 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5075 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5076 or if measurement_type is polynomial.
5077 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5078 * combine.c (make_extraction): Likewise.
5079 * dse.c (find_shift_sequence): Likewise.
5080 * dwarf2out.c (mem_loc_descriptor): Likewise.
5081 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5082 (extract_bit_field, extract_low_bits): Likewise.
5083 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5084 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5085 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5086 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5087 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5088 * reload.c (find_reloads): Likewise.
5089 * reload1.c (alter_reg): Likewise.
5090 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5091 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5092 * tree-if-conv.c (predicate_mem_writes): Likewise.
5093 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5094 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5095 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5096 * valtrack.c (dead_debug_insert_temp): Likewise.
5097 * varasm.c (mergeable_constant_section): Likewise.
5098 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5100 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5101 Alan Hayward <alan.hayward@arm.com>
5102 David Sherwood <david.sherwood@arm.com>
5104 * expr.c (expand_assignment): Cope with polynomial mode sizes
5105 when assigning to a CONCAT.
5107 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5108 Alan Hayward <alan.hayward@arm.com>
5109 David Sherwood <david.sherwood@arm.com>
5111 * machmode.h (mode_precision): Change from unsigned short to
5113 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5115 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5116 or if measurement_type is not polynomial.
5117 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5118 in which the mode is already known to be a scalar_int_mode.
5119 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5120 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5122 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5123 for GET_MODE_PRECISION.
5124 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5125 for GET_MODE_PRECISION.
5126 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5128 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5129 (expand_field_assignment, make_extraction): Likewise.
5130 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5131 (get_last_value): Likewise.
5132 * convert.c (convert_to_integer_1): Likewise.
5133 * cse.c (cse_insn): Likewise.
5134 * expr.c (expand_expr_real_1): Likewise.
5135 * lra-constraints.c (simplify_operand_subreg): Likewise.
5136 * optabs-query.c (can_atomic_load_p): Likewise.
5137 * optabs.c (expand_atomic_load): Likewise.
5138 (expand_atomic_store): Likewise.
5139 * ree.c (combine_reaching_defs): Likewise.
5140 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5141 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5142 * tree.h (type_has_mode_precision_p): Likewise.
5143 * ubsan.c (instrument_si_overflow): Likewise.
5145 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5146 Alan Hayward <alan.hayward@arm.com>
5147 David Sherwood <david.sherwood@arm.com>
5149 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5150 polynomial numbers of units.
5151 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5152 (valid_vector_subparts_p): New function.
5153 (build_vector_type): Remove temporary shim and take the number
5154 of units as a poly_uint64 rather than an int.
5155 (build_opaque_vector_type): Take the number of units as a
5156 poly_uint64 rather than an int.
5157 * tree.c (build_vector_from_ctor): Handle polynomial
5158 TYPE_VECTOR_SUBPARTS.
5159 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5160 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5161 (build_vector_from_val): If the number of units is variable,
5162 use build_vec_duplicate_cst for constant operands and
5163 VEC_DUPLICATE_EXPR otherwise.
5164 (make_vector_type): Remove temporary is_constant ().
5165 (build_vector_type, build_opaque_vector_type): Take the number of
5166 units as a poly_uint64 rather than an int.
5167 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5169 * cfgexpand.c (expand_debug_expr): Likewise.
5170 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5171 (store_constructor, expand_expr_real_1): Likewise.
5172 (const_scalar_mask_from_tree): Likewise.
5173 * fold-const-call.c (fold_const_reduction): Likewise.
5174 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5175 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5176 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5177 (fold_relational_const): Likewise.
5178 (native_interpret_vector): Likewise. Change the size from an
5179 int to an unsigned int.
5180 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5181 TYPE_VECTOR_SUBPARTS.
5182 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5183 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5184 duplicating a non-constant operand into a variable-length vector.
5185 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5186 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5187 * ipa-icf.c (sem_variable::equals): Likewise.
5188 * match.pd: Likewise.
5189 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5190 * print-tree.c (print_node): Likewise.
5191 * stor-layout.c (layout_type): Likewise.
5192 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5193 * tree-cfg.c (verify_gimple_comparison): Likewise.
5194 (verify_gimple_assign_binary): Likewise.
5195 (verify_gimple_assign_ternary): Likewise.
5196 (verify_gimple_assign_single): Likewise.
5197 * tree-pretty-print.c (dump_generic_node): Likewise.
5198 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5199 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5200 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5201 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5202 (vect_shift_permute_load_chain): Likewise.
5203 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5204 (expand_vector_condition, optimize_vector_constructor): Likewise.
5205 (lower_vec_perm, get_compute_type): Likewise.
5206 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5207 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5208 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5209 (vect_recog_mask_conversion_pattern): Likewise.
5210 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5211 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5212 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5213 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5214 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5215 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5216 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5217 (supportable_widening_operation): Likewise.
5218 (supportable_narrowing_operation): Likewise.
5219 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5221 * varasm.c (output_constant): Likewise.
5223 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5224 Alan Hayward <alan.hayward@arm.com>
5225 David Sherwood <david.sherwood@arm.com>
5227 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5228 so that both the length == 3 and length != 3 cases set up their
5229 own permute vectors. Add comments explaining why we know the
5230 number of elements is constant.
5231 (vect_permute_load_chain): Likewise.
