1 2018-02-15 Jakub Jelinek <jakub@redhat.com>
3 PR tree-optimization/84383
4 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
5 dstoff nor call operand_equal_p if dstbase is NULL.
7 PR tree-optimization/84334
8 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
9 also a CONSTANT_CLASS_P, punt.
11 2018-02-14 Jim Wilson <jimw@sifive.com>
13 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
14 first SMALL_OPERAND check. New local min_second_step. Move assert
15 to where locals are set. Add TARGET_RVC support.
16 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
18 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
20 * doc/invoke.texi: Correct -Wformat-overflow code sample.
22 2018-02-14 Martin Sebor <msebor@redhat.com>
24 PR tree-optimization/83698
25 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
26 arrays constrain the offset range to their bounds.
27 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
28 (builtin_access::overlap): Avoid setting the size of overlap if it's
30 (maybe_diag_overlap): Also consider arrays when deciding what values
31 of offsets to include in diagnostics.
33 2018-02-14 Martin Sebor <msebor@redhat.com>
36 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
37 that correspond to the kind of a declaration.
39 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
42 * config/pa/pa.md: Load address of PIC label using the linkage table
43 if the label is nonlocal.
45 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
47 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
48 warning message if user requests -maltivec=be.
49 * doc/invoke.texi: Document deprecation of -maltivec=be.
51 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
54 * config/rs6000/rs6000-c.c: Update definitions for
55 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
56 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
58 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
61 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
62 add _get_ssp intrinsics. Remove argument from
63 __builtin_ia32_rdssp[d|q].
64 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
65 * config/i386/i386-builtin.def: Remove argument from
66 __builtin_ia32_rdssp[d|q].
67 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
68 ix86_expand_special_args_builtin for _rdssp[d|q].
69 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
70 Clear register before usage.
71 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
72 Add documentation for new _get_ssp and _inc_ssp intrinsics.
74 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
76 PR tree-optimization/84357
77 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
78 operand 1 of an ARRAY_REF too.
80 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
83 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
84 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
86 (set_of_reg): New struct.
87 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
88 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
89 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
91 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
92 Split into bitclr, bitset, bitinvert patterns if appropriate.
93 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
94 use rx_fuse_in_memory_bitop.
95 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
96 to named insn, correct maximum insn length.
98 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
101 * machmode.def: Define a complex mode for PARTIAL_INT.
102 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
104 * doc/rtl.texi: Document CSPImode.
105 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
107 (msp430_hard_regno_nregs_with_padding): Likewise.
109 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
112 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
114 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
116 PR rtl-optimization/84169
117 * combine.c (try_combine): New variable split_i2i3. Set it to true if
118 we generated a parallel as new i3 and we split that to new i2 and i3
119 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
120 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
121 those to i2, not i1. Partially rewrite this scan code.
123 2018-02-13 Jakub Jelinek <jakub@redhat.com>
126 * stor-layout.c (place_field): For variable length fields, adjust
127 offset_align afterwards not just based on the field's alignment,
128 but also on the size.
131 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
132 of exps and logs in the use_exp2 case.
134 2018-02-13 Jeff Law <law@redhat.com>
136 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
139 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
142 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
146 * common.opt (gas-loc-support, gas-locview-support): New.
147 (ginline-points, ginternal-reset-location-views): New.
148 * doc/invoke.texi: Document them. Use @itemx where intended.
149 (gvariable-location-views): Adjust.
150 * target.def (reset_location_view): New.
151 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
152 (TARGET_RESET_LOCATION_VIEW): New.
153 * doc/tm.texi: Rebuilt.
154 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
155 (dwarf2out_default_as_locview_support): New.
156 (output_asm_line_debug_info): Use option variables.
157 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
158 (output_loc_list): Likewise.
159 (add_high_low_attributes): Check option variables.
160 Don't output entry view attribute in strict mode.
161 (gen_inlined_subroutine_die): Check option variables.
162 (dwarf2out_inline_entry): Likewise.
163 (init_sections_and_labels): Likewise.
164 (dwarf2out_early_finish): Likewise.
165 (maybe_reset_location_view): New, from...
166 (dwarf2out_var_location): ... here. Call it.
167 * debug.h (dwarf2out_default_as_loc_support): Declare.
168 (dwarf2out_default_as_locview_support): Declare.
169 * hooks.c (hook_int_rtx_insn_0): New.
170 * hooks.h (hook_int_rtx_insn_0): Declare.
171 * toplev.c (process_options): Take -gas-loc-support and
172 -gas-locview-support from dwarf2out. Enable
173 -gvariable-location-views by default only with locview
174 assembler support. Enable -ginternal-reset-location-views by
175 default only if the target defines the corresponding hook.
176 Enable -ginline-points by default if location views are
177 enabled; force it disabled if statement frontiers are
179 * tree-inline.c (expand_call_inline): Check option variables.
180 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
182 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
184 PR tree-optimization/84321
185 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
186 handling. Also check whether the anti-range contains any values
187 that satisfy the mask; switch to a VR_RANGE if not.
189 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
192 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
194 2018-02-13 Martin Jambor <mjambor@suse.cz>
197 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
198 of call statements, also set location of a load to a temporary.
200 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
202 * config/rl78/rl78.c (add_vector_labels): New function.
203 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
204 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
205 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
206 which checks that no arguments are passed.
207 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
208 * doc/extend.texi: Documentation for the new attribute.
210 2018-02-13 Andreas Schwab <schwab@suse.de>
212 * config/riscv/linux.h (CPP_SPEC): Define.
214 2018-02-13 Jakub Jelinek <jakub@redhat.com>
217 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
218 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
219 OPTION_MASK_ISA_AES as first argument to def_builtin_const
220 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
221 instead of OPTION_MASK_ISA_PCLMUL as first argument to
222 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
223 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
224 temporarily for AES and PCLMUL builtins.
226 PR tree-optimization/84339
227 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
228 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
232 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
233 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
234 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
236 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
238 * omp-simd-clone.h: New file.
239 * omp-simd-clone.c: Include omp-simd-clone.h.
240 (expand_simd_clones): No longer static.
241 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
242 cgraph.h and omp-simd-clone.h.
243 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
244 (vect_recog_widen_shift_pattern): Formatting fix.
245 (vect_pattern_recog_1): Don't check optab for calls.
248 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
249 operands[2] into a REG before using gen_lowpart on it.
251 2018-02-12 Jeff Law <law@redhat.com>
254 * config/sh/sh.c (find_barrier): Consider a sibling call
257 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
258 successfully back substituting a reg.
260 2018-02-12 Richard Biener <rguenther@suse.de>
262 PR tree-optimization/84037
263 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
264 parameter, move visited init to caller.
265 (vect_slp_analyze_operations): Separate cost from validity
266 check, initialize visited once for all instances.
267 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
269 * tree-vect-stmts.c (vect_model_simple_cost): Make early
271 (vect_model_promotion_demotion_cost): Likewise.
272 (vectorizable_bswap): Guard cost modeling with !slp_node
273 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
275 (vectorizable_call): Likewise.
276 (vectorizable_conversion): Likewise.
277 (vectorizable_assignment): Likewise.
278 (vectorizable_shift): Likewise.
279 (vectorizable_operation): Likewise.
280 (vectorizable_store): Likewise.
281 (vectorizable_load): Likewise.
282 (vectorizable_condition): Likewise.
283 (vectorizable_comparison): Likewise.
285 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
288 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
289 (ASAN_MARK): Fix fnspec to account for return value, change pointer
290 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
292 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
295 * params.def (inline-min-speedup): Increase from 8 to 15.
296 (max-inline-insns-auto): Decrease from 40 to 30.
297 * ipa-split.c (consider_split): Add some buffer for function to
298 be considered inlining candidate.
299 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
302 2018-02-12 Richard Biener <rguenther@suse.de>
304 PR tree-optimization/84037
305 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
306 matched stmts if we cannot swap the non-matched ones.
308 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
310 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
311 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
312 _mm_maskz_scalef_round_ss): New intrinsics.
313 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
314 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
315 __builtin_ia32_scalefss_round): Remove.
316 (__builtin_ia32_scalefsd_mask_round,
317 __builtin_ia32_scalefss_mask_round): New intrinsics.
318 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
319 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
320 ((match_operand:VF_128 2 "<round_nimm_predicate>"
321 "<round_constraint>")): Changed to ...
322 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
323 "<round_scalar_constraint>")): ... this.
324 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
325 %0, %1, %2<round_op3>}"): Changed to ...
326 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
327 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
328 %2<round_scalar_mask_op3>}"): ... this.
329 * config/i386/subst.md (round_scalar_nimm_predicate): New.
331 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
333 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
334 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
335 (_mm_maskz_sqrt_round_ss): New intrinsics.
336 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
337 (__builtin_ia32_sqrtsd_mask_round)
338 (__builtin_ia32_sqrtss_mask_round): New builtins.
339 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
340 (__builtin_ia32_sqrtss_round): Remove.
341 (__builtin_ia32_sqrtsd_mask_round)
342 (__builtin_ia32_sqrtss_mask_round): New builtins.
343 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
344 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
345 ((match_operand:VF_128 1 "vector_operand"
346 "xBm,<round_constraint>")): Changed to ...
347 ((match_operand:VF_128 1 "vector_operand"
348 "xBm,<round_scalar_constraint>")): ... this.
349 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
350 %0, %2, %<iptr>1<round_op3>}): Changed to ...
351 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
352 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
353 %<iptr>1<round_scalar_mask_op3>}): ... this.
354 ((set_attr "prefix" "<round_prefix>")): Changed to ...
355 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
357 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
360 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
361 Cast vec_cmpeq result to correct type.
362 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
363 Cast vec_cmpgt result to correct type.
365 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
367 * final.c (final_scan_insn_1): Renamed from...
368 (final_scan_insn): ... this. New wrapper, to recover
369 seen from the outermost call in recursive ones.
370 * config/sparc/sparc.c (output_return): Drop seen from call.
371 (output_sibcall): Likewise.
372 * config/visium/visium.c (output_branch): Likewise.
374 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
376 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
379 2018-02-10 Alan Modra <amodra@gmail.com>
382 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
383 Specify LR as an input.
385 2018-02-10 Jakub Jelinek <jakub@redhat.com>
388 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
389 remove_member_access_dummy_vars): New functions.
390 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
391 lower_omp_1, execute_lower_omp): Use them.
393 PR rtl-optimization/84308
394 * shrink-wrap.c (spread_components): Release todo vector.
396 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
398 PR rtl-optimization/57193
399 * ira-color.c (struct allocno_color_data): Add member
400 conflict_allocno_hard_prefs.
401 (update_conflict_allocno_hard_prefs): New.
402 (bucket_allocno_compare_func): Add a preference based on
403 conflict_allocno_hard_prefs.
404 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
405 (color_allocnos): Remove a dead code. Initiate
406 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
408 2018-02-09 Jakub Jelinek <jakub@redhat.com>
411 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
412 constraint from =wa to wa. Avoid a subreg on the output operand,
413 instead use a pseudo and subreg it in a move.
414 (p9_xxbrd_<mode>): Changed to ...
415 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
416 (p9_xxbrd_v2df): New expander.
417 (p9_xxbrw_<mode>): Changed to ...
418 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
419 (p9_xxbrw_v4sf): New expander.
421 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
423 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
425 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
428 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
429 multiply in 32-bit mode.
430 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
431 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
434 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
436 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
437 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
438 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
439 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
441 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
444 * dwarf2out.c (is_trivial_indirect_ref): New function.
445 (dwarf2out_late_global_decl): Do not generate a location
446 attribute for variables that have a non-trivial DECL_VALUE_EXPR
447 and that are not defined in the current unit.
449 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
451 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
452 instead of a libcall for UNORDERED.
454 2018-02-09 Tamar Christina <tamar.christina@arm.com>
457 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
458 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
460 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
463 * config/s390/s390.c (s390_set_current_function): Invoke
464 s390_indirect_branch_settings also if fndecl didn't change.
466 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
468 * config/rs6000/rs6000.md (blockage): Set length to zero.
470 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
472 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
474 2018-02-09 Jakub Jelinek <jakub@redhat.com>
477 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
478 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
482 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
483 PARALLEL incoming that failed vt_get_decl_and_offset check.
486 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
487 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
489 (get_variable_section): For decls in named .bss* sections pass true as
490 second argument to bss_initializer_p.
492 2018-02-09 Marek Polacek <polacek@redhat.com>
493 Jakub Jelinek <jakub@redhat.com>
496 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
497 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
498 Sync some changes from cxx_fold_indirect_ref.
500 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
502 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
504 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
505 (BLOCK_INLINE_ENTRY_LABEL): New.
506 (dwarf2out_var_location): Disregard inline entry markers.
507 (inline_entry_data): New struct.
508 (inline_entry_data_hasher): New hashtable type.
509 (inline_entry_data_hasher::hash): New.
510 (inline_entry_data_hasher::equal): New.
511 (inline_entry_data_table): New variable.
512 (add_high_low_attributes): Add DW_AT_entry_pc and
513 DW_AT_GNU_entry_view attributes if a pending entry is found
514 in inline_entry_data_table. Add old entry_pc attribute only
515 if debug nonbinding markers are disabled.
516 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
518 (block_within_block_p, dwarf2out_inline_entry): New.
519 (dwarf2out_finish): Check that no entries remained in
520 inline_entry_data_table.
521 * final.c (reemit_insn_block_notes): Handle inline entry notes.
522 (final_scan_insn, notice_source_line): Likewise.
523 (rest_of_clean_state): Skip inline entry markers.
524 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
526 * gimple.c (gimple_build_debug_inline_entry): New.
527 * gimple.h (enum gimple_debug_subcode): Add
528 GIMPLE_DEBUG_INLINE_ENTRY.
529 (gimple_build_debug_inline_entry): Declare.
530 (gimple_debug_inline_entry_p): New.
531 (gimple_debug_nonbind_marker_p): Adjust.
532 * insn-notes.def (INLINE_ENTRY): New.
533 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
534 inline entry marker notes.
535 (print_insn): Likewise.
536 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
537 (INSN_DEBUG_MARKER_KIND): Likewise.
538 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
539 * tree-inline.c (expand_call_inline): Build and insert
540 debug_inline_entry stmt.
541 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
542 inline entry blocks early, if nonbind markers are enabled.
543 (dump_scope_block): Dump fragment info.
544 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
545 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
546 (gimple_build_debug_inline_entry): New.
547 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
548 Enable/disable inline entry points too.
549 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
550 (DEBUG_INSN): Describe inline entry markers.
552 * common.opt (gvariable-location-views): New.
553 (gvariable-location-views=incompat5): New.
554 * config.in: Rebuilt.
555 * configure: Rebuilt.
556 * configure.ac: Test assembler for view support.
557 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
558 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
559 * dwarf2out.c (var_loc_view): New typedef.
560 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
561 (dwarf2out_locviews_in_attribute): New.
562 (dwarf2out_locviews_in_loclist): New.
563 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
564 (enum dw_line_info_opcode): Add LI_adv_address.
565 (struct dw_line_info_table): Add view.
566 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
567 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
568 (zero_view_p): New variable.
569 (ZERO_VIEW_P): New macro.
570 (output_asm_line_debug_info): New.
571 (struct var_loc_node): Add view.
572 (add_AT_view_list, AT_loc_list): New.
573 (add_var_loc_to_decl): Add view param. Test it against last.
574 (new_loc_list): Add view params. Record them.
575 (AT_loc_list_ptr): Handle loc and view lists.
576 (view_list_to_loc_list_val_node): New.
577 (print_dw_val): Handle dw_val_class_view_list.
578 (size_of_die): Likewise.
579 (value_format): Likewise.
580 (loc_list_has_views): New.
581 (gen_llsym): Set vl_symbol too.
582 (maybe_gen_llsym, skip_loc_list_entry): New.
583 (dwarf2out_maybe_output_loclist_view_pair): New.
584 (output_loc_list): Output view list or entries too.
585 (output_view_list_offset): New.
586 (output_die): Handle dw_val_class_view_list.
587 (output_dwarf_version): New.
588 (output_compilation_unit_header): Use it.
589 (output_skeleton_debug_sections): Likewise.
590 (output_rnglists, output_line_info): Likewise.
591 (output_pubnames, output_aranges): Update version comments.
592 (output_one_line_info_table): Output view numbers in asm comments.
593 (dw_loc_list): Determine current endview, pass it to new_loc_list.
594 Call maybe_gen_llsym.
595 (loc_list_from_tree_1): Adjust.
596 (add_AT_location_description): Create view list attribute if
597 needed, check it's absent otherwise.
598 (convert_cfa_to_fb_loc_list): Adjust.
599 (maybe_emit_file): Call output_asm_line_debug_info for test.
600 (dwarf2out_var_location): Reset views as needed. Precompute
601 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
603 (new_line_info_table): Reset next view.
604 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
605 (dwarf2out_source_line): Likewise. Output view resets and labels to
606 the assembler, or select appropriate line info opcodes.
607 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
608 (optimize_string_length): Catch it. Adjust.
609 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
610 dw_val_class_view_list, and remove it if no longer needed.
611 (hash_loc_list): Hash view numbers.
612 (loc_list_hasher::equal): Compare them.
613 (optimize_location_lists): Check whether a view list symbol is
614 needed, and whether the locview attribute is present, and
615 whether they match. Remove the locview attribute if no longer
617 (index_location_lists): Call skip_loc_list_entry for test.
618 (dwarf2out_finish): Call output_asm_line_debug_info for test.
619 Use output_dwarf_version.
620 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
621 (struct dw_val_node): Add val_view_list.
622 * final.c (SEEN_NEXT_VIEW): New.
623 (set_next_view_needed): New.
624 (clear_next_view_needed): New.
625 (maybe_output_next_view): New.
626 (final_start_function): Rename to...
627 (final_start_function_1): ... this. Take pointer to FIRST,
628 add SEEN parameter. Emit param bindings in the initial view.
629 (final_start_function): Reintroduce SEEN-less interface.
630 (final): Rename to...
631 (final_1): ... this. Take SEEN parameter. Output final pending
632 next view at the end.
633 (final): Reintroduce seen-less interface.
634 (final_scan_insn): Output pending next view before switching
635 sections or ending a block. Mark the next view as needed when
636 outputting variable locations. Notify debug backend of section
637 changes, and of location view changes.
638 (rest_of_handle_final): Adjust.
639 * toplev.c (process_options): Autodetect value for debug variable
640 location views option. Warn on incompat5 without -gdwarf-5.
641 * doc/invoke.texi (gvariable-location-views): New.