5233 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5234 Alan Hayward <alan.hayward@arm.com>
5235 David Sherwood <david.sherwood@arm.com>
5237 * machmode.h (mode_nunits): Change from unsigned char to
5239 (ONLY_FIXED_SIZE_MODES): New macro.
5240 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5241 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5242 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5244 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5245 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5246 or if measurement_type is not polynomial.
5247 * genmodes.c (ZERO_COEFFS): New macro.
5248 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5250 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5251 Use ZERO_COEFFS when emitting initializers.
5252 * data-streamer.h (bp_pack_poly_value): New function.
5253 (bp_unpack_poly_value): Likewise.
5254 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5255 for GET_MODE_NUNITS.
5256 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5257 for GET_MODE_NUNITS.
5258 * tree.c (make_vector_type): Remove temporary shim and make
5259 the real function take the number of units as a poly_uint64
5261 (build_vector_type_for_mode): Handle polynomial nunits.
5262 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5263 * emit-rtl.c (const_vec_series_p_1): Likewise.
5264 (gen_rtx_CONST_VECTOR): Likewise.
5265 * fold-const.c (test_vec_duplicate_folding): Likewise.
5266 * genrecog.c (validate_pattern): Likewise.
5267 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5268 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5269 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5270 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5271 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5272 * rtlanal.c (subreg_get_info): Likewise.
5273 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5274 (vect_grouped_load_supported): Likewise.
5275 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5276 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5277 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5278 (simplify_const_unary_operation, simplify_binary_operation_1)
5279 (simplify_const_binary_operation, simplify_ternary_operation)
5280 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5281 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5282 instead of CONST_VECTOR_NUNITS.
5283 * varasm.c (output_constant_pool_2): Likewise.
5284 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5285 explicit-encoded elements in the XVEC for variable-length vectors.
5287 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5289 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5291 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5292 Alan Hayward <alan.hayward@arm.com>
5293 David Sherwood <david.sherwood@arm.com>
5295 * coretypes.h (fixed_size_mode): Declare.
5296 (fixed_size_mode_pod): New typedef.
5297 * builtins.h (target_builtins::x_apply_args_mode)
5298 (target_builtins::x_apply_result_mode): Change type to
5299 fixed_size_mode_pod.
5300 * builtins.c (apply_args_size, apply_result_size, result_vector)
5301 (expand_builtin_apply_args_1, expand_builtin_apply)
5302 (expand_builtin_return): Update accordingly.
5304 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5306 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5307 * cselib.c (cselib_hash_rtx): Likewise.
5308 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5309 CONST_VECTOR encoding.
5311 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5312 Jeff Law <law@redhat.com>
5315 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5316 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5317 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5318 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5321 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5322 explicitly probe *sp in a noreturn function if there were any callee
5323 register saves or frame pointer is needed.
5325 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5328 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5329 BLKmode for ternary, binary or unary expressions.
5332 * var-tracking.c (delete_vta_debug_insn): New inline function.
5333 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5334 insns from get_insns () to NULL instead of each bb separately.
5335 Use delete_vta_debug_insn. No longer static.
5336 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5337 delete_vta_debug_insns callers.
5338 * rtl.h (delete_vta_debug_insns): Declare.
5339 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5340 instead of variable_tracking_main.
5342 2018-01-03 Martin Sebor <msebor@redhat.com>
5344 PR tree-optimization/83603
5345 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5346 arguments past the endof the argument list in functions declared
5347 without a prototype.
5348 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5349 Avoid checking when arguments are null.
5351 2018-01-03 Martin Sebor <msebor@redhat.com>
5354 * doc/extend.texi (attribute const): Fix a typo.
5355 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5356 issuing -Wsuggest-attribute for void functions.
5358 2018-01-03 Martin Sebor <msebor@redhat.com>
5360 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5361 offset_int::from instead of wide_int::to_shwi.
5362 (maybe_diag_overlap): Remove assertion.
5363 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5364 * gimple-ssa-sprintf.c (format_directive): Same.
5365 (parse_directive): Same.
5366 (sprintf_dom_walker::compute_format_length): Same.
5367 (try_substitute_return_value): Same.
5369 2018-01-03 Jeff Law <law@redhat.com>
5372 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5373 non-constant residual for zero at runtime and avoid probing in
5374 that case. Reorganize code for trailing problem to mirror handling
5377 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5379 PR tree-optimization/83501
5380 * tree-ssa-strlen.c (get_string_cst): New.
5381 (handle_char_store): Call get_string_cst.
5383 2018-01-03 Martin Liska <mliska@suse.cz>
5385 PR tree-optimization/83593
5386 * tree-ssa-strlen.c: Include tree-cfg.h.
5387 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5388 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5389 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5391 (strlen_dom_walker::before_dom_children): Call
5392 gimple_purge_dead_eh_edges. Dump tranformation with details
5394 (strlen_dom_walker::before_dom_children): Update call by adding
5395 new argument cleanup_eh.
5396 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5398 2018-01-03 Martin Liska <mliska@suse.cz>
5401 * cif-code.def (VARIADIC_THUNK): New enum value.
5402 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5405 2018-01-03 Jan Beulich <jbeulich@suse.com>
5407 * sse.md (mov<mode>_internal): Tighten condition for when to use
5408 vmovdqu<ssescalarsize> for TI and OI modes.
5410 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5412 Update copyright years.
5414 2018-01-03 Martin Liska <mliska@suse.cz>
5417 * ipa-visibility.c (function_and_variable_visibility): Skip
5418 functions with noipa attribure.
5420 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5422 * gcc.c (process_command): Update copyright notice dates.
5423 * gcov-dump.c (print_version): Ditto.
5424 * gcov.c (print_version): Ditto.
5425 * gcov-tool.c (print_version): Ditto.
5426 * gengtype.c (create_file): Ditto.
5427 * doc/cpp.texi: Bump @copying's copyright year.
5428 * doc/cppinternals.texi: Ditto.
5429 * doc/gcc.texi: Ditto.
5430 * doc/gccint.texi: Ditto.
5431 * doc/gcov.texi: Ditto.
5432 * doc/install.texi: Ditto.
5433 * doc/invoke.texi: Ditto.
5435 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5437 * vector-builder.h (vector_builder::m_full_nelts): Change from
5438 unsigned int to poly_uint64.
5439 (vector_builder::full_nelts): Update prototype accordingly.
5440 (vector_builder::new_vector): Likewise.
5441 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5442 (vector_builder::operator ==): Likewise.
5443 (vector_builder::finalize): Likewise.
5444 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5445 Take the number of elements as a poly_uint64 rather than an
5447 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5448 from unsigned int to poly_uint64.
5449 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5450 (vec_perm_indices::new_vector): Likewise.
5451 (vec_perm_indices::length): Likewise.
5452 (vec_perm_indices::nelts_per_input): Likewise.
5453 (vec_perm_indices::input_nelts): Likewise.
5454 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5455 number of elements per input as a poly_uint64 rather than an
5456 unsigned int. Use the original encoding for variable-length
5457 vectors, rather than clamping each individual element.
5458 For the second and subsequent elements in each pattern,
5459 clamp the step and base before clamping their sum.
5460 (vec_perm_indices::series_p): Handle polynomial element counts.
5461 (vec_perm_indices::all_in_range_p): Likewise.
5462 (vec_perm_indices_to_tree): Likewise.
5463 (vec_perm_indices_to_rtx): Likewise.
5464 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5465 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5466 (tree_vector_builder::new_binary_operation): Handle polynomial
5467 element counts. Return false if we need to know the number
5468 of elements at compile time.
5469 * fold-const.c (fold_vec_perm): Punt if the number of elements
5470 isn't known at compile time.
5472 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5474 * vec-perm-indices.h (vec_perm_builder): Change element type
5475 from HOST_WIDE_INT to poly_int64.
5476 (vec_perm_indices::element_type): Update accordingly.
5477 (vec_perm_indices::clamp): Handle polynomial element_types.
5478 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5479 (vec_perm_indices::all_in_range_p): Likewise.
5480 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5482 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5483 polynomial vec_perm_indices element types.
5484 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5485 * fold-const.c (fold_vec_perm): Likewise.
5486 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5487 * tree-vect-generic.c (lower_vec_perm): Likewise.
5488 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5489 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5490 element type to HOST_WIDE_INT.
5492 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5493 Alan Hayward <alan.hayward@arm.com>
5494 David Sherwood <david.sherwood@arm.com>
5496 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5497 rather than an int. Use plus_constant.
5498 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5499 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5501 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5502 Alan Hayward <alan.hayward@arm.com>
5503 David Sherwood <david.sherwood@arm.com>
5505 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5506 a HOST_WIDE_INT to a poly_int64.
5508 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5509 Alan Hayward <alan.hayward@arm.com>
5510 David Sherwood <david.sherwood@arm.com>
5512 * calls.c (load_register_parameters): Cope with polynomial
5513 mode sizes. Require a constant size for BLKmode parameters
5514 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5515 forces a parameter to be padded at the lsb end in order to
5516 fill a complete number of words, require the parameter size
5517 to be ordered wrt UNITS_PER_WORD.
5519 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5520 Alan Hayward <alan.hayward@arm.com>
5521 David Sherwood <david.sherwood@arm.com>
5523 * reload1.c (spill_stack_slot_width): Change element type
5524 from unsigned int to poly_uint64_pod.
5525 (alter_reg): Treat mode sizes as polynomial.
5527 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5528 Alan Hayward <alan.hayward@arm.com>
5529 David Sherwood <david.sherwood@arm.com>
5531 * reload.c (complex_word_subreg_p): New function.
5532 (reload_inner_reg_of_subreg, push_reload): Use it.
5534 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5535 Alan Hayward <alan.hayward@arm.com>
5536 David Sherwood <david.sherwood@arm.com>
5538 * lra-constraints.c (process_alt_operands): Reject matched
5539 operands whose sizes aren't ordered.
5540 (match_reload): Refer to this check here.
5542 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5543 Alan Hayward <alan.hayward@arm.com>
5544 David Sherwood <david.sherwood@arm.com>
5546 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5547 that the mode size is in the set {1, 2, 4, 8, 16}.
5549 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5550 Alan Hayward <alan.hayward@arm.com>
5551 David Sherwood <david.sherwood@arm.com>
5553 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5554 Use plus_constant instead of gen_rtx_PLUS.
5556 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5557 Alan Hayward <alan.hayward@arm.com>
5558 David Sherwood <david.sherwood@arm.com>
5560 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5561 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5562 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5563 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5564 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5565 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5566 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5567 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5568 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5569 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5571 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5572 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5573 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5574 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5575 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5576 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5577 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5578 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5579 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5580 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5582 * expr.c (emit_move_resolve_push): Treat the input and result
5583 of PUSH_ROUNDING as a poly_int64.