642 (gvariable-location-views=incompat5): New.
643 (gno-variable-location-views): New.
645 2018-02-08 David Malcolm <dmalcolm@redhat.com>
647 PR tree-optimization/84136
648 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
649 that the result of find_edge is non-NULL.
651 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
654 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
655 storing integer register in SImode. Fix cost of 256 and 512
656 byte aligned SSE register store.
658 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
660 * config/i386/i386.c (ix86_multiplication_cost): Fix
661 multiplication cost for TARGET_AVX512DQ.
663 2018-02-08 Marek Polacek <polacek@redhat.com>
665 PR tree-optimization/84238
666 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
669 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
671 PR tree-optimization/84265
672 * tree-vect-stmts.c (vectorizable_store): Don't treat
673 VMAT_CONTIGUOUS accesses as grouped.
674 (vectorizable_load): Likewise.
676 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
678 PR tree-optimization/81635
679 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
680 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
681 (test_round_for_mask): New functions.
682 (wide_int_cc_tests): Call test_round_for_mask.
683 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
684 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
685 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
686 range returned by get_range_info.
688 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
691 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
692 * symtab.c: Include builtins.h
693 (symtab_node::output_to_lto_symbol_table_p): Move here
694 from lto-streamer-out.c:output_symbol_p.
695 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
696 (output_symbol_p): Move all logic to symtab.c
697 (produce_symtab): Update.
699 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
701 * config/s390/s390-opts.h (enum indirect_branch): Define.
702 * config/s390/s390-protos.h (s390_return_addr_from_memory)
703 (s390_indirect_branch_via_thunk)
704 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
705 (enum s390_indirect_branch_type): Define.
706 * config/s390/s390.c (struct s390_frame_layout, struct
707 machine_function): Remove.
708 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
709 (indirect_branch_table_label_no, indirect_branch_table_name):
711 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
712 (enum s390_indirect_branch_option): Define.
713 (s390_return_addr_from_memory): New function.
714 (s390_handle_string_attribute): New function.
715 (s390_attribute_table): Add new attribute handler.
716 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
717 (s390_indirect_branch_via_thunk): New function.
718 (s390_indirect_branch_via_inline_thunk): New function.
719 (s390_function_ok_for_sibcall): When jumping via thunk disallow
720 sibling call optimization for non z10 compiles.
721 (s390_emit_call): Force indirect branch target to be a single
722 register. Add r1 clobber for non-z10 compiles.
723 (s390_emit_epilogue): Emit return jump via return_use expander.
724 (s390_reorg): Handle JUMP_INSNs as execute targets.
725 (s390_option_override_internal): Perform validity checks for the
726 new command line options.
727 (s390_indirect_branch_attrvalue): New function.
728 (s390_indirect_branch_settings): New function.
729 (s390_set_current_function): Invoke s390_indirect_branch_settings.
730 (s390_output_indirect_thunk_function): New function.
731 (s390_code_end): Implement target hook.
732 (s390_case_values_threshold): Implement target hook.
733 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
735 * config/s390/s390.h (struct s390_frame_layout)
736 (struct machine_function): Move here from s390.c.
737 (TARGET_INDIRECT_BRANCH_NOBP_RET)
738 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
739 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
740 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
741 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
742 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
743 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
744 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
745 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
746 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
747 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
748 (mnemonic attribute): Add values which aren't recognized
750 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
751 pattern for branch conversion. Fix mnemonic attribute.
752 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
753 indirect branch via thunk if requested.
754 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
755 ("*indirect_jump"): Disable for branch conversion using out of
757 ("indirect_jump_via_thunk<mode>_z10")
758 ("indirect_jump_via_thunk<mode>")
759 ("indirect_jump_via_inlinethunk<mode>_z10")
760 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
761 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
762 ("casesi_jump_via_inlinethunk<mode>_z10")
763 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
764 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
765 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
766 ("*indirect2_jump"): Disable for branch conversion.
767 ("casesi_jump"): Turn into expander and expand patterns for branch
769 ("return_use"): New expander.
770 ("*return"): Emit return via thunk and rename it to ...
771 ("*return<mode>"): ... this one.
772 * config/s390/s390.opt: Add new options and and enum for the
775 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
777 * lra-constraints.c (match_reload): Unconditionally use
778 gen_lowpart_SUBREG, rather than selecting between that
779 and equivalent gen_rtx_SUBREG code.
781 2018-02-08 Richard Biener <rguenther@suse.de>
783 PR tree-optimization/84233
784 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
785 changed flag instead of boguously re-using phi_inserted.
787 2018-02-08 Martin Jambor <mjambor@suse.cz>
789 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
790 static local variables.
792 2018-02-08 Richard Biener <rguenther@suse.de>
794 PR tree-optimization/84278
795 * tree-vect-stmts.c (vectorizable_store): When looking for
796 smaller vector types to perform grouped strided loads/stores
797 make sure the mode is supported by the target.
798 (vectorizable_load): Likewise.
800 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
802 * config/aarch64/aarch64.c (aarch64_components_for_bb):
803 Increase LDP/STP opportunities by adding adjacent callee-saves.
805 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
807 PR rtl-optimization/84068
808 PR rtl-optimization/83459
809 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
811 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
813 PR tree-optimization/84224
814 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
815 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
818 2018-02-07 Iain Sandoe <iain@codesourcery.com>
821 * config/rs6000/altivec.md (*restore_world): Remove LR use.
822 * config/rs6000/predicates.md (restore_world_operation): Adjust op
823 count, remove one USE.
825 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
827 * doc/install.texi (Configuration): Document the
828 --with-long-double-format={ibm,ieee} PowerPC configuration
832 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
833 Convert from define_expand to be define_insn_and_split. Rework
834 float/double/_Float128 conversions to QI/HI/SImode to work with
835 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
836 conversions to QI/HImode types did a store and then a load to
837 truncate the value. For conversions to VSX registers, don't split
838 the insn, instead emit the code directly. Use the code iterator
839 any_fix to combine signed and unsigned conversions.
840 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
841 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
842 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
843 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
844 (fix_<mode>di2_hw): Likewise.
845 (fixuns_<mode>di2_hw): Likewise.
846 (fix_<mode>si2_hw): Likewise.
847 (fixuns_<mode>si2_hw): Likewise.
848 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
849 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
850 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
851 fix<uns>_trunc<SFDF:mode>si2_p8.
852 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
854 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
855 (fix<uns>_<mode>_mem): Likewise.
856 (fctiw<u>z_<mode>_mem): Likewise.
857 (fix<uns>_<mode>_mem): Likewise.
858 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
859 the register allocator from doing a direct move to the GPRs to do
860 a store, and instead use the ISA 3.0 store byte/half-word from
861 vector register instruction. For IEEE 128-bit floating point,
862 also optimize stores of 32-bit ints.
863 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
865 2018-02-07 Alan Hayward <alan.hayward@arm.com>
867 * genextract.c (push_pathstr_operand): New function to support
869 (walk_rtx): Call push_pathstr_operand.
870 (print_path): Support [a-zA-Z].
872 2018-02-07 Richard Biener <rguenther@suse.de>
874 PR tree-optimization/84037
875 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
876 (cse_and_gimplify_to_preheader): Declare.
877 (vect_get_place_in_interleaving_chain): Likewise.
878 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
880 (_loop_vec_info::~_loop_vec_info): Delete it.
881 (cse_and_gimplify_to_preheader): New function.
882 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
883 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
884 (vectorizable_load): Likewise. For grouped stores always base
885 the IV on the first element.
886 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
887 condition before gimplifying.
889 2018-02-07 Jakub Jelinek <jakub@redhat.com>
891 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
892 *DIV_EXPR and *MOD_EXPR.
894 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
897 * config/i386/i386.c (ix86_option_override_internal): Mask out
898 the CF_SET bit when checking -fcf-protection.
900 2018-02-07 Tom de Vries <tom@codesourcery.com>
903 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
906 2018-02-07 Richard Biener <rguenther@suse.de>
908 PR tree-optimization/84204
909 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
912 PR tree-optimization/84205
913 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
914 special-case isl_ast_op_zdiv_r.
916 PR tree-optimization/84223
917 * graphite-scop-detection.c (gather_bbs::before_dom_children):
918 Only add conditions from within the region.
919 (gather_bbs::after_dom_children): Adjust.
921 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
924 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
925 * config/avr/avr.md: Only post-reload split REG-REG moves if
926 either register is REGERAL_REG_P.
928 2018-02-07 Jakub Jelinek <jakub@redhat.com>
930 PR tree-optimization/84235
931 * tree-ssa-scopedtables.c
932 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
933 if the subtraction is performed in floating point type where NaNs are
934 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
935 build 1. Formatting fix.
937 2018-02-06 Jakub Jelinek <jakub@redhat.com>
940 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
941 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
942 and skip it regardless of bb boundaries. Use CALL_P macro,
943 don't test INSN_P (insn) together with CALL_P or JUMP_P check
944 unnecessarily, formatting fix.
946 2018-02-06 Michael Collison <michael.collison@arm.com>
948 * config/arm/thumb2.md:
949 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
950 (*thumb_mov_notscc): Ditto.
952 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
955 * config/rs6000/rs6000.md (su code attribute): Use "u" for
956 unsigned_fix, not "s".
958 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
960 * configure.ac (gcc_fn_eh_frame_ro): New function.
961 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
962 correct .eh_frame permissions.
963 * configure: Regenerate.
965 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
967 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
970 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
972 * config/rs6000/rs6000.c (rs6000_option_override_internal):
973 Display warning message for -mno-speculate-indirect-jumps.
975 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
977 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
979 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
981 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
983 PR tree-optimization/84225
984 * tree-eh.c (find_trapping_overflow): Only call
985 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
987 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
990 * config/i386/i386.c: Reimplement the check of possible options
991 -mibt/-mshstk conbination. Change error messages.
992 * doc/invoke.texi: Fix a typo: remove extra '='.
994 2018-02-06 Marek Polacek <polacek@redhat.com>
996 PR tree-optimization/84228
997 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
999 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1002 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1003 emitted arch directives.
1004 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1005 __ARM_FEATURE_COPROC before changing architectures.
1007 2018-02-06 Richard Biener <rguenther@suse.de>
1009 * config/i386/i386.c (print_reg): Fix typo.
1010 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1012 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1014 * configure: Regenerate.
1016 2018-02-05 Martin Sebor <msebor@redhat.com>
1018 PR tree-optimization/83369
1019 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1022 2018-02-05 Martin Liska <mliska@suse.cz>
1024 * doc/invoke.texi: Cherry-pick upstream r323995.
1026 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1028 * ira.c (ira_init_register_move_cost): Adjust comment.
1030 2018-02-05 Martin Liska <mliska@suse.cz>
1032 PR gcov-profile/84137
1033 * doc/gcov.texi: Fix typo in documentation.
1035 2018-02-05 Martin Liska <mliska@suse.cz>
1037 PR gcov-profile/83879
1038 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1039 using dlopen functionality.
1041 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1043 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1044 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1045 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1046 _mm_maskz_range_round_ss): New intrinsics.
1047 (__builtin_ia32_rangesd128_round)
1048 (__builtin_ia32_rangess128_round): Remove.
1049 (__builtin_ia32_rangesd128_mask_round,
1050 __builtin_ia32_rangess128_mask_round): New builtins.
1051 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1052 __builtin_ia32_rangess128_round): Remove.
1053 (__builtin_ia32_rangesd128_mask_round,
1054 __builtin_ia32_rangess128_mask_round): New builtins.
1055 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1056 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1057 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1058 "<round_saeonly_constraint>")): Changed to ...
1059 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1060 "<round_saeonly_scalar_constraint>")): ... this.
1061 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1062 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1063 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1064 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1065 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1067 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1069 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1071 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1072 Remove all values except native, 8540 and 8548.
1074 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1076 * config/i386/i386.c (ix86_output_function_return): Pass
1077 INVALID_REGNUM, instead of -1, as invalid register number to
1078 indirect_thunk_name and output_indirect_thunk.
1080 2018-02-02 Julia Koval <julia.koval@intel.com>
1082 * config.gcc: Add -march=icelake.
1083 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1084 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1085 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1086 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1087 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1088 (processor_target_table): Add icelake.
1089 (ix86_option_override_internal): Handle new PTAs.
1090 (get_builtin_code_for_version): Handle icelake.
1091 (M_INTEL_COREI7_ICELAKE): New.
1092 (fold_builtin_cpu): Handle icelake.
1093 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1094 * doc/invoke.texi: Add -march=icelake.
1096 2018-02-02 Julia Koval <julia.koval@intel.com>
1098 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1099 to wide_int_bitmask.
1100 * wide-int-bitmask.h: New.
1102 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1105 * config/i386/i386.md: Replace Pmode with word_mode in
1106 builtin_setjmp_setup and builtin_longjmp to support x32.
1108 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1112 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1114 (rs6000_supported_cpu_names): New static variable.
1115 (linux_cpu_translation_table): Likewise.
1116 (elf_platform) <cpu>: Define new static variable and use it.
1117 Translate kernel AT_PLATFORM name to canonical name if needed.
1118 Error if platform name is unknown.
1120 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1123 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1125 2018-02-01 Jeff Law <law@redhat.com>
1128 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1129 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1130 the scratch if RELEASE_VIA_POP is false.
1131 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1132 If we have to save a temporary register, decrement SIZE appropriately.
1133 Pass new arguments to release_scratch_register_on_entry.
1134 (ix86_adjust_stack_and_probe): Likewise.
1135 (ix86_emit_probe_stack_range): Pass new arguments to
1136 release_scratch_register_on_entry.
1138 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1140 PR rtl-optimization/84157
1141 * combine.c (change_zero_ext): Use REG_P predicate in
1142 front of HARD_REGISTER_P predicate.
1144 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1146 * config/avr/avr.c (avr_option_override): Move disabling of
1147 -fdelete-null-pointer-checks to...
1148 * common/config/avr/avr-common.c (avr_option_optimization_table):
1151 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1153 PR tree-optimization/81635
1154 * tree-data-ref.c (split_constant_offset_1): For types that
1155 wrap on overflow, try to use range info to prove that wrapping
1158 2018-02-01 Renlin Li <renlin.li@arm.com>
1161 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1163 (aarch64_register_move_cost): Likewise.
1164 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1166 (REG_CLASS_NAMES): Likewise.
1167 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1168 TAILCALL_ADDR_REGS. Remove IP registers.
1169 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1171 2018-02-01 Richard Biener <rguenther@suse.de>
1173 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1174 for specifying RPO order and allow NULL for that.
1175 * domwalk.c (dom_walker::dom_walker): Likewise.
1176 (dom_walker::walk): Handle NULL RPO order.
1177 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1179 (rewrite_update_dom_walker): Likewise.
1180 (mark_def_dom_walker): Likewise.
1182 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1184 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1185 (aarch64_maybe_expand_sve_subreg_move): Declare.
1186 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1187 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1189 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1190 that are semantically a reverse operation.
1191 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1192 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1193 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1195 (aarch64_can_change_mode_class): For big-endian, forbid changes
1196 between two SVE modes if they have different element sizes.
1198 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1200 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1201 the TImode handling for big-endian targets.
1203 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1205 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1206 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1208 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1209 Remove BSWAP handing for big-endian targets and use the form of
1210 LD1RQ appropariate for the mode.
1212 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1214 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1215 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1218 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1221 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1222 check for operands that need to go through aarch64_sve_reload_be.
1224 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1226 PR tree-optimization/81661
1227 PR tree-optimization/84117
1228 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1229 * tree-eh.c: Include gimplify.h.
1230 (find_trapping_overflow, replace_trapping_overflow,
1231 rewrite_to_non_trapping_overflow): New functions.
1232 * tree-vect-loop.c: Include tree-eh.h.
1233 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1234 * tree-data-ref.c: Include tree-eh.h.
1235 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1237 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1239 PR rtl-optimization/84123
1240 * combine.c (change_zero_ext): Check if hard register satisfies
1241 can_change_dest_mode before calling gen_lowpart_SUBREG.
1243 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1246 * ira.c (ira_init_register_move_cost): Remove assert.
1248 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1250 PR rtl-optimization/84071
1251 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1252 * doc/tm.texi: Regenerate.
1254 2018-01-31 Richard Biener <rguenther@suse.de>
1256 PR tree-optimization/84132
1257 * tree-data-ref.c (analyze_miv_subscript): Properly
1258 check whether evolution_function_is_affine_multivariate_p
1259 before calling gcd_of_steps_may_divide_p.
1261 2018-01-31 Julia Koval <julia.koval@intel.com>
1264 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1265 * config/i386/i386.md (rdpid_rex64) New.
1266 (rdpid): Make 32bit only.
1268 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1271 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1272 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1274 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1277 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1279 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1281 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1283 PR rtl-optimization/84071
1284 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1285 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1287 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1289 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1290 (arc_attribute_table): Add 'aux' attribute.
1291 (arc_in_small_data_p): Consider aux like variables.
1292 (arc_is_aux_reg_p): New function.
1293 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1294 (arc_get_aux_arg): New function.
1295 (prepare_move_operands): Handle aux-register access.
1296 (arc_handle_aux_attribute): New function.
1297 * doc/extend.texi (ARC Variable attributes): Add subsection.
1299 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1301 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1302 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1303 (arc_attribute_table): Add 'uncached' attribute.
1304 (arc_print_operand): Print '.di' flag for uncached memory
1306 (arc_in_small_data_p): Do not consider for small data the uncached
1308 (arc_is_uncached_mem_p): New function.
1309 * config/arc/predicates.md (compact_store_memory_operand): Check
1310 for uncached memory accesses.
1311 (nonvol_nonimm_operand): Likewise.
1312 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1314 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1317 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1318 falign-loops=): Add Optimization flag.
1320 2018-01-30 Jeff Law <law@redhat.com>
1323 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1324 INT_REGISTERS_SAVED. Check it prior to calling
1325 get_scratch_register_on_entry.
1326 (ix86_adjust_stack_and_probe): Similarly.
1327 (ix86_emit_probe_stack_range): Similarly.
1328 (ix86_expand_prologue): Corresponding changes.
1330 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1333 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1334 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1336 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1339 * lra-constraints.c (curr_insn_transform): Process AND in the
1342 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1344 PR rtl-optimization/83986
1345 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1346 dependence against last_pending_memory_flush in addition to
1349 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1351 PR tree-optimization/81611
1352 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1355 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1357 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1360 2018-01-30 Richard Biener <rguenther@suse.de>
1361 Jakub Jelinek <jakub@redhat.com>
1363 PR tree-optimization/84111
1364 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1365 inner loops added during recursion, as they don't have up-to-date
1368 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1371 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1372 (can_inline_edge_by_limits_p): ... here.