5584 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5585 (emit_push_insn): Likewise.
5586 * lra-eliminations.c (mark_not_eliminable): Likewise.
5587 * recog.c (push_operand): Likewise.
5588 * reload1.c (elimination_effects): Likewise.
5589 * rtlanal.c (nonzero_bits1): Likewise.
5590 * calls.c (store_one_arg): Likewise. Require the padding to be
5591 known at compile time.
5593 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5594 Alan Hayward <alan.hayward@arm.com>
5595 David Sherwood <david.sherwood@arm.com>
5597 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5598 Use plus_constant instead of gen_rtx_PLUS.
5600 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5601 Alan Hayward <alan.hayward@arm.com>
5602 David Sherwood <david.sherwood@arm.com>
5604 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5607 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5608 Alan Hayward <alan.hayward@arm.com>
5609 David Sherwood <david.sherwood@arm.com>
5611 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5612 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5613 via stack temporaries. Treat the mode size as polynomial too.
5615 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5616 Alan Hayward <alan.hayward@arm.com>
5617 David Sherwood <david.sherwood@arm.com>
5619 * expr.c (expand_expr_real_2): When handling conversions involving
5620 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5621 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5622 as a poly_uint64 too.
5624 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5625 Alan Hayward <alan.hayward@arm.com>
5626 David Sherwood <david.sherwood@arm.com>
5628 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5630 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5631 Alan Hayward <alan.hayward@arm.com>
5632 David Sherwood <david.sherwood@arm.com>
5634 * combine.c (can_change_dest_mode): Handle polynomial
5635 REGMODE_NATURAL_SIZE.
5636 * expmed.c (store_bit_field_1): Likewise.
5637 * expr.c (store_constructor): Likewise.
5638 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5639 and polynomial REGMODE_NATURAL_SIZE.
5640 (gen_lowpart_common): Likewise.
5641 * reginfo.c (record_subregs_of_mode): Likewise.
5642 * rtlanal.c (read_modify_subreg_p): Likewise.
5644 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5645 Alan Hayward <alan.hayward@arm.com>
5646 David Sherwood <david.sherwood@arm.com>
5648 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5649 numbers of elements.
5651 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5652 Alan Hayward <alan.hayward@arm.com>
5653 David Sherwood <david.sherwood@arm.com>
5655 * match.pd: Cope with polynomial numbers of vector elements.
5657 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5658 Alan Hayward <alan.hayward@arm.com>
5659 David Sherwood <david.sherwood@arm.com>
5661 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5662 in a POINTER_PLUS_EXPR.
5664 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5665 Alan Hayward <alan.hayward@arm.com>
5666 David Sherwood <david.sherwood@arm.com>
5668 * omp-simd-clone.c (simd_clone_subparts): New function.
5669 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5670 (ipa_simd_modify_function_body): Likewise.
5672 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5673 Alan Hayward <alan.hayward@arm.com>
5674 David Sherwood <david.sherwood@arm.com>
5676 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5677 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5678 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5679 (expand_vector_condition, vector_element): Likewise.
5680 (subparts_gt): New function.
5681 (get_compute_type): Use subparts_gt.
5682 (count_type_subparts): Delete.
5683 (expand_vector_operations_1): Use subparts_gt instead of
5684 count_type_subparts.
5686 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5687 Alan Hayward <alan.hayward@arm.com>
5688 David Sherwood <david.sherwood@arm.com>
5690 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5691 (vect_compile_time_alias): ...this new function. Do the calculation
5692 on poly_ints rather than trees.
5693 (vect_prune_runtime_alias_test_list): Update call accordingly.
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-slp.c (vect_build_slp_tree_1): Handle polynomial
5701 (vect_schedule_slp_instance): Likewise.
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-slp.c (vect_get_and_check_slp_defs): Reject
5708 constant and extern definitions for variable-length vectors.
5709 (vect_get_constant_vectors): Note that the number of units
5710 is known to be constant.
5712 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5713 Alan Hayward <alan.hayward@arm.com>
5714 David Sherwood <david.sherwood@arm.com>
5716 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5717 of units as polynomial. Choose between WIDE and NARROW based
5720 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5721 Alan Hayward <alan.hayward@arm.com>
5722 David Sherwood <david.sherwood@arm.com>
5724 * tree-vect-stmts.c (simd_clone_subparts): New function.
5725 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5727 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5728 Alan Hayward <alan.hayward@arm.com>
5729 David Sherwood <david.sherwood@arm.com>
5731 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5732 vectors as polynomial. Use build_index_vector for
5735 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5736 Alan Hayward <alan.hayward@arm.com>
5737 David Sherwood <david.sherwood@arm.com>
5739 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5740 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5741 for variable-length vectors.
5742 (vectorizable_mask_load_store): Treat the number of units as
5743 polynomial, asserting that it is constant if the condition has
5744 already been enforced.
5745 (vectorizable_store, vectorizable_load): Likewise.
5747 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5748 Alan Hayward <alan.hayward@arm.com>
5749 David Sherwood <david.sherwood@arm.com>
5751 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5752 of units as polynomial. Punt if we can't tell at compile time
5753 which vector contains the final result.
5755 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5756 Alan Hayward <alan.hayward@arm.com>
5757 David Sherwood <david.sherwood@arm.com>
5759 * tree-vect-loop.c (vectorizable_induction): Treat the number
5760 of units as polynomial. Punt on SLP inductions. Use an integer
5761 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5762 cast of such a series for variable-length floating-point
5765 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5766 Alan Hayward <alan.hayward@arm.com>
5767 David Sherwood <david.sherwood@arm.com>
5769 * tree.h (build_index_vector): Declare.
5770 * tree.c (build_index_vector): New function.
5771 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5772 of units as polynomial, forcibly converting it to a constant if
5773 vectorizable_reduction has already enforced the condition.
5774 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5775 to create a {1,2,3,...} vector.
5776 (vectorizable_reduction): Treat the number of units as polynomial.
5777 Choose vectype_in based on the largest scalar element size rather
5778 than the smallest number of units. Enforce the restrictions
5781 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5782 Alan Hayward <alan.hayward@arm.com>
5783 David Sherwood <david.sherwood@arm.com>
5785 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5786 number of units as polynomial.
5788 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5789 Alan Hayward <alan.hayward@arm.com>
5790 David Sherwood <david.sherwood@arm.com>
5792 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5793 * target.def (autovectorize_vector_sizes): Return the vector sizes
5794 by pointer, using vector_sizes rather than a bitmask.
5795 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5796 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5797 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5799 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5800 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5801 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5802 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5803 * omp-general.c (omp_max_vf): Likewise.
5804 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5805 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5806 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5807 * tree-vect-slp.c (vect_slp_bb): Likewise.
5808 * doc/tm.texi: Regenerate.
5809 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5811 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5812 the vector size as a poly_uint64 rather than an unsigned int.
5813 (current_vector_size): Change from an unsigned int to a poly_uint64.
5814 (get_vectype_for_scalar_type): Update accordingly.
5815 * tree.h (build_truth_vector_type): Take the size and number of
5816 units as a poly_uint64 rather than an unsigned int.
5817 (build_vector_type): Add a temporary overload that takes
5818 the number of units as a poly_uint64 rather than an unsigned int.
5819 * tree.c (make_vector_type): Likewise.
5820 (build_truth_vector_type): Take the number of units as a poly_uint64
5821 rather than an unsigned int.
5823 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5824 Alan Hayward <alan.hayward@arm.com>
5825 David Sherwood <david.sherwood@arm.com>
5827 * target.def (get_mask_mode): Take the number of units and length
5828 as poly_uint64s rather than unsigned ints.
5829 * targhooks.h (default_get_mask_mode): Update accordingly.
5830 * targhooks.c (default_get_mask_mode): Likewise.
5831 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5832 * doc/tm.texi: Regenerate.
5834 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5835 Alan Hayward <alan.hayward@arm.com>
5836 David Sherwood <david.sherwood@arm.com>
5838 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5839 * omp-general.c (omp_max_vf): Likewise.
5840 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5841 (expand_omp_simd): Handle polynomial safelen.
5842 * omp-low.c (omplow_simd_context): Add a default constructor.
5843 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5844 (lower_rec_simd_input_clauses): Update accordingly.
5845 (lower_rec_input_clauses): Likewise.
5847 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5848 Alan Hayward <alan.hayward@arm.com>
5849 David Sherwood <david.sherwood@arm.com>
5851 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5852 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5853 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5854 (vect_analyze_slp_cost): Likewise.
5855 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5856 (vect_model_load_cost): Likewise.
5858 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5859 Alan Hayward <alan.hayward@arm.com>
5860 David Sherwood <david.sherwood@arm.com>
5862 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5863 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5864 from an unsigned int * to a poly_uint64_pod *.
5865 (calculate_unrolling_factor): New function.
5866 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5868 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5869 Alan Hayward <alan.hayward@arm.com>
5870 David Sherwood <david.sherwood@arm.com>
5872 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5873 from an unsigned int to a poly_uint64.
5874 (_loop_vec_info::slp_unrolling_factor): Likewise.
5875 (_loop_vec_info::vectorization_factor): Change from an int
5877 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5878 (vect_get_num_vectors): New function.
5879 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5880 (vect_get_num_copies): Use vect_get_num_vectors.
5881 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5882 to an unsigned int *.
5883 (vect_analyze_data_refs): Change min_vf from an int * to a
5885 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5886 than an unsigned HOST_WIDE_INT.
5887 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5888 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5889 to an unsigned int *.
5890 (vect_analyze_data_ref_dependences): Likewise.
5891 (vect_compute_data_ref_alignment): Handle polynomial vf.
5892 (vect_enhance_data_refs_alignment): Likewise.
5893 (vect_prune_runtime_alias_test_list): Likewise.
5894 (vect_shift_permute_load_chain): Likewise.
5895 (vect_supportable_dr_alignment): Likewise.
5896 (dependence_distance_ge_vf): Take the vectorization factor as a
5897 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5898 (vect_analyze_data_refs): Change min_vf from an int * to a
5900 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5901 vfm1 as a poly_uint64 rather than an int. Make the same change
5902 for the returned bound_scalar.
5903 (vect_gen_vector_loop_niters): Handle polynomial vf.
5904 (vect_do_peeling): Likewise. Update call to
5905 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5906 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5908 * tree-vect-loop.c (vect_determine_vectorization_factor)
5909 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5910 (vect_get_known_peeling_cost): Likewise.