1373 (can_early_inline_edge_p, check_callers,
1374 update_caller_keys, update_callee_keys, recursive_inlining,
1375 add_new_edges_to_heap, speculation_useful_p,
1376 inline_small_functions,
1377 inline_small_functions, flatten_function,
1378 inline_to_all_callers_1): Update.
1380 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1382 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1385 2018-01-30 Richard Biener <rguenther@suse.de>
1387 PR tree-optimization/83008
1388 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1389 invariant and constant vector uses in stmts when they need
1392 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1395 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1396 * configure: Regenerate.
1398 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1400 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1402 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1403 Use gen_rtx_REG rather than gen_lowpart.
1405 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1407 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1408 rather than 0 when creating partial subregs.
1410 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1412 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1415 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1418 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1419 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1420 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1421 flags. This restores the settings used before the 2017-07-24.
1422 Turning off pre increment/decrement/modify allows IVOPTS to
1423 optimize DF/SF loops where the index is an int.
1425 2018-01-29 Richard Biener <rguenther@suse.de>
1426 Kelvin Nilsen <kelvin@gcc.gnu.org>
1429 * tree-vect-stmts.c (vectorizable_call): Don't call
1430 targetm.vectorize_builtin_md_vectorized_function if callee is
1433 2018-01-22 Carl Love <cel@us.ibm.com>
1435 * doc/extend.tex: Fix typo in second arg in
1436 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1438 2018-01-29 Richard Biener <rguenther@suse.de>
1440 PR tree-optimization/84086
1441 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1442 (flush_ssaname_freelist): When SSA names were released reset
1443 the SCEV hash table.
1445 2018-01-29 Richard Biener <rguenther@suse.de>
1447 PR tree-optimization/84057
1448 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1449 removed paths when removing edges.
1451 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1453 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1454 -mfunction-return=@var{choice}.
1456 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1459 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1461 (layout::print_source_line): Likewise.
1462 (test_get_line_width_without_trailing_whitespace): Add test cases.
1464 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1467 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1470 2018-01-26 Jim Wilson <jimw@sifive.com>
1472 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1474 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1477 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1479 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1480 and CMP + SUB-immediate -> SUBS.
1482 2018-01-26 Martin Sebor <msebor@redhat.com>
1484 PR tree-optimization/83896
1485 * tree-ssa-strlen.c (get_string_len): Rename...
1486 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1487 Avoid assuming length is constant.
1488 (handle_char_store): Use HOST_WIDE_INT for string length.
1490 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1493 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1494 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1496 2018-01-26 Richard Biener <rguenther@suse.de>
1498 PR rtl-optimization/84003
1499 * dse.c (record_store): Only record redundant stores when
1500 the earlier store aliases at least all accesses the later one does.
1502 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1504 PR rtl-optimization/83985
1505 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1506 REG_CFA_RESTORE insns.
1507 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1508 REG_CFA_RESTORE insns here.
1511 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1512 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1514 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1516 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1517 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1518 (arc_init): Likewise.
1519 (arc_override_options): Likewise.
1520 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1522 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1524 * config/arc/arc.h (TARGET_DBNZ): Define.
1525 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1526 properly set the tune attribute.
1527 (dbnz): Use TARGET_DBNZ guard.
1528 * config/arc/arc.opt (mtune): Add core3 option.
1530 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1532 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1533 recognize new pic like addresses.
1534 (arc_delegitimize_address): Clean up.
1536 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1538 * config/arc/arc-arches.def: Option mrf16 valid for all
1540 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1541 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1542 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1543 * config/arc/arc-tables.opt: Regenerate.
1544 * config/arc/arc.c (arc_conditional_register_usage): Handle
1545 reduced register file case.
1546 (arc_file_start): Set must have build attributes.
1547 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1549 * config/arc/arc.opt (mrf16): Add new option.
1550 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1551 * config/arc/genmultilib.awk: Handle new mrf16 option.
1552 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1553 * config/arc/t-multilib: Regenerate.
1554 * doc/invoke.texi (ARC Options): Document mrf16 option.
1556 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1558 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1559 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1560 (arc_attribute_table): Add 'secure_call' attribute.
1561 (arc_print_operand): Print secure call operand.
1562 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1564 (arc_is_secure_call_p): New function. * config/arc/arc.md
1565 (call_i): Add support for sjli instruction.
1566 (call_value_i): Likewise.
1567 * config/arc/constraints.md (Csc): New constraint.
1569 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1570 John Eric Martin <John.Martin@emmicro-us.com>
1572 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1573 * config/arc/arc.c (_arc_jli_section): New struct.
1574 (arc_jli_section): New type.
1575 (rc_jli_sections): New static variable.
1576 (arc_handle_jli_attribute): New function.
1577 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1578 (arc_file_end): New function.
1579 (TARGET_ASM_FILE_END): Define.
1580 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1581 (arc_add_jli_section): New function.
1582 (jli_call_scan): Likewise.
1583 (arc_reorg): Call jli_call_scan.
1584 (arc_output_addsi): Remove 'S' from printing asm operand.
1585 (arc_is_jli_call_p): New function.
1586 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1588 (movhi_insn): Likewise.
1589 (movsi_insn): Likewise.
1590 (movsi_set_cc_insn): Likewise.
1591 (loadqi_update): Likewise.
1592 (load_zeroextendqisi_update): Likewise.
1593 (load_signextendqisi_update): Likewise.
1594 (loadhi_update): Likewise.
1595 (load_zeroextendhisi_update): Likewise.
1596 (load_signextendhisi_update): Likewise.
1597 (loadsi_update): Likewise.
1598 (loadsf_update): Likewise.
1599 (movsicc_insn): Likewise.
1600 (bset_insn): Likewise.
1601 (bxor_insn): Likewise.
1602 (bclr_insn): Likewise.
1603 (bmsk_insn): Likewise.
1604 (bicsi3_insn): Likewise.
1605 (cmpsi_cc_c_insn): Likewise.
1606 (movsi_ne): Likewise.
1607 (movsi_cond_exec): Likewise.
1608 (clrsbsi2): Likewise.
1619 (call_i): Remove 'S' asm letter, add jli instruction.
1620 (call_value_i): Likewise.
1621 * config/arc/arc.op (mjli-always): New option.
1622 * config/arc/constraints.md (Cji): New constraint.
1623 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1625 (subsf3_fpx): Likewise.
1626 (mulsf3_fpx): Likewise.
1627 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1629 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1631 * doc/invoke.texi (ARC): Document mjli-always option.
1633 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1635 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
1636 avoid addition with 0 and use incw and decw where possible.
1638 2018-01-26 Richard Biener <rguenther@suse.de>
1640 PR tree-optimization/81082
1641 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1642 association if it requires casting to unsigned.
1643 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1644 from fold_plusminus_mult_expr to catch important cases late when
1645 range info is available.
1647 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1649 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1650 * configure.ac (hidden_linkonce): New test.
1651 * configure: Regenerate.
1652 * config.in: Regenerate.
1654 2018-01-26 Julia Koval <julia.koval@intel.com>
1656 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1657 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1658 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1659 _mm_mask_bitshuffle_epi64_mask): Fix type.
1660 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1661 USI_FTYPE_V4DI_V4DI_USI): Remove.
1662 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1663 __builtin_ia32_vpshufbitqmb256_mask,
1664 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1665 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1666 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1668 2018-01-26 Alan Modra <amodra@gmail.com>
1671 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1672 UNSPEC_VBPERMQ. Sort other unspecs.
1674 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1676 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1678 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1681 * predict.c (drop_profile): Do not push/pop cfun; update also
1683 (handle_missing_profiles): Fix logic looking for zero profiles.
1685 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1688 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1689 on functions with #pragma omp declare simd or functions with simd
1691 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1692 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1693 Remove trailing \n from warning_at calls.
1695 2018-01-25 Tom de Vries <tom@codesourcery.com>
1698 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1699 for neutered workers.
1701 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1704 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1705 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1707 2018-01-24 Jeff Law <law@redhat.com>
1710 * i386.c (get_probe_interval): Move to earlier point.
1711 (ix86_compute_frame_layout): If -fstack-clash-protection and
1712 the frame is larger than the probe interval, then use pushes
1713 to save registers rather than reg->mem moves.
1714 (ix86_expand_prologue): Remove conditional for int_registers_saved
1717 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1720 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1721 min/max for never referenced object.
1723 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1726 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1728 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1729 attributes from DECL_ATTRIBUTES (decl) without affecting
1730 DECL_ATTRIBUTES (current_function_decl).
1731 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1732 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1734 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1736 PR tree-optimization/83979
1737 * fold-const.c (fold_comparison): Use constant_boolean_node
1738 instead of boolean_{true,false}_node.
1740 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1742 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1745 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1747 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1748 Simplify the clause that sets the length attribute.
1749 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1750 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1751 clause that sets the length attribute.
1752 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1754 2018-01-24 Tom de Vries <tom@codesourcery.com>
1757 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1758 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1759 Add strict parameter.
1760 (prevent_branch_around_nothing): Insert dummy insn between branch to
1761 label and label with no ptx insn inbetween.
1762 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1764 2018-01-24 Tom de Vries <tom@codesourcery.com>
1767 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1768 for neutered threads in warp.
1769 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1771 2018-01-24 Richard Biener <rguenther@suse.de>
1773 PR tree-optimization/83176
1774 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1777 2018-01-24 Richard Biener <rguenther@suse.de>
1779 PR tree-optimization/82819
1780 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1781 code generating pluses that are no-ops in the target precision.
1783 2018-01-24 Richard Biener <rguenther@suse.de>
1786 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1788 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1790 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1791 to merge probabilities.
1792 * predict.c (probably_never_executed): Also mark as cold functions
1793 with global 0 profile and guessed local profile.
1794 * profile-count.c (profile_probability::combine_with_count): New
1796 * profile-count.h (profile_probability::operator*,
1797 profile_probability::operator*=, profile_probability::operator/,
1798 profile_probability::operator/=): Reduce precision to adjusted
1799 and set value to guessed on contradictory divisions.
1800 (profile_probability::combine_with_freq): Remove.
1801 (profile_probability::combine_wiht_count): Declare.
1802 (profile_count::force_nonzero):: Set to adjusted.
1803 (profile_count::probability_in):: Set quality to adjusted.
1804 * tree-ssa-tail-merge.c (replace_block_by): Use
1807 2018-01-23 Andrew Waterman <andrew@sifive.com>
1808 Jim Wilson <jimw@sifive.com>
1810 * config/riscv/riscv.c (riscv_stack_boundary): New.
1811 (riscv_option_override): Set riscv_stack_boundary. Handle
1812 riscv_preferred_stack_boundary_arg.
1813 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1814 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1815 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1816 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1817 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1818 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1820 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1823 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1824 of struct ix86_frame.
1825 (ix86_expand_epilogue): Likewise. Add a local variable for
1826 the reg_save_offset field in struct ix86_frame.
1828 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1830 PR tree-optimization/82604
1831 * tree-loop-distribution.c (enum partition_kind): New enum item
1832 PKIND_PARTIAL_MEMSET.
1833 (partition_builtin_p): Support above new enum item.
1834 (generate_code_for_partition): Ditto.
1835 (compute_access_range): Differentiate cases that equality can be
1836 proven at all loops, the innermost loops or no loops.
1837 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1838 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1839 (finalize_partitions, distribute_loop): Don't fuse partition of
1840 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1841 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1844 2018-01-23 Martin Liska <mliska@suse.cz>
1846 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1847 order to ignore the predictor.
1848 (PRED_POLYMORPHIC_CALL): Likewise.
1849 (PRED_RECURSIVE_CALL): Likewise.
1851 2018-01-23 Martin Liska <mliska@suse.cz>
1853 * tree-profile.c (tree_profiling): Print function header to
1854 aware reader which function we are working on.
1855 * value-prof.c (gimple_find_values_to_profile): Do not print
1856 not interesting value histograms.
1858 2018-01-23 Martin Liska <mliska@suse.cz>
1860 * profile-count.h (enum profile_quality): Add
1861 profile_uninitialized as the first value. Do not number values
1862 as they are zero based.
1863 (profile_count::verify): Update sanity check.
1864 (profile_probability::verify): Likewise.
1866 2018-01-23 Nathan Sidwell <nathan@acm.org>
1868 * doc/invoke.texi (ffor-scope): Deprecate.
1870 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1872 PR tree-optimization/83510
1873 * domwalk.c (set_all_edges_as_executable): New function.
1874 (dom_walker::dom_walker): Convert bool param
1875 "skip_unreachable_blocks" to enum reachability. Move setup of
1876 edge flags to set_all_edges_as_executable and only do it when
1877 reachability is REACHABLE_BLOCKS.
1878 * domwalk.h (enum dom_walker::reachability): New enum.
1879 (dom_walker::dom_walker): Convert bool param
1880 "skip_unreachable_blocks" to enum reachability.
1881 (set_all_edges_as_executable): New decl.
1882 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1883 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1885 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1886 but converting true to REACHABLE_BLOCKS.
1887 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1889 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1890 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1891 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1893 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1894 if check_all_array_refs will be called.
1896 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1898 * tree.c (selftest::test_location_wrappers): Add more test
1901 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1903 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1904 (selftest::test_bit_in_range): Likewise.
1906 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1909 * doc/sourcebuild.texi (vect_float): Say that the selector
1910 only describes the situation when -funsafe-math-optimizations is on.
1911 (vect_float_strict): Document.
1913 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1915 PR tree-optimization/83965
1916 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1917 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1918 instead of checking only for a reduction.
1919 (vect_recog_widen_sum_pattern): Likewise.
1921 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1923 * predict.c (probably_never_executed): Only use precise profile info.
1924 (compute_function_frequency): Skip after inlining hack since we now
1925 have quality checking.
1927 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1929 * profile-count.h (profile_probability::very_unlikely,
1930 profile_probability::unlikely, profile_probability::even): Set
1931 precision to guessed.
1933 2018-01-23 Richard Biener <rguenther@suse.de>
1935 PR tree-optimization/83963
1936 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1937 Properly terminate dominator walk when crossing the exit edge not
1938 when visiting its source block.
1940 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1943 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1944 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1946 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1948 PR tree-optimization/83957
1949 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1950 semicolon after for body surrounded by braces.
1952 PR tree-optimization/83081
1953 * profile-count.h (profile_probability::split): New method.
1954 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1955 Use profile_probability::split.
1956 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1957 when splitting a single conditional jump into 2.
1959 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1961 PR tree-optimization/69452
1962 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1965 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1967 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
1968 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
1969 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
1971 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1973 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
1974 * config/rl78/rl78.md (movdi): New define_expand.
1975 * config/rl78/rl78.c (rl78_split_movdi): New function.
1977 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1980 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1982 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1983 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1984 128-bit to produce an UNSPEC move to get the double word with the
1985 signbit and then a shift directly to do signbit.
1986 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1987 implementation with a new version that just does either a direct
1988 move or a regular move. Move memory interface to separate insns.
1989 Move insns so they are next to the expander.
1990 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1991 with signbit move. Split big and little endian case.
1992 (signbit<mode>2_dm_mem_le): Likewise.
1993 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1994 (signbit<mode>2_dm2): Likewise.
1996 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1998 * config/rl78/rl78.md (anddi3): New define_expand.
2000 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2002 * config/rl78/rl78.md (umindi3): New define_expand.
2004 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2006 * config/rl78/rl78.md (smindi3): New define_expand.
2008 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2010 * config/rl78/rl78.md (smaxdi3): New define_expand.
2012 2018-01-22 Carl Love <cel@us.ibm.com>
2014 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2015 LVX_V1TI): Add macro expansion.
2016 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2017 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2018 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2019 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2020 Change check to determine if the instruction is a byte reversing
2021 entry. Fix typo in comment.
2022 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2023 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2024 Add def_builtin calls for new builtins.
2025 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2026 Add define_insn expansion.
2028 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2030 * config/rl78/rl78.md (umaxdi3): New define_expand.
2032 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2034 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2035 for non-QImode registers.
2037 2018-01-22 Richard Biener <rguenther@suse.de>
2039 PR tree-optimization/83963
2040 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2041 including the loop exit block.
2042 (scop_detection::merge_sese): Likewise.
2043 (scop_detection::add_scop): Do it here instead.
2045 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2047 * doc/sourcebuild.texi (arm_softfloat): Document.
2049 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2052 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2053 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2054 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2056 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2057 David Edelsohn <dje.gcc@gmail.com>
2060 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2061 Change "crset eq" to "crset 2".
2062 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2063 (*call_indirect_aix<mode>_nospec): Likewise.
2064 (*call_value_indirect_aix<mode>_nospec): Likewise.
2065 (*call_indirect_elfv2<mode>_nospec): Likewise.
2066 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2067 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2068 change assembly output from . to $.
2069 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2070 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2071 (*tablejump<mode>_internal1_nospec): Likewise.
2073 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2076 * config/sh/sh_optimize_sett_clrt.cc:
2077 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2079 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2081 PR tree-optimization/83940
2082 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2083 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2084 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2085 use it to pass back the definition type.
2086 (vect_check_store_rhs): Likewise rhs_dt_out.
2087 (vect_build_gather_load_calls): Add a mask_dt argument and use
2088 it instead of a call to vect_is_simple_use.
2089 (vectorizable_store): Update calls to vect_check_load_store_mask
2090 and vect_check_store_rhs. Use the dt returned by the latter instead
2091 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2092 instead of calls to vect_is_simple_use. Pass the scalar rather
2093 than the vector operand to vect_is_simple_use when handling
2094 second and subsequent copies of an rhs value.
2095 (vectorizable_load): Update calls to vect_check_load_store_mask
2096 and vect_build_gather_load_calls. Use the cached mask_dt and
2097 gs_info.offset_dt instead of calls to vect_is_simple_use.
2099 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2102 * tree-emutls.c: Include gimplify.h.
2103 (lower_emutls_2): New function.
2104 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2105 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2106 it before further processing.
2109 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2110 UINTVAL (trueop1) instead of INTVAL (op1).
2112 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2116 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2117 INCOMING_FRAME_SP_OFFSET if not defined.
2118 (scan_trace): Add ENTRY argument. If true and
2119 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2120 emit a note to adjust the CFA offset.
2121 (create_cfi_notes): Adjust scan_trace callers.
2122 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2123 INCOMING_FRAME_SP_OFFSET in the CIE.
2124 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2125 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2127 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2128 * doc/tm.texi: Regenerated.