5911 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5912 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5913 (vect_transform_loop): Likewise. Use the lowest possible VF when
5914 updating the upper bounds of the loop.
5915 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5917 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5918 polynomial unroll factors.
5919 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5920 (vect_make_slp_decision): Likewise.
5921 (vect_supported_load_permutation_p): Likewise, and polynomial
5923 (vect_analyze_slp_cost): Handle polynomial vf.
5924 (vect_slp_analyze_node_operations): Likewise.
5925 (vect_slp_analyze_bb_1): Likewise.
5926 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5927 than an unsigned HOST_WIDE_INT.
5928 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5929 (vectorizable_load): Handle polynomial vf.
5930 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5932 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5934 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5935 Alan Hayward <alan.hayward@arm.com>
5936 David Sherwood <david.sherwood@arm.com>
5938 * match.pd: Handle bit operations involving three constants
5939 and try to fold one pair.
5941 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5943 * tree-vect-loop-manip.c: Include gimple-fold.h.
5944 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5945 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5946 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5947 Add a path that uses a step of VF instead of 1, but disable it
5949 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5950 and niters_no_overflow parameters. Update calls to
5951 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5952 Create a new SSA name if the latter choses to use a ste other
5953 than zero, and return it via niters_vector_mult_vf_var.
5954 * tree-vect-loop.c (vect_transform_loop): Update calls to
5955 vect_do_peeling, vect_gen_vector_loop_niters and
5956 slpeel_make_loop_iterate_ntimes.
5957 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5958 (vect_gen_vector_loop_niters): Update declarations after above changes.
5960 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5962 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5963 128-bit round to integer instructions.
5964 (ceil<mode>2): Likewise.
5965 (btrunc<mode>2): Likewise.
5966 (round<mode>2): Likewise.
5968 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5970 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5971 unaligned VSX load/store on P8/P9.
5972 (expand_block_clear): Allow the use of unaligned VSX
5973 load/store on P8/P9.
5975 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5977 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5979 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5980 swap associated with both a load and a store.
5982 2018-01-02 Andrew Waterman <andrew@sifive.com>
5984 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5985 * config/riscv/riscv.md (clear_cache): Use it.
5987 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5989 * web.c: Remove out-of-date comment.
5991 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5993 * expr.c (fixup_args_size_notes): Check that any existing
5994 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5995 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5996 (emit_single_push_insn): ...here.
5998 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6000 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6001 (const_vector_encoded_nelts): New function.
6002 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6003 (const_vector_int_elt, const_vector_elt): Declare.
6004 * emit-rtl.c (const_vector_int_elt_1): New function.
6005 (const_vector_elt): Likewise.
6006 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6007 of CONST_VECTOR_ELT.
6009 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6011 * expr.c: Include rtx-vector-builder.h.
6012 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6013 directly on the tree encoding.
6014 (const_vector_from_tree): Likewise.
6015 * optabs.c: Include rtx-vector-builder.h.
6016 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6017 sequence of "u" values.
6018 * vec-perm-indices.c: Include rtx-vector-builder.h.
6019 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6020 directly on the vec_perm_indices encoding.
6022 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6024 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6025 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6026 * rtx-vector-builder.h: New file.
6027 * rtx-vector-builder.c: Likewise.
6028 * rtl.h (rtx_def::u2): Add a const_vector field.
6029 (CONST_VECTOR_NPATTERNS): New macro.
6030 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6031 (CONST_VECTOR_DUPLICATE_P): Likewise.
6032 (CONST_VECTOR_STEPPED_P): Likewise.
6033 (CONST_VECTOR_ENCODED_ELT): Likewise.
6034 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6035 (unwrap_const_vec_duplicate): Likewise.
6036 (const_vec_series_p): Check for a non-duplicated vector encoding.
6037 Say that the function only returns true for integer vectors.
6038 * emit-rtl.c: Include rtx-vector-builder.h.
6039 (gen_const_vec_duplicate_1): Delete.
6040 (gen_const_vector): Call gen_const_vec_duplicate instead of
6041 gen_const_vec_duplicate_1.
6042 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6043 (gen_const_vec_duplicate): Use rtx_vector_builder.
6044 (gen_const_vec_series): Likewise.
6045 (gen_rtx_CONST_VECTOR): Likewise.
6046 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6047 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6048 Build a new vector rather than modifying a CONST_VECTOR in-place.
6049 (handle_special_swappables): Update call accordingly.
6050 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6051 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6052 Build a new vector rather than modifying a CONST_VECTOR in-place.
6053 (handle_special_swappables): Update call accordingly.
6055 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6057 * simplify-rtx.c (simplify_const_binary_operation): Use
6058 CONST_VECTOR_ELT instead of XVECEXP.
6060 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6062 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6063 the selector elements to be different from the data elements
6064 if the selector is a VECTOR_CST.
6065 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6066 ssizetype for the selector.
6068 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6070 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6071 before testing each element individually.
6072 * tree-vect-generic.c (lower_vec_perm): Likewise.
6074 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6076 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6077 * selftest-run-tests.c (selftest::run_tests): Call it.
6078 * vector-builder.h (vector_builder::operator ==): New function.
6079 (vector_builder::operator !=): Likewise.
6080 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6081 (vec_perm_indices::all_from_input_p): New function.