2130 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2132 PR rtl-optimization/83147
2133 * lra-constraints.c (remove_inheritance_pseudos): Use
2134 lra_substitute_pseudo_within_insn.
2136 2018-01-19 Tom de Vries <tom@codesourcery.com>
2137 Cesar Philippidis <cesar@codesourcery.com>
2140 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2142 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2145 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2146 spaces for function labels.
2148 2018-01-19 Martin Liska <mliska@suse.cz>
2150 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2151 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2152 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2153 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2154 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2155 (PRED_CONST_RETURN): Change from 69 to 65.
2156 (PRED_NULL_RETURN): Change from 91 to 71.
2157 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2158 (PRED_LOOP_GUARD): Change from 66 to 73.
2160 2018-01-19 Martin Liska <mliska@suse.cz>
2162 * predict.c (predict_insn_def): Add new assert.
2163 (struct branch_predictor): Change type to signed integer.
2164 (test_prediction_value_range): Amend test to cover
2166 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2167 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2168 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2169 (PRED_LOOP_IV_COMPARE): Likewise.
2170 * predict.h (PROB_UNINITIALIZED): Define new constant.
2172 2018-01-19 Martin Liska <mliska@suse.cz>
2174 * predict.c (dump_prediction): Add new format for
2175 analyze_brprob.py script which is enabled with -details
2177 * profile-count.h (precise_p): New function.
2179 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2181 PR tree-optimization/83922
2182 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2183 there are no statements that need masking.
2184 (vect_active_double_reduction_p): New function.
2185 (vect_analyze_loop_operations): Use it when handling phis that
2186 are not in the loop header.
2188 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2190 PR tree-optimization/83914
2191 * tree-vect-loop.c (vectorizable_induction): Don't convert
2192 init_expr or apply the peeling adjustment for inductions
2193 that are nested within the vectorized loop.
2195 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2197 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2200 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2204 * function.h (gimplify_parameters): Add gimple_seq * argument.
2205 * function.c: Include gimple.h and options.h.
2206 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2207 for the added local temporaries if needed.
2208 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2209 if there are any parameter cleanups, wrap whole body into a
2210 try/finally with the cleanups.
2212 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2215 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2216 Use GET_MODE_CLASS for scalar floating point.
2218 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2222 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2223 Fix call of call_cgraph_insertion_hooks.
2225 2018-01-18 Martin Sebor <msebor@redhat.com>
2227 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2229 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2232 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2235 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2238 * common.opt: (-ffile-prefix-map): New option.
2239 * opts.c (common_handle_option): Defer it.
2240 * opts-global.c (handle_common_deferred_options): Handle it.
2241 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2242 * file-prefix-map.h: New file.
2243 (remap_debug_filename, add_debug_prefix_map): ...here.
2244 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2245 * final.c (debug_prefix_map, add_debug_prefix_map
2246 remap_debug_filename): Move to...
2247 * file-prefix-map.c: New file.
2248 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2249 generalize, get rid of alloca(), use strrchr() instead of strchr().
2250 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2251 Implement in terms of add_prefix_map().
2252 (remap_macro_filename, remap_debug_filename): Implement in term of
2254 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2255 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2256 * dbxout.c: Include file-prefix-map.h.
2257 * varasm.c: Likewise.
2258 * vmsdbgout.c: Likewise.
2259 * xcoffout.c: Likewise.
2260 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2261 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2262 * doc/invoke.texi (-ffile-prefix-map): Document.
2263 (-fdebug-prefix-map): Update description.
2265 2018-01-18 Martin Liska <mliska@suse.cz>
2267 * config/i386/i386.c (indirect_thunk_name): Document that also
2269 (output_indirect_thunk): Document why both instructions
2270 (pause and lfence) are generated.
2272 2018-01-18 Richard Biener <rguenther@suse.de>
2274 PR tree-optimization/83887
2275 * graphite-scop-detection.c
2276 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2277 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2278 (scop_detection::merge_sese): Re-implement with a flood-fill
2279 algorithm that properly finds a SESE region if it exists.
2281 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2284 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2285 pointer_diff optimizations use view_convert instead of convert.
2287 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2289 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2290 Generate different code for -mno-speculate-indirect-jumps.
2291 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2292 (*call_indirect_aix<mode>): Disable for
2293 -mno-speculate-indirect-jumps.
2294 (*call_indirect_aix<mode>_nospec): New define_insn.
2295 (*call_value_indirect_aix<mode>): Disable for
2296 -mno-speculate-indirect-jumps.
2297 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2298 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2299 -mno-speculate-indirect-jumps.
2300 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2302 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2304 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2305 long double type, set the flags for noting the default long double
2306 type, even if we don't pass or return a long double type.
2308 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2311 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2314 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2316 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2317 support for merge[hl].
2318 (fold_mergehl_helper): New helper function.
2319 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2320 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2321 (altivec_vmrglw_direct): Add xxmrglw insn.
2323 2018-01-17 Andrew Waterman <andrew@sifive.com>
2325 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2326 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2328 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2331 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2332 call the lto_location_cache before reading the
2333 DECL_SOURCE_LOCATION of the types.
2335 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2336 Richard Sandiford <richard.sandiford@linaro.org>
2338 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2339 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2340 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2341 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2342 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2344 * config/aarch64/constraints.md (aarch64_movti_operand):
2346 * config/aarch64/predicates.md (Uti): Add new constraint.
2348 2018-01-17 Carl Love <cel@us.ibm.com>
2350 * config/rs6000/vsx.md (define_expand xl_len_r,
2351 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2352 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2354 (define_expand, define_insn): Move the shift left from the
2355 define_insn to the define_expand for lxvl and stxvl instructions.
2356 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2357 and XL_LEN_R definitions to PURE.
2359 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2361 * config/i386/i386.c (indirect_thunk_name): Declare regno
2362 as unsigned int. Compare regno with INVALID_REGNUM.
2363 (output_indirect_thunk): Ditto.
2364 (output_indirect_thunk_function): Ditto.
2365 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2366 in the call to output_indirect_thunk_function.
2368 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2371 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2372 rather than the size of inner_type to determine the stack slot size
2373 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2375 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2378 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2381 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2383 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2384 endian Linux systems to optionally enable multilibs for selecting
2385 the long double type if the user configured an explicit type.
2386 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2387 have no long double multilibs if not defined.
2388 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2389 warn if the user used -mabi={ieee,ibm}longdouble and we built
2390 multilibs for long double.
2391 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2392 appropriate multilib option.
2393 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2395 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2396 for building long double multilibs.
2397 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2399 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2401 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2404 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2406 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2409 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2412 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2415 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2417 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2418 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2420 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2422 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2423 different rtl trees depending on TARGET_64BIT.
2424 (rs6000_gen_lvx): Likewise.
2426 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2428 * config/visium/visium.md (nop): Tweak comment.
2429 (hazard_nop): Likewise.
2431 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2433 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2434 -mspeculate-indirect-jumps.
2435 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2436 for -mno-speculate-indirect-jumps.
2437 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2438 (*call_value_indirect_elfv2<mode>): Disable for
2439 -mno-speculate-indirect-jumps.
2440 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2441 (indirect_jump): Emit different RTL for
2442 -mno-speculate-indirect-jumps.
2443 (*indirect_jump<mode>): Disable for
2444 -mno-speculate-indirect-jumps.
2445 (*indirect_jump<mode>_nospec): New define_insn.
2446 (tablejump): Emit different RTL for
2447 -mno-speculate-indirect-jumps.
2448 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2449 (tablejumpsi_nospec): New define_expand.
2450 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2451 (tablejumpdi_nospec): New define_expand.
2452 (*tablejump<mode>_internal1): Disable for
2453 -mno-speculate-indirect-jumps.
2454 (*tablejump<mode>_internal1_nospec): New define_insn.
2455 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2458 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2460 * caller-save.c (insert_save): Drop unnecessary parameter. All
2463 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2464 Richard Biener <rguenth@suse.de>
2467 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2468 return early, inline manually is_gimple_sizepos. Make sure if we
2469 call gimplify_expr we don't end up with a gimple constant.
2470 * tree.c (variably_modified_type_p): Don't return true for
2471 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2472 * gimplify.h (is_gimple_sizepos): Remove.
2474 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2476 PR tree-optimization/83857
2477 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2478 vectorizable_live_operation for pure SLP statements.
2479 (vectorizable_live_operation): Handle PHIs.
2481 2018-01-16 Richard Biener <rguenther@suse.de>
2483 PR tree-optimization/83867
2484 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2485 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2487 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2490 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2491 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2492 If off is not INTEGER_CST, issue a may not be aligned warning
2493 rather than isn't aligned. Use isn%'t rather than isn't.
2494 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2496 <case MULT_EXPR>: Improve the case when bottom and one of the
2497 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2498 operand, in that case check if the other operand is multiple of
2499 bottom divided by the INTEGER_CST operand.
2501 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2504 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2505 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2506 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2507 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2508 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2509 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2510 (pa_function_arg_size): New function.
2512 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2514 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2515 in a separate statement.
2517 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2519 PR tree-optimization/83847
2520 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2521 group gathers and scatters.
2523 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2525 PR rtl-optimization/86620
2526 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2528 PR rtl-optimization/83213
2529 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2530 to last if both are JUMP_INSNs.
2532 PR tree-optimization/83843
2533 * gimple-ssa-store-merging.c
2534 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2535 store_immediate_info for bswap/nop orig_stores.
2537 2018-01-15 Andrew Waterman <andrew@sifive.com>
2539 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2541 <UDIV>: Increase cost if !TARGET_DIV.
2543 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2545 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2546 (define_attr "cr_logical_3op"): New.
2547 (cceq_ior_compare): Adjust.
2548 (cceq_ior_compare_complement): Adjust.
2549 (*cceq_rev_compare): Adjust.
2550 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2551 (is_cracked_insn): Adjust.
2552 (insn_must_be_first_in_group): Adjust.
2553 * config/rs6000/40x.md: Adjust.
2554 * config/rs6000/440.md: Adjust.
2555 * config/rs6000/476.md: Adjust.
2556 * config/rs6000/601.md: Adjust.
2557 * config/rs6000/603.md: Adjust.
2558 * config/rs6000/6xx.md: Adjust.
2559 * config/rs6000/7450.md: Adjust.
2560 * config/rs6000/7xx.md: Adjust.
2561 * config/rs6000/8540.md: Adjust.
2562 * config/rs6000/cell.md: Adjust.
2563 * config/rs6000/e300c2c3.md: Adjust.
2564 * config/rs6000/e500mc.md: Adjust.
2565 * config/rs6000/e500mc64.md: Adjust.
2566 * config/rs6000/e5500.md: Adjust.
2567 * config/rs6000/e6500.md: Adjust.
2568 * config/rs6000/mpc.md: Adjust.
2569 * config/rs6000/power4.md: Adjust.
2570 * config/rs6000/power5.md: Adjust.
2571 * config/rs6000/power6.md: Adjust.
2572 * config/rs6000/power7.md: Adjust.
2573 * config/rs6000/power8.md: Adjust.
2574 * config/rs6000/power9.md: Adjust.
2575 * config/rs6000/rs64.md: Adjust.
2576 * config/rs6000/titan.md: Adjust.
2578 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2580 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2581 ix86_indirect_branch_register logic.
2583 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2585 * config/i386/constraints.md (Bs): Update
2586 ix86_indirect_branch_register check. Don't check
2587 ix86_indirect_branch_register with GOT_memory_operand.
2589 * config/i386/predicates.md (GOT_memory_operand): Don't check
2590 ix86_indirect_branch_register here.
2591 (GOT32_symbol_operand): Likewise.
2593 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2595 * config/i386/predicates.md (constant_call_address_operand):
2596 Rewrite ix86_indirect_branch_register logic.
2597 (sibcall_insn_operand): Likewise.
2599 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2601 * config/i386/constraints.md (Bs): Replace
2602 ix86_indirect_branch_thunk_register with
2603 ix86_indirect_branch_register.
2605 * config/i386/i386.md (indirect_jump): Likewise.
2606 (tablejump): Likewise.
2607 (*sibcall_memory): Likewise.
2608 (*sibcall_value_memory): Likewise.
2609 Peepholes of indirect call and jump via memory: Likewise.
2610 * config/i386/i386.opt: Likewise.
2611 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2612 (GOT_memory_operand): Likewise.
2613 (call_insn_operand): Likewise.
2614 (sibcall_insn_operand): Likewise.
2615 (GOT32_symbol_operand): Likewise.
2617 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2620 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2621 type rather than type addr's type points to.
2622 (expand_omp_atomic_mutex): Likewise.
2623 (expand_omp_atomic): Likewise.
2625 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2628 * config/i386/i386.c (output_indirect_thunk_function): Use
2629 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2630 for __x86_return_thunk.
2632 2018-01-15 Richard Biener <rguenther@suse.de>
2635 * expmed.c (extract_bit_field_1): Fix typo.
2637 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2640 * config/arm/iterators.md (VF): New mode iterator.
2641 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2642 Remove integer-related logic from pattern.
2643 (neon_vabd<mode>_3): Likewise.
2645 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2648 * common.opt (fstrict-overflow): No longer an alias.
2649 (fwrapv-pointer): New option.
2650 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2651 also for pointer types based on flag_wrapv_pointer.
2652 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2653 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2654 opts->x_flag_wrapv got set.
2655 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2656 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2657 POINTER_TYPE_OVERFLOW_UNDEFINED.
2658 * match.pd: Likewise in address comparison pattern.
2659 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2661 2018-01-15 Richard Biener <rguenther@suse.de>
2664 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2665 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2666 Reset type names to their identifier if their TYPE_DECL doesn't
2667 have linkage (and thus is used for ODR and devirt).
2668 (save_debug_info_for_decl): Remove.
2669 (save_debug_info_for_type): Likewise.
2670 (add_tree_to_fld_list): Adjust.
2671 * tree-pretty-print.c (dump_generic_node): Make dumping of
2672 type names more robust.
2674 2018-01-15 Richard Biener <rguenther@suse.de>
2676 * BASE-VER: Bump to 8.0.1.
2678 2018-01-14 Martin Sebor <msebor@redhat.com>
2681 * builtins.c (check_access): Avoid warning when the no-warning bit
2684 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2686 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2687 * ira-color (allocno_hard_regs_compare): Likewise.
2689 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2692 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2693 Use .pushsection/.popsection.
2695 2018-01-14 Martin Sebor <msebor@redhat.com>
2698 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2700 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2702 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2703 entry from extra_headers.
2704 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2705 extra_headers, make the list bitwise identical to the i?86-*-* one.
2707 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2709 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2710 -mcmodel=large with -mindirect-branch=thunk,
2711 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2712 -mfunction-return=thunk-extern.
2713 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2714 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2715 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2717 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2719 * config/i386/i386.c (print_reg): Print the name of the full
2720 integer register without '%'.
2721 (ix86_print_operand): Handle 'V'.
2722 * doc/extend.texi: Document 'V' modifier.
2724 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2726 * config/i386/constraints.md (Bs): Disallow memory operand for
2727 -mindirect-branch-register.
2729 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2730 (GOT_memory_operand): Likewise.
2731 (call_insn_operand): Likewise.
2732 (sibcall_insn_operand): Likewise.
2733 (GOT32_symbol_operand): Likewise.
2734 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2735 for -mindirect-branch-register.
2736 (tablejump): Likewise.
2737 (*sibcall_memory): Likewise.
2738 (*sibcall_value_memory): Likewise.
2739 Disallow peepholes of indirect call and jump via memory for
2740 -mindirect-branch-register.
2741 (*call_pop): Replace m with Bw.
2742 (*call_value_pop): Likewise.
2743 (*sibcall_pop_memory): Replace m with Bs.
2744 * config/i386/i386.opt (mindirect-branch-register): New option.
2745 * doc/invoke.texi: Document -mindirect-branch-register option.
2747 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2749 * config/i386/i386-protos.h (ix86_output_function_return): New.
2750 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2751 set function_return_type.
2752 (indirect_thunk_name): Add ret_p to indicate thunk for function
2754 (output_indirect_thunk_function): Pass false to
2755 indirect_thunk_name.
2756 (ix86_output_indirect_branch_via_reg): Likewise.
2757 (ix86_output_indirect_branch_via_push): Likewise.
2758 (output_indirect_thunk_function): Create alias for function
2759 return thunk if regno < 0.
2760 (ix86_output_function_return): New function.
2761 (ix86_handle_fndecl_attribute): Handle function_return.
2762 (ix86_attribute_table): Add function_return.
2763 * config/i386/i386.h (machine_function): Add
2764 function_return_type.
2765 * config/i386/i386.md (simple_return_internal): Use
2766 ix86_output_function_return.
2767 (simple_return_internal_long): Likewise.
2768 * config/i386/i386.opt (mfunction-return=): New option.
2769 (indirect_branch): Mention -mfunction-return=.
2770 * doc/extend.texi: Document function_return function attribute.
2771 * doc/invoke.texi: Document -mfunction-return= option.
2773 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2775 * config/i386/i386-opts.h (indirect_branch): New.
2776 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2777 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2778 with local indirect jump when converting indirect call and jump.
2779 (ix86_set_indirect_branch_type): New.
2780 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2781 (indirectlabelno): New.
2782 (indirect_thunk_needed): Likewise.
2783 (indirect_thunk_bnd_needed): Likewise.
2784 (indirect_thunks_used): Likewise.
2785 (indirect_thunks_bnd_used): Likewise.
2786 (INDIRECT_LABEL): Likewise.
2787 (indirect_thunk_name): Likewise.
2788 (output_indirect_thunk): Likewise.
2789 (output_indirect_thunk_function): Likewise.
2790 (ix86_output_indirect_branch_via_reg): Likewise.
2791 (ix86_output_indirect_branch_via_push): Likewise.
2792 (ix86_output_indirect_branch): Likewise.
2793 (ix86_output_indirect_jmp): Likewise.
2794 (ix86_code_end): Call output_indirect_thunk_function if needed.
2795 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2797 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2798 (ix86_attribute_table): Add indirect_branch.
2799 * config/i386/i386.h (machine_function): Add indirect_branch_type
2800 and has_local_indirect_jump.
2801 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2803 (tablejump): Likewise.
2804 (*indirect_jump): Use ix86_output_indirect_jmp.
2805 (*tablejump_1): Likewise.
2806 (simple_return_indirect_internal): Likewise.
2807 * config/i386/i386.opt (mindirect-branch=): New option.
2808 (indirect_branch): New.
2811 (thunk-inline): Likewise.
2812 (thunk-extern): Likewise.
2813 * doc/extend.texi: Document indirect_branch function attribute.
2814 * doc/invoke.texi: Document -mindirect-branch= option.
2816 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2819 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2821 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2823 * ipa-inline.c (want_inline_small_function_p): Return false if
2824 inlining has already failed with CIF_FINAL_ERROR.
2825 (update_caller_keys): Call want_inline_small_function_p before
2827 (update_callee_keys): Likewise.
2829 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2831 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2833 (rs6000_quadword_masked_address_p): Likewise.
2834 (quad_aligned_load_p): Likewise.
2835 (quad_aligned_store_p): Likewise.
2836 (const_load_sequence_p): Add comment to describe the outer-most loop.
2837 (mimic_memory_attributes_and_flags): New function.
2838 (rs6000_gen_stvx): Likewise.
2839 (replace_swapped_aligned_store): Likewise.
2840 (rs6000_gen_lvx): Likewise.
2841 (replace_swapped_aligned_load): Likewise.
2842 (replace_swapped_load_constant): Capitalize argument name in
2843 comment describing this function.
2844 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2845 and stores that access quad-word aligned addresses and replace
2846 with stvx or lvx instructions when appropriate.
2847 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2848 New function prototype.
2849 (rs6000_quadword_masked_address_p): Likewise.
2850 (rs6000_gen_lvx): Likewise.
2851 (rs6000_gen_stvx): Likewise.
2852 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2853 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2854 when memory address is aligned.
2855 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2856 this split to select lvx instruction when memory address is aligned.
2857 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2858 instruction when memory address is aligned.
2859 (*vsx_le_perm_load_v16qi): Likewise.
2860 (four unnamed splitters): Modify to select the stvx instruction
2861 when memory is aligned.
2863 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2865 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2866 which appears in the queue multiple times.
2868 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2869 Alan Hayward <alan.hayward@arm.com>
2870 David Sherwood <david.sherwood@arm.com>
2872 * tree-vectorizer.h (vec_lower_bound): New structure.
2873 (_loop_vec_info): Add check_nonzero and lower_bounds.
2874 (LOOP_VINFO_CHECK_NONZERO): New macro.
2875 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2876 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2877 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2878 fields. Make seg_len the distance travelled, not including the
2880 (dr_direction_indicator): Declare.
2881 (dr_zero_step_indicator): Likewise.
2882 (dr_known_forward_stride_p): Likewise.
2883 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2885 (runtime_alias_check_p): Allow runtime alias checks with
2887 (operator ==): Compare access_size and align.
2888 (prune_runtime_alias_test_list): Rework for new distinction between
2889 the access_size and seg_len.
2890 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2892 (get_segment_min_max): New function.
2893 (create_intersect_range_checks): Use it.
2894 (dr_step_indicator): New function.
2895 (dr_direction_indicator): Likewise.
2896 (dr_zero_step_indicator): Likewise.
2897 (dr_known_forward_stride_p): Likewise.
2898 * tree-loop-distribution.c (data_ref_segment_size): Return
2899 DR_STEP * (niters - 1).
2900 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2902 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2903 (vect_preserves_scalar_order_p): New function, split out from...
2904 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2905 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2906 (vect_vfa_access_size): New function.
2907 (vect_vfa_align): Likewise.
2908 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2909 (dump_lower_bound): New function.
2910 (vect_check_lower_bound): Likewise.
2911 (vect_small_gap_p): Likewise.
2912 (vectorizable_with_step_bound_p): Likewise.
2913 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2914 depencies if the vectorization factor is 1. Convert the checks
2915 for nonzero steps into checks on the bounds of DR_STEP. Try using
2916 a bunds check for variable steps if the minimum required step is
2917 relatively small. Update calls to the dr_with_seg_len
2918 constructor and to vect_compile_time_alias.
2919 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2921 (vect_loop_versioning): Call it.
2922 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2924 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2926 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2927 Alan Hayward <alan.hayward@arm.com>
2928 David Sherwood <david.sherwood@arm.com>
2930 * doc/sourcebuild.texi (vect_scatter_store): Document.
2931 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2933 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2935 * genopinit.c (main): Add supports_vec_scatter_store and
2936 supports_vec_scatter_store_cached to target_optabs.
2937 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2938 IFN_MASK_SCATTER_STORE.
2939 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2941 * internal-fn.h (internal_store_fn_p): Declare.
2942 (internal_fn_stored_value_index): Likewise.
2943 * internal-fn.c (scatter_store_direct): New macro.
2944 (expand_scatter_store_optab_fn): New function.
2945 (direct_scatter_store_optab_supported_p): New macro.
2946 (internal_store_fn_p): New function.
2947 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2948 IFN_MASK_SCATTER_STORE.
2949 (internal_fn_mask_index): Likewise.
2950 (internal_fn_stored_value_index): New function.
2951 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2953 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2954 * optabs-query.c (supports_vec_scatter_store_p): New function.
2955 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2956 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2957 true for scatter stores.
2958 (vect_gather_scatter_fn_p): Handle scatter stores too.
2959 (vect_check_gather_scatter): Consider using scatter stores if
2960 supports_vec_scatter_store_p.
2961 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2963 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2964 internal_fn_stored_value_index.
2965 (check_load_store_masking): Handle scatter stores too.
2966 (vect_get_store_rhs): Make public.
2967 (vectorizable_call): Use internal_store_fn_p.
2968 (vectorizable_store): Handle scatter store internal functions.
2969 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2970 when deciding whether the end of the group has been reached.
2971 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2972 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2973 (mask_scatter_store<mode>): New insns.
2975 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2976 Alan Hayward <alan.hayward@arm.com>
2977 David Sherwood <david.sherwood@arm.com>
2979 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2980 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2981 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2983 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2984 Use vect_truncate_gather_scatter_offset if we can't treat the
2985 operation as a normal gather load or scatter store.
2986 (get_group_load_store_type): Take the gather_scatter_info
2987 as argument. Try using a gather load or scatter store for
2988 single-element groups.
2989 (get_load_store_type): Update calls to get_group_load_store_type
2990 and vect_use_strided_gather_scatters_p.
2992 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2993 Alan Hayward <alan.hayward@arm.com>
2994 David Sherwood <david.sherwood@arm.com>
2996 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2997 optional tree argument.
2998 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3000 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3001 but continue to use the current value as a fallback.
3002 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3003 to compare the updates.
3004 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3005 (get_load_store_type): Use it when handling a strided access.
3006 (vect_get_strided_load_store_ops): New function.
3007 (vect_get_data_ptr_increment): Likewise.
3008 (vectorizable_load): Handle strided gather loads. Always pass
3009 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3011 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3012 Alan Hayward <alan.hayward@arm.com>
3013 David Sherwood <david.sherwood@arm.com>
3015 * doc/md.texi (gather_load@var{m}): Document.
3016 (mask_gather_load@var{m}): Likewise.
3017 * genopinit.c (main): Add supports_vec_gather_load and
3018 supports_vec_gather_load_cached to target_optabs.
3019 * optabs-tree.c (init_tree_optimization_optabs): Use
3020 ggc_cleared_alloc to allocate target_optabs.
3021 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3022 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3024 * internal-fn.h (internal_load_fn_p): Declare.
3025 (internal_gather_scatter_fn_p): Likewise.
3026 (internal_fn_mask_index): Likewise.
3027 (internal_gather_scatter_fn_supported_p): Likewise.
3028 * internal-fn.c (gather_load_direct): New macro.
3029 (expand_gather_load_optab_fn): New function.
3030 (direct_gather_load_optab_supported_p): New macro.
3031 (direct_internal_fn_optab): New function.
3032 (internal_load_fn_p): Likewise.
3033 (internal_gather_scatter_fn_p): Likewise.
3034 (internal_fn_mask_index): Likewise.
3035 (internal_gather_scatter_fn_supported_p): Likewise.
3036 * optabs-query.c (supports_at_least_one_mode_p): New function.
3037 (supports_vec_gather_load_p): Likewise.
3038 * optabs-query.h (supports_vec_gather_load_p): Declare.
3039 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3040 and memory_type field.
3041 (NUM_PATTERNS): Bump to 15.
3042 * tree-vect-data-refs.c: Include internal-fn.h.
3043 (vect_gather_scatter_fn_p): New function.
3044 (vect_describe_gather_scatter_call): Likewise.
3045 (vect_check_gather_scatter): Try using internal functions for
3046 gather loads. Recognize existing calls to a gather load function.
3047 (vect_analyze_data_refs): Consider using gather loads if
3048 supports_vec_gather_load_p.
3049 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3050 (vect_get_gather_scatter_offset_type): Likewise.
3051 (vect_convert_mask_for_vectype): Likewise.
3052 (vect_add_conversion_to_patterm): Likewise.
3053 (vect_try_gather_scatter_pattern): Likewise.
3054 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3055 (vect_vect_recog_func_ptrs): Add it.
3056 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3057 internal_fn_mask_index and internal_gather_scatter_fn_p.
3058 (check_load_store_masking): Take the gather_scatter_info as an
3059 argument and handle gather loads.
3060 (vect_get_gather_scatter_ops): New function.
3061 (vectorizable_call): Check internal_load_fn_p.
3062 (vectorizable_load): Likewise. Handle gather load internal
3064 (vectorizable_store): Update call to check_load_store_masking.
3065 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3066 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3067 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3068 (aarch64_gather_scale_operand_d): New predicates.
3069 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3070 (mask_gather_load<mode>): New insns.
3072 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3073 Alan Hayward <alan.hayward@arm.com>
3074 David Sherwood <david.sherwood@arm.com>
3076 * optabs.def (fold_left_plus_optab): New optab.
3077 * doc/md.texi (fold_left_plus_@var{m}): Document.
3078 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3079 * internal-fn.c (fold_left_direct): Define.
3080 (expand_fold_left_optab_fn): Likewise.
3081 (direct_fold_left_optab_supported_p): Likewise.
3082 * fold-const-call.c (fold_const_fold_left): New function.
3083 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3084 * tree-parloops.c (valid_reduction_p): New function.
3085 (gather_scalar_reductions): Use it.
3086 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3087 (vect_finish_replace_stmt): Declare.
3088 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3089 (needs_fold_left_reduction_p): New function, split out from...
3090 (vect_is_simple_reduction): ...here. Accept reductions that
3091 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3092 (vect_force_simple_reduction): Also store the reduction type in
3093 the assignment's STMT_VINFO_REDUC_TYPE.
3094 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3095 (merge_with_identity): New function.
3096 (vect_expand_fold_left): Likewise.
3097 (vectorize_fold_left_reduction): Likewise.
3098 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3099 scalar phi in place for it. Check for target support and reject
3100 cases that would reassociate the operation. Defer the transform
3101 phase to vectorize_fold_left_reduction.
3102 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3103 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3104 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3106 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3108 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3109 call to ifc_temp_var.
3111 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3112 Alan Hayward <alan.hayward@arm.com>
3113 David Sherwood <david.sherwood@arm.com>
3115 * target.def (legitimize_address_displacement): Take the original
3116 offset as a poly_int.
3117 * targhooks.h (default_legitimize_address_displacement): Update
3119 * targhooks.c (default_legitimize_address_displacement): Likewise.
3120 * doc/tm.texi: Regenerate.
3121 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3122 as an argument, moving assert of ad->disp == ad->disp_term to...
3123 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3124 Try calling targetm.legitimize_address_displacement before expanding
3125 the address rather than afterwards, and adjust for the new interface.
3126 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3127 Match the new hook interface. Handle SVE addresses.
3128 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3131 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3133 * Makefile.in (OBJS): Add early-remat.o.
3134 * target.def (select_early_remat_modes): New hook.
3135 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3136 * doc/tm.texi: Regenerate.
3137 * targhooks.h (default_select_early_remat_modes): Declare.
3138 * targhooks.c (default_select_early_remat_modes): New function.
3139 * timevar.def (TV_EARLY_REMAT): New timevar.
3140 * passes.def (pass_early_remat): New pass.
3141 * tree-pass.h (make_pass_early_remat): Declare.
3142 * early-remat.c: New file.
3143 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3145 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3147 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3148 Alan Hayward <alan.hayward@arm.com>
3149 David Sherwood <david.sherwood@arm.com>
3151 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3152 vfm1 with a bound_epilog parameter.
3153 (vect_do_peeling): Update calls accordingly, and move the prologue
3154 call earlier in the function. Treat the base bound_epilog as 0 for
3155 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3156 this base when peeling for gaps.
3157 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3158 with fully-masked loops.
3159 (vect_estimate_min_profitable_iters): Handle the single peeled
3160 iteration in that case.
3162 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3163 Alan Hayward <alan.hayward@arm.com>
3164 David Sherwood <david.sherwood@arm.com>
3166 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3167 single-element interleaving even if the size is not a power of 2.
3168 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3169 accesses for single-element interleaving if the group size is
3172 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3173 Alan Hayward <alan.hayward@arm.com>
3174 David Sherwood <david.sherwood@arm.com>
3176 * doc/md.texi (fold_extract_last_@var{m}): Document.
3177 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3178 * optabs.def (fold_extract_last_optab): New optab.
3179 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3180 * internal-fn.c (fold_extract_direct): New macro.
3181 (expand_fold_extract_optab_fn): Likewise.
3182 (direct_fold_extract_optab_supported_p): Likewise.
3183 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3184 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3185 EXTRACT_LAST_REDUCTION.
3186 (get_initial_def_for_reduction): Do not create an initial vector
3187 for EXTRACT_LAST_REDUCTION reductions.
3188 (vectorizable_reduction): Leave the scalar phi in place for
3189 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3190 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3191 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3192 transform phase to vectorizable_condition.
3193 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3195 (vect_finish_stmt_generation): ...here.
3196 (vect_finish_replace_stmt): New function.
3197 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3198 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3200 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3202 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3203 Alan Hayward <alan.hayward@arm.com>
3204 David Sherwood <david.sherwood@arm.com>
3206 * doc/md.texi (extract_last_@var{m}): Document.
3207 * optabs.def (extract_last_optab): New optab.
3208 * internal-fn.def (EXTRACT_LAST): New internal function.
3209 * internal-fn.c (cond_unary_direct): New macro.
3210 (expand_cond_unary_optab_fn): Likewise.
3211 (direct_cond_unary_optab_supported_p): Likewise.
3212 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3213 loops using EXTRACT_LAST.
3214 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3215 (extract_last_<mode>): ...this optab.
3216 (vec_extract<mode><Vel>): Update accordingly.
3218 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3219 Alan Hayward <alan.hayward@arm.com>
3220 David Sherwood <david.sherwood@arm.com>
3222 * target.def (empty_mask_is_expensive): New hook.
3223 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3224 * doc/tm.texi: Regenerate.
3225 * targhooks.h (default_empty_mask_is_expensive): Declare.
3226 * targhooks.c (default_empty_mask_is_expensive): New function.
3227 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3228 if the target says that empty masks are expensive.
3229 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3231 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3233 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3234 Alan Hayward <alan.hayward@arm.com>
3235 David Sherwood <david.sherwood@arm.com>
3237 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3238 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3239 (vect_use_loop_mask_for_alignment_p): New function.
3240 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3241 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3242 niters_skip argument. Make sure that the first niters_skip elements
3243 of the first iteration are inactive.
3244 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3245 Update call to vect_set_loop_masks_directly.
3246 (get_misalign_in_elems): New function, split out from...
3247 (vect_gen_prolog_loop_niters): ...here.
3248 (vect_update_init_of_dr): Take a code argument that specifies whether
3249 the adjustment should be added or subtracted.
3250 (vect_update_init_of_drs): Likewise.
3251 (vect_prepare_for_masked_peels): New function.
3252 (vect_do_peeling): Skip prologue peeling if we're using a mask
3253 instead. Update call to vect_update_inits_of_drs.
3254 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3256 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3257 alignment. Do not include the number of peeled iterations in
3258 the minimum threshold in that case.
3259 (vectorizable_induction): Adjust the start value down by
3260 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3261 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3262 Take the number of skipped iterations into account when calculating
3264 * tree-vect-stmts.c (vect_gen_while_not): New function.
3266 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3267 Alan Hayward <alan.hayward@arm.com>
3268 David Sherwood <david.sherwood@arm.com>
3270 * doc/sourcebuild.texi (vect_fully_masked): Document.
3271 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3273 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3275 (vect_analyze_loop_2): ...here. Don't check the vectorization
3276 factor against the number of loop iterations if the loop is
3279 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3280 Alan Hayward <alan.hayward@arm.com>
3281 David Sherwood <david.sherwood@arm.com>
3283 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3284 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3285 (dump_groups): Update accordingly.
3286 (iv_use::mem_type): New member variable.
3287 (address_p): New function.
3288 (record_use): Add a mem_type argument and initialize the new
3290 (record_group_use): Add a mem_type argument. Use address_p.
3291 Remove obsolete null checks of base_object. Update call to record_use.
3292 (find_interesting_uses_op): Update call to record_group_use.
3293 (find_interesting_uses_cond): Likewise.
3294 (find_interesting_uses_address): Likewise.
3295 (get_mem_type_for_internal_fn): New function.
3296 (find_address_like_use): Likewise.
3297 (find_interesting_uses_stmt): Try find_address_like_use before
3298 calling find_interesting_uses_op.
3299 (addr_offset_valid_p): Use the iv mem_type field as the type
3300 of the addressed memory.
3301 (add_autoinc_candidates): Likewise.
3302 (get_address_cost): Likewise.
3303 (split_small_address_groups_p): Use address_p.
3304 (split_address_groups): Likewise.
3305 (add_iv_candidate_for_use): Likewise.
3306 (autoinc_possible_for_pair): Likewise.
3307 (rewrite_groups): Likewise.
3308 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3309 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3310 (get_alias_ptr_type_for_ptr_address): New function.
3311 (rewrite_use_address): Rewrite address uses in calls that were
3312 identified by find_address_like_use.
3314 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3315 Alan Hayward <alan.hayward@arm.com>
3316 David Sherwood <david.sherwood@arm.com>
3318 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3320 * gimple-expr.h (is_gimple_addressable: Likewise.
3321 * gimple-expr.c (is_gimple_address): Likewise.
3322 * internal-fn.c (expand_call_mem_ref): New function.
3323 (expand_mask_load_optab_fn): Use it.
3324 (expand_mask_store_optab_fn): Likewise.
3326 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3327 Alan Hayward <alan.hayward@arm.com>
3328 David Sherwood <david.sherwood@arm.com>
3330 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3331 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3332 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3333 (cond_umax@var{mode}): Document.