6082 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6083 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6084 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6085 instead of reading the VECTOR_CST directly. Detect whether both
6086 vector inputs are the same before constructing the vec_perm_indices,
6087 and update the number of inputs argument accordingly. Use the
6088 utility functions added above. Only construct sel2 if we need to.
6090 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6092 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6093 the broadcast of the low byte.
6094 (expand_mult_highpart): Use an explicit encoding for the permutes.
6095 * optabs-query.c (can_mult_highpart_p): Likewise.
6096 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6097 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6098 (vectorizable_bswap): Likewise.
6099 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6100 explicit encoding for the power-of-2 permutes.
6101 (vect_permute_store_chain): Likewise.
6102 (vect_grouped_load_supported): Likewise.
6103 (vect_permute_load_chain): Likewise.
6105 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6107 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6108 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6109 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6110 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6111 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6112 (vect_gen_perm_mask_any): Likewise.
6114 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6116 * int-vector-builder.h: New file.
6117 * vec-perm-indices.h: Include int-vector-builder.h.
6118 (vec_perm_indices): Redefine as an int_vector_builder.
6119 (auto_vec_perm_indices): Delete.
6120 (vec_perm_builder): Redefine as a stand-alone class.
6121 (vec_perm_indices::vec_perm_indices): New function.
6122 (vec_perm_indices::clamp): Likewise.
6123 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6124 (vec_perm_indices::new_vector): New function.
6125 (vec_perm_indices::new_expanded_vector): Update for new
6126 vec_perm_indices class.
6127 (vec_perm_indices::rotate_inputs): New function.
6128 (vec_perm_indices::all_in_range_p): Operate directly on the
6129 encoded form, without computing elided elements.
6130 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6131 encoding. Update for new vec_perm_indices class.
6132 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6133 the given vec_perm_builder.
6134 (expand_vec_perm_var): Update vec_perm_builder constructor.
6135 (expand_mult_highpart): Use vec_perm_builder instead of
6136 auto_vec_perm_indices.
6137 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6138 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6139 or double series encoding as appropriate.
6140 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6141 vec_perm_indices instead of auto_vec_perm_indices.
6142 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6143 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6144 (vect_permute_store_chain): Likewise.
6145 (vect_grouped_load_supported): Likewise.
6146 (vect_permute_load_chain): Likewise.
6147 (vect_shift_permute_load_chain): Likewise.
6148 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6149 (vect_transform_slp_perm_load): Likewise.
6150 (vect_schedule_slp_instance): Likewise.
6151 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6152 (vectorizable_mask_load_store): Likewise.
6153 (vectorizable_bswap): Likewise.
6154 (vectorizable_store): Likewise.
6155 (vectorizable_load): Likewise.
6156 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6157 vec_perm_indices instead of auto_vec_perm_indices. Use
6158 tree_to_vec_perm_builder to read the vector from a tree.
6159 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6160 vec_perm_builder instead of a vec_perm_indices.
6161 (have_whole_vector_shift): Use vec_perm_builder and
6162 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6163 truncation to calc_vec_perm_mask_for_shift.
6164 (vect_create_epilog_for_reduction): Likewise.
6165 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6166 from auto_vec_perm_indices to vec_perm_indices.
6167 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6168 instead of changing individual elements.
6169 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6170 the vector in d.perm.
6171 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6172 from auto_vec_perm_indices to vec_perm_indices.
6173 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6174 instead of changing individual elements.
6175 (arm_vectorize_vec_perm_const): Use new_vector to install
6176 the vector in d.perm.
6177 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6178 Update vec_perm_builder constructor.
6179 (rs6000_expand_interleave): Likewise.
6180 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6181 (rs6000_expand_interleave): Likewise.
6183 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6185 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6186 to qimode could truncate the indices.
6187 * optabs.c (expand_vec_perm_var): Likewise.
6189 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6191 * Makefile.in (OBJS): Add vec-perm-indices.o.
6192 * vec-perm-indices.h: New file.
6193 * vec-perm-indices.c: Likewise.
6194 * target.h (vec_perm_indices): Replace with a forward class
6196 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6197 * optabs.h: Include vec-perm-indices.h.
6198 (expand_vec_perm): Delete.
6199 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6200 (expand_vec_perm_const): Declare.
6201 * target.def (vec_perm_const_ok): Replace with...
6202 (vec_perm_const): ...this new hook.
6203 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6204 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6205 * doc/tm.texi: Regenerate.
6206 * optabs.def (vec_perm_const): Delete.
6207 * doc/md.texi (vec_perm_const): Likewise.
6208 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6209 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6210 expand_vec_perm for constant permutation vectors. Assert that
6211 the mode of variable permutation vectors is the integer equivalent
6212 of the mode that is being permuted.
6213 * optabs-query.h (selector_fits_mode_p): Declare.
6214 * optabs-query.c: Include vec-perm-indices.h.
6215 (selector_fits_mode_p): New function.
6216 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6217 is defined, instead of checking whether the vec_perm_const_optab
6218 exists. Use targetm.vectorize.vec_perm_const instead of
6219 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6220 fit in the vector mode before using a variable permute.
6221 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6222 vec_perm_indices instead of an rtx.
6223 (expand_vec_perm): Replace with...
6224 (expand_vec_perm_const): ...this new function. Take the selector
6225 as a vec_perm_indices rather than an rtx. Also take the mode of
6226 the selector. Update call to shift_amt_for_vec_perm_mask.
6227 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6228 Use vec_perm_indices::new_expanded_vector to expand the original
6229 selector into bytes. Check whether the indices fit in the vector
6230 mode before using a variable permute.
6231 (expand_vec_perm_var): Make global.
6232 (expand_mult_highpart): Use expand_vec_perm_const.
6233 * fold-const.c: Includes vec-perm-indices.h.
6234 * tree-ssa-forwprop.c: Likewise.
6235 * tree-vect-data-refs.c: Likewise.
6236 * tree-vect-generic.c: Likewise.
6237 * tree-vect-loop.c: Likewise.
6238 * tree-vect-slp.c: Likewise.
6239 * tree-vect-stmts.c: Likewise.
6240 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6242 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6243 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6244 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6245 (aarch64_vectorize_vec_perm_const): ...this new function.
6246 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6247 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6248 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6249 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6250 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6251 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6252 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6254 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6255 check for NEON modes.
6256 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6257 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6258 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6259 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6261 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6262 the old VEC_PERM_CONST conditions.
6263 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6264 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6265 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6266 (ia64_vectorize_vec_perm_const_ok): Merge into...
6267 (ia64_vectorize_vec_perm_const): ...this new function.
6268 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6269 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6270 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6271 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6272 * config/mips/mips.c (mips_expand_vec_perm_const)
6273 (mips_vectorize_vec_perm_const_ok): Merge into...
6274 (mips_vectorize_vec_perm_const): ...this new function.
6275 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6276 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6277 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6278 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6279 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6280 (rs6000_expand_vec_perm_const): Delete.
6281 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6283 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6284 (altivec_expand_vec_perm_const_le): Take each operand individually.
6285 Operate on constant selectors rather than rtxes.
6286 (altivec_expand_vec_perm_const): Likewise. Update call to
6287 altivec_expand_vec_perm_const_le.
6288 (rs6000_expand_vec_perm_const): Delete.
6289 (rs6000_vectorize_vec_perm_const_ok): Delete.
6290 (rs6000_vectorize_vec_perm_const): New function.
6291 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6292 an element count and rtx array.
6293 (rs6000_expand_extract_even): Update call accordingly.
6294 (rs6000_expand_interleave): Likewise.
6295 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6296 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6297 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6298 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6299 (rs6000_expand_vec_perm_const): Delete.
6300 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6301 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6302 (altivec_expand_vec_perm_const_le): Take each operand individually.
6303 Operate on constant selectors rather than rtxes.
6304 (altivec_expand_vec_perm_const): Likewise. Update call to
6305 altivec_expand_vec_perm_const_le.
6306 (rs6000_expand_vec_perm_const): Delete.
6307 (rs6000_vectorize_vec_perm_const_ok): Delete.
6308 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6309 reference to the SPE evmerge intructions.
6310 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6311 an element count and rtx array.
6312 (rs6000_expand_extract_even): Update call accordingly.
6313 (rs6000_expand_interleave): Likewise.
6314 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6315 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6317 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6319 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6321 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6322 vector mode and that that mode matches the mode of the data
6324 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6325 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6326 directly using expand_vec_perm_1 when forcing selectors into
6328 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6330 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6332 * optabs-query.h (can_vec_perm_p): Delete.
6333 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6334 * optabs-query.c (can_vec_perm_p): Split into...
6335 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6336 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6337 particular selector is valid.
6338 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6339 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6340 (vect_grouped_load_supported): Likewise.
6341 (vect_shift_permute_load_chain): Likewise.
6342 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6343 (vect_transform_slp_perm_load): Likewise.
6344 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6345 (vectorizable_bswap): Likewise.
6346 (vect_gen_perm_mask_checked): Likewise.
6347 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6348 implementations of variable permutation vectors into account
6349 when deciding which selector to use.
6350 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6351 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6352 with a false third argument.
6353 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6354 to test whether the constant selector is valid and can_vec_perm_var_p
6355 to test whether a variable selector is valid.
6357 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6359 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6360 * optabs-query.c (can_vec_perm_p): Likewise.
6361 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6362 instead of vec_perm_indices.
6363 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6364 (vect_gen_perm_mask_checked): Likewise,
6365 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6366 (vect_gen_perm_mask_checked): Likewise,
6368 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6370 * optabs-query.h (qimode_for_vec_perm): Declare.
6371 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6372 (qimode_for_vec_perm): ...this new function.
6373 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6375 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6377 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6378 does not have a conditional at the top.
6380 2018-01-02 Richard Biener <rguenther@suse.de>
6382 * ipa-inline.c (big_speedup_p): Fix expression.
6384 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6387 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6390 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6394 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6395 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6396 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6397 cond_taken_branch_cost 3->4.
6399 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6401 PR tree-optimization/83581
6402 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6403 TODO_cleanup_cfg if any changes have been made.
6406 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6407 convert_modes if target mode has the right side, but different mode
6411 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6412 last argument when extracting from CONCAT. If either from_real or
6413 from_imag is NULL, use expansion through memory. If result is not
6414 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6415 the parts directly to inner mode, if even that fails, use expansion
6419 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6420 check for bswap in mode rather than HImode and use that in expand_unop
6423 Copyright (C) 2018 Free Software Foundation, Inc.
6425 Copying and distribution of this file, with or without modification,
6426 are permitted in any medium without royalty provided the copyright
6427 notice and this notice are preserved.