3334 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3335 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3336 (cond_umin_optab, cond_umax_optab): New optabs.
3337 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3338 (COND_IOR, COND_XOR): New internal functions.
3339 * internal-fn.h (get_conditional_internal_fn): Declare.
3340 * internal-fn.c (cond_binary_direct): New macro.
3341 (expand_cond_binary_optab_fn): Likewise.
3342 (direct_cond_binary_optab_supported_p): Likewise.
3343 (get_conditional_internal_fn): New function.
3344 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3345 Cope with reduction statements that are vectorized as calls rather
3347 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3348 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3349 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3350 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3351 (UNSPEC_COND_EOR): New unspecs.
3352 (optab): Add mappings for them.
3353 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3354 (sve_int_op, sve_fp_op): New int attributes.
3356 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3357 Alan Hayward <alan.hayward@arm.com>
3358 David Sherwood <david.sherwood@arm.com>
3360 * optabs.def (while_ult_optab): New optab.
3361 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3362 * internal-fn.def (WHILE_ULT): New internal function.
3363 * internal-fn.h (direct_internal_fn_supported_p): New override
3364 that takes two types as argument.
3365 * internal-fn.c (while_direct): New macro.
3366 (expand_while_optab_fn): New function.
3367 (convert_optab_supported_p): Likewise.
3368 (direct_while_optab_supported_p): New macro.
3369 * wide-int.h (wi::udiv_ceil): New function.
3370 * tree-vectorizer.h (rgroup_masks): New structure.
3371 (vec_loop_masks): New typedef.
3372 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3374 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3375 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3376 (vect_max_vf): New function.
3377 (slpeel_make_loop_iterate_ntimes): Delete.
3378 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3379 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3380 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3381 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3382 internal-fn.h, stor-layout.h and optabs-query.h.
3383 (vect_set_loop_mask): New function.
3384 (add_preheader_seq): Likewise.
3385 (add_header_seq): Likewise.
3386 (interleave_supported_p): Likewise.
3387 (vect_maybe_permute_loop_masks): Likewise.
3388 (vect_set_loop_masks_directly): Likewise.
3389 (vect_set_loop_condition_masked): Likewise.
3390 (vect_set_loop_condition_unmasked): New function, split out from
3391 slpeel_make_loop_iterate_ntimes.
3392 (slpeel_make_loop_iterate_ntimes): Rename to..
3393 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3394 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3395 (vect_do_peeling): Update call accordingly.
3396 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3398 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3399 mask_compare_type, can_fully_mask_p and fully_masked_p.
3400 (release_vec_loop_masks): New function.
3401 (_loop_vec_info): Use it to free the loop masks.
3402 (can_produce_all_loop_masks_p): New function.
3403 (vect_get_max_nscalars_per_iter): Likewise.
3404 (vect_verify_full_masking): Likewise.
3405 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3406 retries, and free the mask rgroups before retrying. Check loop-wide
3407 reasons for disallowing fully-masked loops. Make the final decision
3408 about whether use a fully-masked loop or not.
3409 (vect_estimate_min_profitable_iters): Do not assume that peeling
3410 for the number of iterations will be needed for fully-masked loops.
3411 (vectorizable_reduction): Disable fully-masked loops.
3412 (vectorizable_live_operation): Likewise.
3413 (vect_halve_mask_nunits): New function.
3414 (vect_double_mask_nunits): Likewise.
3415 (vect_record_loop_mask): Likewise.
3416 (vect_get_loop_mask): Likewise.
3417 (vect_transform_loop): Handle the case in which the final loop
3418 iteration might handle a partial vector. Call vect_set_loop_condition
3419 instead of slpeel_make_loop_iterate_ntimes.
3420 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3421 (check_load_store_masking): New function.
3422 (prepare_load_store_mask): Likewise.
3423 (vectorizable_store): Handle fully-masked loops.
3424 (vectorizable_load): Likewise.
3425 (supportable_widening_operation): Use vect_halve_mask_nunits for
3427 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3428 (vect_gen_while): New function.
3429 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3430 (aarch64_uqdec<mode>): New insn.
3432 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3433 Alan Hayward <alan.hayward@arm.com>
3434 David Sherwood <david.sherwood@arm.com>
3436 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3437 (reduc_xor_scal_optab): New optabs.
3438 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3439 (reduc_xor_scal_@var{m}): Document.
3440 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3441 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3443 * fold-const-call.c (fold_const_call): Handle them.
3444 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3445 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3446 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3447 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3448 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3449 (UNSPEC_XORV): New unspecs.
3450 (optab): Add entries for them.
3451 (BITWISEV): New int iterator.
3452 (bit_reduc_op): New int attributes.
3454 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3455 Alan Hayward <alan.hayward@arm.com>
3456 David Sherwood <david.sherwood@arm.com>
3458 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3459 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3460 * optabs.def (vec_shl_insert_optab): New optab.
3461 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3462 (duplicate_and_interleave): Likewise.
3463 * tree-vect-loop.c: Include internal-fn.h.
3464 (neutral_op_for_slp_reduction): New function, split out from
3465 get_initial_defs_for_reduction.
3466 (get_initial_def_for_reduction): Handle option 2 for variable-length
3467 vectors by loading the neutral value into a vector and then shifting
3468 the initial value into element 0.
3469 (get_initial_defs_for_reduction): Replace the code argument with
3470 the neutral value calculated by neutral_op_for_slp_reduction.
3471 Use gimple_build_vector for constant-length vectors.
3472 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3473 but the first group_size elements have a neutral value.
3474 Use duplicate_and_interleave otherwise.
3475 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3476 Update call to get_initial_defs_for_reduction. Handle SLP
3477 reductions for variable-length vectors by creating one vector
3478 result for each scalar result, with the elements associated
3479 with other scalar results stubbed out with the neutral value.
3480 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3481 Require IFN_VEC_SHL_INSERT for double reductions on
3482 variable-length vectors, or SLP reductions that have
3483 a neutral value. Require can_duplicate_and_interleave_p
3484 support for variable-length unchained SLP reductions if there
3485 is no neutral value, such as for MIN/MAX reductions. Also require
3486 the number of vector elements to be a multiple of the number of
3487 SLP statements when doing variable-length unchained SLP reductions.
3488 Update call to vect_create_epilog_for_reduction.
3489 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3490 and remove initial values.
3491 (duplicate_and_interleave): Make public.
3492 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3493 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3495 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3496 Alan Hayward <alan.hayward@arm.com>
3497 David Sherwood <david.sherwood@arm.com>
3499 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3500 (can_duplicate_and_interleave_p): New function.
3501 (vect_get_and_check_slp_defs): Take the vector of statements
3502 rather than just the current one. Remove excess parentheses.
3503 Restriction rejectinon of vect_constant_def and vect_external_def
3504 for variable-length vectors to boolean types, or types for which
3505 can_duplicate_and_interleave_p is false.
3506 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3507 (duplicate_and_interleave): New function.
3508 (vect_get_constant_vectors): Use gimple_build_vector for
3509 constant-length vectors and suitable variable-length constant
3510 vectors. Use duplicate_and_interleave for other variable-length
3511 vectors. Don't defer the update when inserting new statements.
3513 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3514 Alan Hayward <alan.hayward@arm.com>
3515 David Sherwood <david.sherwood@arm.com>
3517 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3518 min_profitable_iters doesn't go negative.
3520 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3521 Alan Hayward <alan.hayward@arm.com>
3522 David Sherwood <david.sherwood@arm.com>
3524 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3525 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3526 * optabs.def (vec_mask_load_lanes_optab): New optab.
3527 (vec_mask_store_lanes_optab): Likewise.
3528 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3529 (MASK_STORE_LANES): Likewise.
3530 * internal-fn.c (mask_load_lanes_direct): New macro.
3531 (mask_store_lanes_direct): Likewise.
3532 (expand_mask_load_optab_fn): Handle masked operations.
3533 (expand_mask_load_lanes_optab_fn): New macro.
3534 (expand_mask_store_optab_fn): Handle masked operations.
3535 (expand_mask_store_lanes_optab_fn): New macro.
3536 (direct_mask_load_lanes_optab_supported_p): Likewise.
3537 (direct_mask_store_lanes_optab_supported_p): Likewise.
3538 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3540 (vect_load_lanes_supported): Likewise.
3541 * tree-vect-data-refs.c (strip_conversion): New function.
3542 (can_group_stmts_p): Likewise.
3543 (vect_analyze_data_ref_accesses): Use it instead of checking
3544 for a pair of assignments.
3545 (vect_store_lanes_supported): Take a masked_p parameter.
3546 (vect_load_lanes_supported): Likewise.
3547 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3548 vect_store_lanes_supported and vect_load_lanes_supported.
3549 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3550 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3551 parameter. Don't allow gaps for masked accesses.
3552 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3553 and vect_load_lanes_supported.
3554 (get_load_store_type): Take a masked_p parameter and update
3555 call to get_group_load_store_type.
3556 (vectorizable_store): Update call to get_load_store_type.
3557 Handle IFN_MASK_STORE_LANES.
3558 (vectorizable_load): Update call to get_load_store_type.
3559 Handle IFN_MASK_LOAD_LANES.
3561 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3562 Alan Hayward <alan.hayward@arm.com>
3563 David Sherwood <david.sherwood@arm.com>
3565 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3567 * config/aarch64/aarch64-protos.h
3568 (aarch64_sve_struct_memory_operand_p): Declare.
3569 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3570 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3571 (VPRED, vpred): Handle SVE structure modes.
3572 * config/aarch64/constraints.md (Utx): New constraint.
3573 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3574 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3575 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3576 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3577 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3578 structure modes. Split into pieces after RA.
3579 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3580 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3582 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3583 SVE structure modes.
3584 (aarch64_classify_address): Likewise.
3585 (sizetochar): Move earlier in file.
3586 (aarch64_print_operand): Handle SVE register lists.
3587 (aarch64_array_mode): New function.
3588 (aarch64_sve_struct_memory_operand_p): Likewise.
3589 (TARGET_ARRAY_MODE): Redefine.
3591 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3592 Alan Hayward <alan.hayward@arm.com>
3593 David Sherwood <david.sherwood@arm.com>
3595 * target.def (array_mode): New target hook.
3596 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3597 * doc/tm.texi: Regenerate.
3598 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3599 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3600 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3602 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3605 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3606 Alan Hayward <alan.hayward@arm.com>
3607 David Sherwood <david.sherwood@arm.com>
3609 * fold-const.c (fold_binary_loc): Check the argument types
3610 rather than the result type when testing for a vector operation.
3612 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3614 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3615 * doc/tm.texi: Regenerate.
3617 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3618 Alan Hayward <alan.hayward@arm.com>
3619 David Sherwood <david.sherwood@arm.com>
3621 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3622 (sve): Document new AArch64 extension.
3623 * doc/md.texi (w): Extend the description of the AArch64
3624 constraint to include SVE vectors.
3625 (Upl, Upa): Document new AArch64 predicate constraints.
3626 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3628 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3629 (msve-vector-bits=): New option.
3630 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3631 SVE when these are disabled.
3632 (sve): New extension.
3633 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3634 modes. Adjust their number of units based on aarch64_sve_vg.
3635 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3636 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3637 aarch64_addr_query_type.
3638 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3639 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3640 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3641 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3642 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3643 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3644 (aarch64_simd_imm_zero_p): Delete.
3645 (aarch64_check_zero_based_sve_index_immediate): Declare.
3646 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3647 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3648 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3649 (aarch64_sve_float_mul_immediate_p): Likewise.
3650 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3652 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3653 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3654 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3655 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3656 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3657 (aarch64_regmode_natural_size): Likewise.
3658 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3659 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3661 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3662 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3663 for VG and the SVE predicate registers.
3664 (V_ALIASES): Add a "z"-prefixed alias.
3665 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3666 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3667 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3668 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3669 (REG_CLASS_NAMES): Add entries for them.
3670 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3671 and the predicate registers.
3672 (aarch64_sve_vg): Declare.
3673 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3674 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3675 (REGMODE_NATURAL_SIZE): Define.
3676 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3678 * config/aarch64/aarch64.c: Include cfgrtl.h.
3679 (simd_immediate_info): Add a constructor for series vectors,
3680 and an associated step field.
3681 (aarch64_sve_vg): New variable.
3682 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3683 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3684 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3685 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3686 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3687 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3688 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3689 (aarch64_get_mask_mode): New functions.
3690 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3691 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3692 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3693 predicate modes and predicate registers. Explicitly restrict
3694 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3695 to store a vector mode if it is recognized by
3696 aarch64_classify_vector_mode.
3697 (aarch64_regmode_natural_size): New function.
3698 (aarch64_hard_regno_caller_save_mode): Return the original mode
3700 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3701 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3702 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3703 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3705 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3706 does not overlap dest if the function is frame-related. Handle
3708 (aarch64_split_add_offset): New function.
3709 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3710 them aarch64_add_offset.
3711 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3712 and update call to aarch64_sub_sp.
3713 (aarch64_add_cfa_expression): New function.
3714 (aarch64_expand_prologue): Pass extra temporary registers to the
3715 functions above. Handle the case in which we need to emit new
3716 DW_CFA_expressions for registers that were originally saved
3717 relative to the stack pointer, but now have to be expressed
3718 relative to the frame pointer.
3719 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3721 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3722 IP0 and IP1 values for SVE frames.
3723 (aarch64_expand_vec_series): New function.
3724 (aarch64_expand_sve_widened_duplicate): Likewise.
3725 (aarch64_expand_sve_const_vector): Likewise.
3726 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3727 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3728 into the register, rather than emitting a SET directly.
3729 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3730 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3731 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3732 (offset_9bit_signed_scaled_p): New functions.
3733 (aarch64_replicate_bitmask_imm): New function.
3734 (aarch64_bitmask_imm): Use it.
3735 (aarch64_cannot_force_const_mem): Reject expressions involving
3736 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3737 (aarch64_classify_index): Handle SVE indices, by requiring
3738 a plain register index with a scale that matches the element size.
3739 (aarch64_classify_address): Handle SVE addresses. Assert that
3740 the mode of the address is VOIDmode or an integer mode.
3741 Update call to aarch64_classify_symbol.
3742 (aarch64_classify_symbolic_expression): Update call to
3743 aarch64_classify_symbol.
3744 (aarch64_const_vec_all_in_range_p): New function.
3745 (aarch64_print_vector_float_operand): Likewise.
3746 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3747 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3748 and the FP immediates 1.0 and 0.5.
3749 (aarch64_print_address_internal): Handle SVE addresses.
3750 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3751 (aarch64_regno_regclass): Handle predicate registers.
3752 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3754 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3755 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3756 (aarch64_convert_sve_vector_bits): New function.
3757 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3758 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3760 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3761 Handle SVE vector and predicate modes. Accept VL-based constants
3762 that need only one temporary register, and VL offsets that require
3763 no temporary registers.
3764 (aarch64_conditional_register_usage): Mark the predicate registers
3765 as fixed if SVE isn't available.
3766 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3767 Return true for SVE vector and predicate modes.
3768 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3769 rather than an unsigned int. Handle SVE modes.
3770 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3772 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3774 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3775 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3776 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3777 (aarch64_sve_float_mul_immediate_p): New functions.
3778 (aarch64_sve_valid_immediate): New function.
3779 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3780 Explicitly reject structure modes. Check for INDEX constants.
3781 Handle PTRUE and PFALSE constants.
3782 (aarch64_check_zero_based_sve_index_immediate): New function.
3783 (aarch64_simd_imm_zero_p): Delete.
3784 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3785 vector modes. Accept constants in the range of CNT[BHWD].
3786 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3787 ask for an Advanced SIMD mode.
3788 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3789 (aarch64_simd_vector_alignment): Handle SVE predicates.
3790 (aarch64_vectorize_preferred_vector_alignment): New function.
3791 (aarch64_simd_vector_alignment_reachable): Use it instead of
3793 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3794 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3796 (MAX_VECT_LEN): Delete.
3797 (expand_vec_perm_d): Add a vec_flags field.
3798 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3799 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3800 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3802 (aarch64_evpc_rev): Rename to...
3803 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3804 (aarch64_evpc_rev_global): New function.
3805 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3806 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3808 (aarch64_evpc_sve_tbl): New function.
3809 (aarch64_expand_vec_perm_const_1): Update after rename of
3810 aarch64_evpc_rev. Handle SVE permutes too, trying
3811 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3812 than aarch64_evpc_tbl.
3813 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3814 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3815 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3816 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3817 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3818 (aarch64_expand_sve_vcond): New functions.
3819 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3820 of aarch64_vector_mode_p.
3821 (aarch64_dwarf_poly_indeterminate_value): New function.
3822 (aarch64_compute_pressure_classes): Likewise.
3823 (aarch64_can_change_mode_class): Likewise.
3824 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3825 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3826 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3827 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3828 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3829 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3830 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3831 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3833 (Dn, Dl, Dr): Accept const as well as const_vector.
3834 (Dz): Likewise. Compare against CONST0_RTX.
3835 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3836 of "vector" where appropriate.
3837 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3838 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3839 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3840 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3841 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3842 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3843 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3844 (v_int_equiv): Extend to SVE modes.
3845 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3847 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3848 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3849 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3850 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3851 (SVE_COND_FP_CMP): New int iterators.
3852 (perm_hilo): Handle the new unpack unspecs.
3853 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3855 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3856 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3857 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3858 (aarch64_equality_operator, aarch64_constant_vector_operand)
3859 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3860 (aarch64_sve_nonimmediate_operand): Likewise.
3861 (aarch64_sve_general_operand): Likewise.
3862 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3863 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3864 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3865 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3866 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3867 (aarch64_sve_float_arith_immediate): Likewise.
3868 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3869 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3870 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3871 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3872 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3873 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3874 (aarch64_sve_float_arith_operand): Likewise.
3875 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3876 (aarch64_sve_float_mul_operand): Likewise.
3877 (aarch64_sve_vec_perm_operand): Likewise.
3878 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3879 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3880 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3881 as well as const_vector.
3882 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3883 in file. Use CONST0_RTX and CONSTM1_RTX.
3884 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3885 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3886 Use aarch64_simd_imm_zero.
3887 * config/aarch64/aarch64-sve.md: New file.
3888 * config/aarch64/aarch64.md: Include it.
3889 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3890 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3891 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3892 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3893 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3894 (sve): New attribute.
3895 (enabled): Disable instructions with the sve attribute unless
3897 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3898 aarch64_expand_mov_immediate.
3899 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3900 CNT[BHSD] immediates.
3901 (movti): Split CONST_POLY_INT moves into two halves.
3902 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3903 Split additions that need a temporary here if the destination
3904 is the stack pointer.
3905 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3906 (*add<mode>3_poly_1): New instruction.
3907 (set_clobber_cc): New expander.
3909 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3911 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3912 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3913 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3914 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3915 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3916 Change innermode from fixed_mode_size to machine_mode.
3917 (simplify_subreg): Update call accordingly. Handle a constant-sized
3918 subreg of a variable-length CONST_VECTOR.
3920 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3921 Alan Hayward <alan.hayward@arm.com>
3922 David Sherwood <david.sherwood@arm.com>
3924 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3925 (add_offset_to_base): New function, split out from...
3926 (create_mem_ref): ...here. When handling a scale other than 1,
3927 check first whether the address is valid without the offset.
3928 Add it into the base if so, leaving the index and scale as-is.
3930 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3933 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3934 fold_for_warn before checking if arg2 is INTEGER_CST.
3936 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3938 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3939 (store_multiple_operation): Delete.
3940 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3941 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3942 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3943 guarded by TARGET_STRING.
3944 (rs6000_output_load_multiple): Delete.
3945 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3946 OPTION_MASK_STRING / TARGET_STRING handling.
3947 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3948 (const rs6000_opt_masks) <"string">: Change mask to 0.
3949 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3950 (MASK_STRING): Delete.
3951 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3953 (load_multiple): Delete.
3960 (store_multiple): Delete.
3967 (movmemsi_8reg): Delete.
3968 (corresponding unnamed define_insn): Delete.
3969 (movmemsi_6reg): Delete.
3970 (corresponding unnamed define_insn): Delete.
3971 (movmemsi_4reg): Delete.
3972 (corresponding unnamed define_insn): Delete.
3973 (movmemsi_2reg): Delete.
3974 (corresponding unnamed define_insn): Delete.
3975 (movmemsi_1reg): Delete.
3976 (corresponding unnamed define_insn): Delete.
3977 * config/rs6000/rs6000.opt (mno-string): New.
3978 (mstring): Replace by deprecation warning stub.
3979 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3981 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3983 * regrename.c (regrename_do_replace): If replacing the same
3984 reg multiple times, try to reuse last created gen_raw_REG.
3987 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3988 main to workaround a bug in GDB.
3990 2018-01-12 Tom de Vries <tom@codesourcery.com>
3993 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3995 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3997 PR rtl-optimization/80481
3998 * ira-color.c (get_cap_member): New function.
3999 (allocnos_conflict_by_live_ranges_p): Use it.
4000 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4001 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4003 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4006 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4007 (*saddl_se_1): Ditto.
4009 (*ssubl_se_1): Ditto.
4011 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4013 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4014 rather than wi::to_widest for DR_INITs.
4015 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4016 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4017 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4019 (vect_analyze_group_access_1): Note that here.
4021 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4023 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4024 polynomial type sizes.
4026 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4028 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4029 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4030 (gimple_add_tmp_var): Likewise.
4032 2018-01-12 Martin Liska <mliska@suse.cz>
4034 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4035 (gimple_alloc_sizes): Likewise.
4036 (dump_gimple_statistics): Use PRIu64 in printf format.
4037 * gimple.h: Change uint64_t to int.
4039 2018-01-12 Martin Liska <mliska@suse.cz>
4041 * tree-core.h: Use uint64_t instead of int.
4042 * tree.c (tree_node_counts): Likewise.
4043 (tree_node_sizes): Likewise.
4044 (dump_tree_statistics): Use PRIu64 in printf format.
4046 2018-01-12 Martin Liska <mliska@suse.cz>
4048 * Makefile.in: As qsort_chk is implemented in vec.c, add
4049 vec.o to linkage of gencfn-macros.
4050 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4051 passing the info to record_node_allocation_statistics.
4052 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4054 * ggc-common.c (struct ggc_usage): Add operator== and use
4055 it in operator< and compare function.
4056 * mem-stats.h (struct mem_usage): Likewise.
4057 * vec.c (struct vec_usage): Remove operator< and compare
4058 function. Can be simply inherited.
4060 2018-01-12 Martin Jambor <mjambor@suse.cz>
4063 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4064 * tree-ssa-math-opts.c: Include domwalk.h.
4065 (convert_mult_to_fma_1): New function.
4066 (fma_transformation_info): New type.
4067 (fma_deferring_state): Likewise.
4068 (cancel_fma_deferring): New function.
4069 (result_of_phi): Likewise.
4070 (last_fma_candidate_feeds_initial_phi): Likewise.
4071 (convert_mult_to_fma): Added deferring logic, split actual
4072 transformation to convert_mult_to_fma_1.
4073 (math_opts_dom_walker): New type.
4074 (math_opts_dom_walker::after_dom_children): New method, body moved
4075 here from pass_optimize_widening_mul::execute, added deferring logic
4077 (pass_optimize_widening_mul::execute): Moved most of code to
4078 math_opts_dom_walker::after_dom_children.
4079 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4080 * config/i386/i386.c (ix86_option_override_internal): Added
4081 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4083 2018-01-12 Richard Biener <rguenther@suse.de>
4086 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4087 inline instance vars.
4089 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4092 * config/rx/rx.c (rx_is_restricted_memory_address):
4095 2018-01-12 Richard Biener <rguenther@suse.de>
4097 PR tree-optimization/80846
4098 * target.def (split_reduction): New target hook.
4099 * targhooks.c (default_split_reduction): New function.
4100 * targhooks.h (default_split_reduction): Declare.
4101 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4102 target requests first reduce vectors by combining low and high
4104 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4105 (get_vectype_for_scalar_type_and_size): Export.
4106 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4107 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4108 * doc/tm.texi: Regenerate.
4109 * config/i386/i386.c (ix86_split_reduction): Implement
4110 TARGET_VECTORIZE_SPLIT_REDUCTION.
4112 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4115 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4116 in PIC mode except for TARGET_VXWORKS_RTP.
4117 * config/sparc/sparc.c: Include cfgrtl.h.
4118 (TARGET_INIT_PIC_REG): Define.
4119 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4120 (sparc_pic_register_p): New predicate.
4121 (sparc_legitimate_address_p): Use it.
4122 (sparc_legitimize_pic_address): Likewise.
4123 (sparc_delegitimize_address): Likewise.
4124 (sparc_mode_dependent_address_p): Likewise.
4125 (gen_load_pcrel_sym): Remove 4th parameter.
4126 (load_got_register): Adjust call to above. Remove obsolete stuff.
4127 (sparc_expand_prologue): Do not call load_got_register here.
4128 (sparc_flat_expand_prologue): Likewise.
4129 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4130 (sparc_use_pseudo_pic_reg): New function.
4131 (sparc_init_pic_reg): Likewise.
4132 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4133 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4135 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4137 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4138 Add item for branch_cost.
4140 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4142 PR rtl-optimization/83565
4143 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4144 not extend the result to a larger mode for rotate operations.
4145 (num_sign_bit_copies1): Likewise.
4147 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4150 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4152 Use values-Xc.o for -pedantic.
4153 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4155 2018-01-12 Martin Liska <mliska@suse.cz>
4158 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4160 (possible_polymorphic_call_targets): Use it.
4161 (ipa_devirt): Likewise.
4163 2018-01-12 Martin Liska <mliska@suse.cz>
4165 * profile-count.h (enum profile_quality): Use 0 as invalid
4166 enum value of profile_quality.
4168 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4170 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4171 -mext-string options.
4173 2018-01-12 Richard Biener <rguenther@suse.de>
4175 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4176 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4177 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4179 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4181 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4183 * configure.ac (--with-long-double-format): Add support for the
4184 configuration option to change the default long double format on
4186 * config.gcc (powerpc*-linux*-*): Likewise.
4187 * configure: Regenerate.
4188 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4189 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4190 used without modification.
4192 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4194 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4195 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4196 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4197 MISC_BUILTIN_SPEC_BARRIER.
4198 (rs6000_init_builtins): Likewise.
4199 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4201 (speculation_barrier): New define_insn.
4202 * doc/extend.texi: Document __builtin_speculation_barrier.
4204 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4207 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4208 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4209 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4211 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4212 integral modes instead of "ss" and "sd".
4213 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4214 vectors with 32-bit and 64-bit elements.
4215 (vecdupssescalarmodesuffix): New mode attribute.
4216 (vec_dup<mode>): Use it.
4218 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4221 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4222 frame if argument is passed on stack.
4224 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4227 * ree.c (combine_reaching_defs): Optimize also
4228 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4229 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4231 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4234 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4236 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4239 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4240 after they are computed.
4242 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4244 PR tree-optimization/83695
4245 * gimple-loop-linterchange.cc
4246 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4247 reset cached scev information after interchange.
4248 (pass_linterchange::execute): Remove call to scev_reset_htab.
4250 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4252 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4253 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4254 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4255 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4256 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4257 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4258 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4259 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4260 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4261 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4262 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4263 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4264 (V_lane_reg): Likewise.
4265 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4267 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4268 (vfmal_lane_low<mode>_intrinsic,
4269 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4270 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4271 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4272 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4273 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4274 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4276 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4278 * config/arm/arm-cpus.in (fp16fml): New feature.
4279 (ALL_SIMD): Add fp16fml.
4280 (armv8.2-a): Add fp16fml as an option.
4281 (armv8.3-a): Likewise.
4282 (armv8.4-a): Add fp16fml as part of fp16.
4283 * config/arm/arm.h (TARGET_FP16FML): Define.
4284 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4286 * config/arm/arm-modes.def (V2HF): Define.
4287 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4288 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4289 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4290 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4291 vfmsl_low, vfmsl_high): New set of builtins.
4292 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4293 (vfml_op): New code attribute.
4294 (VFMLHALVES): New int iterator.
4295 (VFML, VFMLSEL): New mode attributes.
4296 (V_reg): Define mapping for V2HF.
4297 (V_hi, V_lo): New mode attributes.
4298 (VF_constraint): Likewise.
4299 (vfml_half, vfml_half_selector): New int attributes.
4300 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4302 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4303 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4305 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4306 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4307 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4308 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4310 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4311 Document new effective target and option set.
4313 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4315 * config/arm/arm-cpus.in (armv8_4): New feature.
4316 (ARMv8_4a): New fgroup.
4317 (armv8.4-a): New arch.
4318 * config/arm/arm-tables.opt: Regenerate.
4319 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4320 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4321 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4322 Add matching rules for -march=armv8.4-a and extensions.
4323 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4325 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4328 * config/rx/rx.md (BW): New mode attribute.
4329 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4331 2018-01-11 Richard Biener <rguenther@suse.de>
4333 PR tree-optimization/83435
4334 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4335 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4336 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4338 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4339 Alan Hayward <alan.hayward@arm.com>
4340 David Sherwood <david.sherwood@arm.com>
4342 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4344 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4345 (aarch64_print_address_internal): Use it to check for a zero offset.
4347 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4348 Alan Hayward <alan.hayward@arm.com>
4349 David Sherwood <david.sherwood@arm.com>
4351 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4352 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4353 Return a poly_int64 rather than a HOST_WIDE_INT.
4354 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4355 rather than a HOST_WIDE_INT.
4356 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4357 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4358 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4359 final_offset from HOST_WIDE_INT to poly_int64.
4360 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4361 to_constant when getting the number of units in an Advanced SIMD
4363 (aarch64_builtin_vectorized_function): Check for a constant number
4365 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4367 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4368 attribute instead of GET_MODE_NUNITS.
4369 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4370 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4371 GET_MODE_SIZE for fixed-size registers.
4372 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4373 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4374 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4375 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4376 (aarch64_print_operand, aarch64_print_address_internal)
4377 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4378 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4379 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4380 Handle polynomial GET_MODE_SIZE.
4381 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4382 wider than SImode without modification.
4383 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4384 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4385 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4386 passing and returning SVE modes.
4387 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4388 rather than GEN_INT.
4389 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4390 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4391 (aarch64_allocate_and_probe_stack_space): Likewise.
4392 (aarch64_layout_frame): Cope with polynomial offsets.
4393 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4394 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4396 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4397 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4398 poly_int64 rather than a HOST_WIDE_INT.
4399 (aarch64_get_separate_components, aarch64_process_components)
4400 (aarch64_expand_prologue, aarch64_expand_epilogue)
4401 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4402 (aarch64_anchor_offset): New function, split out from...
4403 (aarch64_legitimize_address): ...here.
4404 (aarch64_builtin_vectorization_cost): Handle polynomial
4405 TYPE_VECTOR_SUBPARTS.
4406 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4408 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4409 number of elements from the PARALLEL rather than the mode.
4410 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4411 rather than GET_MODE_BITSIZE.
4412 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4413 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4414 (aarch64_expand_vec_perm_const_1): Handle polynomial
4415 d->perm.length () and d->perm elements.
4416 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4417 Apply to_constant to d->perm elements.
4418 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4419 polynomial CONST_VECTOR_NUNITS.
4420 (aarch64_move_pointer): Take amount as a poly_int64 rather
4422 (aarch64_progress_pointer): Avoid temporary variable.
4423 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4424 the mode attribute instead of GET_MODE.
4426 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4427 Alan Hayward <alan.hayward@arm.com>
4428 David Sherwood <david.sherwood@arm.com>
4430 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4431 x exists before using it.
4432 (aarch64_add_constant_internal): Rename to...
4433 (aarch64_add_offset_1): ...this. Replace regnum with separate
4434 src and dest rtxes. Handle the case in which they're different,
4435 including when the offset is zero. Replace scratchreg with an rtx.
4436 Use 2 additions if there is no spare register into which we can
4437 move a 16-bit constant.
4438 (aarch64_add_constant): Delete.
4439 (aarch64_add_offset): Replace reg with separate src and dest
4440 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4441 Use aarch64_add_offset_1.
4442 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4443 an rtx rather than an int. Take the delta as a poly_int64
4444 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4445 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4446 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4447 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4448 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4450 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4451 aarch64_add_constant.
4453 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4455 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4456 Use scalar_float_mode.
4458 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4460 * config/aarch64/aarch64-simd.md
4461 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4462 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4463 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4464 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4465 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4466 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4467 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4468 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4469 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4470 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4472 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4475 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4476 targ_options->x_arm_arch_string is non NULL.
4478 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4480 * config/aarch64/aarch64.h
4481 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4483 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4486 * expmed.c (emit_store_flag_force): Swap if const op0
4487 and change VOIDmode to mode of op0.
4489 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4491 PR rtl-optimization/83761
4492 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4493 than bytes to mode_for_size.
4495 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4498 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4499 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4502 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4505 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4506 when in layout mode.
4507 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4508 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4511 2018-01-10 Michael Collison <michael.collison@arm.com>
4513 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4514 * config/aarch64/aarch64-option-extension.def: Add
4515 AARCH64_OPT_EXTENSION of 'fp16fml'.
4516 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4517 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4518 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4519 * config/aarch64/constraints.md (Ui7): New constraint.
4520 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4521 (VFMLA_SEL_W): Ditto.
4524 (VFMLA16_LOW): New int iterator.
4525 (VFMLA16_HIGH): Ditto.
4526 (UNSPEC_FMLAL): New unspec.
4527 (UNSPEC_FMLSL): Ditto.
4528 (UNSPEC_FMLAL2): Ditto.
4529 (UNSPEC_FMLSL2): Ditto.
4530 (f16mac): New code attribute.
4531 * config/aarch64/aarch64-simd-builtins.def
4532 (aarch64_fmlal_lowv2sf): Ditto.
4533 (aarch64_fmlsl_lowv2sf): Ditto.
4534 (aarch64_fmlalq_lowv4sf): Ditto.
4535 (aarch64_fmlslq_lowv4sf): Ditto.
4536 (aarch64_fmlal_highv2sf): Ditto.
4537 (aarch64_fmlsl_highv2sf): Ditto.
4538 (aarch64_fmlalq_highv4sf): Ditto.
4539 (aarch64_fmlslq_highv4sf): Ditto.
4540 (aarch64_fmlal_lane_lowv2sf): Ditto.
4541 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4542 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4543 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4544 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4545 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4546 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4547 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4548 (aarch64_fmlal_lane_highv2sf): Ditto.
4549 (aarch64_fmlsl_lane_highv2sf): Ditto.
4550 (aarch64_fmlal_laneq_highv2sf): Ditto.
4551 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4552 (aarch64_fmlalq_lane_highv4sf): Ditto.
4553 (aarch64_fmlsl_lane_highv4sf): Ditto.
4554 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4555 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4556 * config/aarch64/aarch64-simd.md:
4557 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4558 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4559 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4560 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4561 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4562 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4563 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4564 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4565 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4566 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4567 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4568 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4569 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4570 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4571 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4572 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4573 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4574 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4575 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4576 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4577 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4578 (vfmlsl_low_u32): Ditto.
4579 (vfmlalq_low_u32): Ditto.
4580 (vfmlslq_low_u32): Ditto.
4581 (vfmlal_high_u32): Ditto.
4582 (vfmlsl_high_u32): Ditto.
4583 (vfmlalq_high_u32): Ditto.
4584 (vfmlslq_high_u32): Ditto.
4585 (vfmlal_lane_low_u32): Ditto.
4586 (vfmlsl_lane_low_u32): Ditto.
4587 (vfmlal_laneq_low_u32): Ditto.
4588 (vfmlsl_laneq_low_u32): Ditto.
4589 (vfmlalq_lane_low_u32): Ditto.
4590 (vfmlslq_lane_low_u32): Ditto.
4591 (vfmlalq_laneq_low_u32): Ditto.
4592 (vfmlslq_laneq_low_u32): Ditto.
4593 (vfmlal_lane_high_u32): Ditto.
4594 (vfmlsl_lane_high_u32): Ditto.
4595 (vfmlal_laneq_high_u32): Ditto.
4596 (vfmlsl_laneq_high_u32): Ditto.
4597 (vfmlalq_lane_high_u32): Ditto.
4598 (vfmlslq_lane_high_u32): Ditto.
4599 (vfmlalq_laneq_high_u32): Ditto.
4600 (vfmlslq_laneq_high_u32): Ditto.
4601 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4602 (AARCH64_FL_FOR_ARCH8_4): New.
4603 (AARCH64_ISA_F16FML): New ISA flag.
4604 (TARGET_F16FML): New feature flag for fp16fml.
4605 (doc/invoke.texi): Document new fp16fml option.
4607 2018-01-10 Michael Collison <michael.collison@arm.com>
4609 * config/aarch64/aarch64-builtins.c:
4610 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4611 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4612 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4613 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4614 (AARCH64_ISA_SHA3): New ISA flag.
4615 (TARGET_SHA3): New feature flag for sha3.
4616 * config/aarch64/iterators.md (sha512_op): New int attribute.
4617 (CRYPTO_SHA512): New int iterator.
4618 (UNSPEC_SHA512H): New unspec.
4619 (UNSPEC_SHA512H2): Ditto.
4620 (UNSPEC_SHA512SU0): Ditto.
4621 (UNSPEC_SHA512SU1): Ditto.
4622 * config/aarch64/aarch64-simd-builtins.def
4623 (aarch64_crypto_sha512hqv2di): New builtin.
4624 (aarch64_crypto_sha512h2qv2di): Ditto.
4625 (aarch64_crypto_sha512su0qv2di): Ditto.
4626 (aarch64_crypto_sha512su1qv2di): Ditto.
4627 (aarch64_eor3qv8hi): Ditto.
4628 (aarch64_rax1qv2di): Ditto.
4629 (aarch64_xarqv2di): Ditto.
4630 (aarch64_bcaxqv8hi): Ditto.
4631 * config/aarch64/aarch64-simd.md:
4632 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4633 (aarch64_crypto_sha512su0qv2di): Ditto.
4634 (aarch64_crypto_sha512su1qv2di): Ditto.
4635 (aarch64_eor3qv8hi): Ditto.
4636 (aarch64_rax1qv2di): Ditto.
4637 (aarch64_xarqv2di): Ditto.
4638 (aarch64_bcaxqv8hi): Ditto.
4639 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4640 (vsha512h2q_u64): Ditto.
4641 (vsha512su0q_u64): Ditto.
4642 (vsha512su1q_u64): Ditto.
4643 (veor3q_u16): Ditto.
4644 (vrax1q_u64): Ditto.
4646 (vbcaxq_u16): Ditto.
4647 * config/arm/types.md (crypto_sha512): New type attribute.
4648 (crypto_sha3): Ditto.
4649 (doc/invoke.texi): Document new sha3 option.
4651 2018-01-10 Michael Collison <michael.collison@arm.com>
4653 * config/aarch64/aarch64-builtins.c:
4654 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4655 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4656 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4657 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4658 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4659 (AARCH64_ISA_SM4): New ISA flag.
4660 (TARGET_SM4): New feature flag for sm4.
4661 * config/aarch64/aarch64-simd-builtins.def
4662 (aarch64_sm3ss1qv4si): Ditto.
4663 (aarch64_sm3tt1aq4si): Ditto.
4664 (aarch64_sm3tt1bq4si): Ditto.
4665 (aarch64_sm3tt2aq4si): Ditto.
4666 (aarch64_sm3tt2bq4si): Ditto.
4667 (aarch64_sm3partw1qv4si): Ditto.
4668 (aarch64_sm3partw2qv4si): Ditto.
4669 (aarch64_sm4eqv4si): Ditto.
4670 (aarch64_sm4ekeyqv4si): Ditto.
4671 * config/aarch64/aarch64-simd.md:
4672 (aarch64_sm3ss1qv4si): Ditto.
4673 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4674 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4675 (aarch64_sm4eqv4si): Ditto.
4676 (aarch64_sm4ekeyqv4si): Ditto.
4677 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4678 (sm3part_op): Ditto.
4679 (CRYPTO_SM3TT): Ditto.
4680 (CRYPTO_SM3PART): Ditto.
4681 (UNSPEC_SM3SS1): New unspec.
4682 (UNSPEC_SM3TT1A): Ditto.
4683 (UNSPEC_SM3TT1B): Ditto.
4684 (UNSPEC_SM3TT2A): Ditto.
4685 (UNSPEC_SM3TT2B): Ditto.
4686 (UNSPEC_SM3PARTW1): Ditto.
4687 (UNSPEC_SM3PARTW2): Ditto.
4688 (UNSPEC_SM4E): Ditto.
4689 (UNSPEC_SM4EKEY): Ditto.
4690 * config/aarch64/constraints.md (Ui2): New constraint.
4691 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4692 * config/arm/types.md (crypto_sm3): New type attribute.
4693 (crypto_sm4): Ditto.
4694 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4695 (vsm3tt1aq_u32): Ditto.
4696 (vsm3tt1bq_u32): Ditto.
4697 (vsm3tt2aq_u32): Ditto.
4698 (vsm3tt2bq_u32): Ditto.
4699 (vsm3partw1q_u32): Ditto.
4700 (vsm3partw2q_u32): Ditto.
4701 (vsm4eq_u32): Ditto.
4702 (vsm4ekeyq_u32): Ditto.
4703 (doc/invoke.texi): Document new sm4 option.
4705 2018-01-10 Michael Collison <michael.collison@arm.com>
4707 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4708 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4709 (AARCH64_FL_FOR_ARCH8_4): New.
4710 (AARCH64_FL_V8_4): New flag.
4711 (doc/invoke.texi): Document new armv8.4-a option.
4713 2018-01-10 Michael Collison <michael.collison@arm.com>
4715 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4716 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4717 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4718 * config/aarch64/aarch64-option-extension.def: Add
4719 AARCH64_OPT_EXTENSION of 'sha2'.
4720 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4721 (crypto): Disable sha2 and aes if crypto disabled.
4722 (crypto): Enable aes and sha2 if enabled.
4723 (simd): Disable sha2 and aes if simd disabled.
4724 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4726 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4727 (TARGET_SHA2): New feature flag for sha2.
4728 (TARGET_AES): New feature flag for aes.
4729 * config/aarch64/aarch64-simd.md:
4730 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4731 conditional on TARGET_AES.
4732 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4733 (aarch64_crypto_sha1hsi): Make pattern conditional
4735 (aarch64_crypto_sha1hv4si): Ditto.
4736 (aarch64_be_crypto_sha1hv4si): Ditto.
4737 (aarch64_crypto_sha1su1v4si): Ditto.
4738 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4739 (aarch64_crypto_sha1su0v4si): Ditto.
4740 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4741 (aarch64_crypto_sha256su0v4si): Ditto.
4742 (aarch64_crypto_sha256su1v4si): Ditto.
4743 (doc/invoke.texi): Document new aes and sha2 options.
4745 2018-01-10 Martin Sebor <msebor@redhat.com>
4747 PR tree-optimization/83781
4748 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4751 2018-01-11 Martin Sebor <msebor@gmail.com>
4752 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4754 PR tree-optimization/83501
4755 PR tree-optimization/81703
4757 * tree-ssa-strlen.c (get_string_cst): Rename...
4758 (get_string_len): ...to this. Handle global constants.
4759 (handle_char_store): Adjust.
4761 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4762 Jim Wilson <jimw@sifive.com>
4764 * config/riscv/riscv-protos.h (riscv_output_return): New.
4765 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4766 (riscv_attribute_table, riscv_output_return),
4767 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4768 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4769 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4771 (riscv_expand_prologue): Add early return for naked function.
4772 (riscv_expand_epilogue): Likewise.
4773 (riscv_function_ok_for_sibcall): Return false for naked function.
4774 (riscv_set_current_function): New.
4775 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4776 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4777 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4778 * doc/extend.texi (RISC-V Function Attributes): New.
4780 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4782 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4783 check for 128-bit long double before checking TCmode.
4784 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4785 128-bit long doubles before checking TFmode or TCmode.
4786 (FLOAT128_IBM_P): Likewise.
4788 2018-01-10 Martin Sebor <msebor@redhat.com>
4790 PR tree-optimization/83671
4791 * builtins.c (c_strlen): Unconditionally return zero for the empty
4793 Use -Warray-bounds for warnings.
4794 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4795 for non-constant array indices with COMPONENT_REF, arrays of
4796 arrays, and pointers to arrays.
4797 (gimple_fold_builtin_strlen): Determine and set length range for
4798 non-constant character arrays.
4800 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4803 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4806 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4808 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4810 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4813 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4814 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4815 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4816 indexed_or_indirect_operand predicate.
4817 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4818 (*vsx_le_perm_load_v8hi): Likewise.
4819 (*vsx_le_perm_load_v16qi): Likewise.
4820 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4821 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4822 (*vsx_le_perm_store_v8hi): Likewise.
4823 (*vsx_le_perm_store_v16qi): Likewise.
4824 (eight unnamed splitters): Likewise.
4826 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4828 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4829 * config/rs6000/emmintrin.h: Likewise.
4830 * config/rs6000/mmintrin.h: Likewise.
4831 * config/rs6000/xmmintrin.h: Likewise.
4833 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4836 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4838 * tree.c (tree_nop_conversion): Return true for location wrapper
4840 (maybe_wrap_with_location): New function.
4841 (selftest::check_strip_nops): New function.
4842 (selftest::test_location_wrappers): New function.
4843 (selftest::tree_c_tests): Call it.
4844 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4845 (maybe_wrap_with_location): New decl.
4846 (EXPR_LOCATION_WRAPPER_P): New macro.
4847 (location_wrapper_p): New inline function.
4848 (tree_strip_any_location_wrapper): New inline function.
4850 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4853 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4854 stack_realign_offset for the largest alignment of stack slot
4856 (ix86_find_max_used_stack_alignment): New function.
4857 (ix86_finalize_stack_frame_flags): Use it. Set
4858 max_used_stack_alignment if we don't realign stack.
4859 * config/i386/i386.h (machine_function): Add
4860 max_used_stack_alignment.
4862 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4864 * config/arm/arm.opt (-mbranch-cost): New option.
4865 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4868 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4871 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4872 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4874 2018-01-10 Richard Biener <rguenther@suse.de>
4877 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4878 early out so it also covers the case where we have a non-NULL
4881 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4883 PR tree-optimization/83753
4884 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4885 for non-strided grouped accesses if the number of elements is 1.
4887 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4890 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4891 * i386.h (TARGET_USE_GATHER): Define.
4892 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4894 2018-01-10 Martin Liska <mliska@suse.cz>
4897 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4898 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4900 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4901 CLEANUP_NO_PARTITIONING is not set.
4903 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4905 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4906 for vectors, as a partial revert of r254296.
4907 * rtl.h (const_vec_p): Delete.
4908 (const_vec_duplicate_p): Don't test for vector CONSTs.
4909 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4910 * expmed.c (make_tree): Likewise.
4913 * common.md (E, F): Use CONSTANT_P instead of checking for
4915 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4916 checking for CONST_VECTOR.
4918 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4921 * predict.c (force_edge_cold): Handle in more sane way edges
4924 2018-01-09 Carl Love <cel@us.ibm.com>
4926 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4928 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4929 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4930 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4931 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4932 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4933 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4934 * config/rs6000/rs6000-protos.h: Add extern defition for
4935 rs6000_generate_float2_double_code.
4936 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4938 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4939 (float2_v2df): Add define_expand.
4941 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4944 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4945 op_mode in the force_to_mode call.
4947 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4949 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4950 instead of checking each element individually.
4951 (aarch64_evpc_uzp): Likewise.
4952 (aarch64_evpc_zip): Likewise.
4953 (aarch64_evpc_ext): Likewise.
4954 (aarch64_evpc_rev): Likewise.
4955 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4956 instead of checking each element individually. Return true without
4958 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4959 whether all selected elements come from the same input, instead of
4960 checking each element individually. Remove calls to gen_rtx_REG,
4961 start_sequence and end_sequence and instead assert that no rtl is
4964 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4966 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4967 order of HIGH and CONST checks.
4969 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4971 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4972 if the destination isn't an SSA_NAME.
4974 2018-01-09 Richard Biener <rguenther@suse.de>
4976 PR tree-optimization/83668
4977 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4979 (canonicalize_loop_form): ... here, renamed from ...
4980 (canonicalize_loop_closed_ssa_form): ... this and amended to
4981 swap successor edges for loop exit blocks to make us use
4982 the RPO order we need for initial schedule generation.
4984 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4986 PR tree-optimization/64811
4987 * match.pd: When optimizing comparisons with Inf, avoid
4988 introducing or losing exceptions from comparisons with NaN.
4990 2018-01-09 Martin Liska <mliska@suse.cz>
4993 * asan.c (shadow_mem_size): Add gcc_assert.
4995 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4997 Don't save registers in main().
5000 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5001 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5002 * config/avr/avr.c (avr_set_current_function): Don't error if
5003 naked, OS_task or OS_main are specified at the same time.
5004 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5006 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5008 * common/config/avr/avr-common.c (avr_option_optimization_table):
5009 Switch on -mmain-is-OS_task for optimizing compilations.
5011 2018-01-09 Richard Biener <rguenther@suse.de>
5013 PR tree-optimization/83572
5014 * graphite.c: Include cfganal.h.
5015 (graphite_transform_loops): Connect infinite loops to exit
5016 and remove fake edges at the end.
5018 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5020 * ipa-inline.c (edge_badness): Revert accidental checkin.
5022 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5025 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5026 symbols; not inline clones.
5028 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5031 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5032 hard registers. Formatting fixes.
5034 PR preprocessor/83722
5035 * gcc.c (try_generate_repro): Pass
5036 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5037 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5040 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5041 Kito Cheng <kito.cheng@gmail.com>
5043 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5044 (riscv_leaf_function_p): Delete.
5045 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5047 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5049 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5051 (do_ifelse): New function.
5052 (do_isel): New function.
5053 (do_sub3): New function.
5054 (do_add3): New function.
5055 (do_load_mask_compare): New function.
5056 (do_overlap_load_compare): New function.
5057 (expand_compare_loop): New function.
5058 (expand_block_compare): Call expand_compare_loop() when appropriate.
5059 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5061 (-mblock-compare-inline-loop-limit): New option.
5063 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5066 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5067 Reverse order of second and third operands in first alternative.
5068 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5069 of first and second elements in UNSPEC_VPERMR vector.
5070 (altivec_expand_vec_perm_le): Likewise.
5072 2018-01-08 Jeff Law <law@redhat.com>
5074 PR rtl-optimizatin/81308
5075 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5076 (process_switch): If group_case_labels makes a change, then set
5078 (pass_convert_switch::execute): If a switch is converted, then
5079 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5081 PR rtl-optimization/81308
5082 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5085 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5087 PR target/83663 - Revert r255946
5088 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5089 generation for cases where splatting a value is not useful.
5090 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5091 across a vec_duplicate and a paradoxical subreg forming a vector
5092 mode to a vec_concat.
5094 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5096 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5097 -march=armv8.3-a variants.
5098 * config/arm/t-multilib: Likewise.
5099 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5101 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5103 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5105 (cceq_ior_compare_complement): Give it a name so I can use it, and
5106 change boolean_or_operator predicate to boolean_operator so it can
5107 be used to generate a crand.
5108 (eqne): New code iterator.
5109 (bd/bd_neg): New code_attrs.
5110 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5111 a single define_insn.
5112 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5113 decrement (bdnzt/bdnzf/bdzt/bdzf).
5114 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5115 with the new names of the branch decrement patterns, and added the
5116 names of the branch decrement conditional patterns.
5118 2018-01-08 Richard Biener <rguenther@suse.de>
5120 PR tree-optimization/83563
5121 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5124 2018-01-08 Richard Biener <rguenther@suse.de>
5127 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5129 2018-01-08 Richard Biener <rguenther@suse.de>
5131 PR tree-optimization/83685
5132 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5133 references to abnormals.
5135 2018-01-08 Richard Biener <rguenther@suse.de>
5138 * dwarf2out.c (output_indirect_strings): Handle empty
5139 skeleton_debug_str_hash.
5140 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5142 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5144 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5145 (emit_store_direct): Likewise.
5146 (arc_trampoline_adjust_address): Likewise.
5147 (arc_asm_trampoline_template): New function.
5148 (arc_initialize_trampoline): Use asm_trampoline_template.
5149 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5150 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5151 * config/arc/arc.md (flush_icache): Delete pattern.
5153 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5155 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5156 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5159 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5162 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5163 by not USED_FOR_TARGET.
5164 (make_pass_resolve_sw_modes): Likewise.
5166 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5168 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5171 2018-01-08 Richard Biener <rguenther@suse.de>
5174 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5176 2018-01-08 Richard Biener <rguenther@suse.de>
5179 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5181 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5184 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5185 basic blocks with a small number of successors.
5186 (convert_control_dep_chain_into_preds): Improve handling of
5188 (dump_predicates): Split apart into...
5189 (dump_pred_chain): ...here...
5190 (dump_pred_info): ...and here.
5191 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5192 (can_chain_union_be_invalidated_p): Improve check for invalidation
5194 (uninit_uses_cannot_happen): Avoid unnecessary if
5195 convert_control_dep_chain_into_preds yielded nothing.
5197 2018-01-06 Martin Sebor <msebor@redhat.com>
5199 PR tree-optimization/83640
5200 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5201 subtracting negative offset from size.
5202 (builtin_access::overlap): Adjust offset bounds of the access to fall
5203 within the size of the object if possible.
5205 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5207 PR rtl-optimization/83699
5208 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5209 extract_bit_field_as_subreg to cases in which the extracted
5210 value is also a vector.
5212 * lra-constraints.c (process_alt_operands): Test for the equivalence
5213 substitutions when detecting a possible reload cycle.
5215 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5218 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5219 by default if flag_selective_schedling{,2}. Formatting fixes.
5221 PR rtl-optimization/83682
5222 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5223 if it has non-VECTOR_MODE element mode.
5224 (vec_duplicate_p): Likewise.
5227 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5228 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5230 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5233 * config/i386/i386-builtin.def
5234 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5235 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5236 Require also OPTION_MASK_ISA_AVX512F in addition to
5237 OPTION_MASK_ISA_GFNI.
5238 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5239 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5240 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5241 to OPTION_MASK_ISA_GFNI.
5242 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5243 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5244 OPTION_MASK_ISA_AVX512BW.
5245 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5246 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5247 addition to OPTION_MASK_ISA_GFNI.
5248 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5249 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5250 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5251 to OPTION_MASK_ISA_GFNI.
5252 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5253 a requirement for all ISAs rather than any of them with a few
5255 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5257 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5258 bitmasks to be enabled with 3 exceptions, instead of requiring any
5259 enabled ISA with lots of exceptions.
5260 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5261 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5262 Change avx512bw in isa attribute to avx512f.
5263 * config/i386/sgxintrin.h: Add license boilerplate.
5264 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5265 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5266 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5267 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5269 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5270 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5271 temporarily sse2 rather than sse if not enabled already.
5274 * config/i386/sse.md (VI248_VLBW): Rename to ...
5275 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5276 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5277 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5278 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5279 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5280 mode iterator instead of VI248_VLBW.
5282 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5284 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5285 (record_modified): Skip clobbers; add debug output.
5286 (param_change_prob): Use sreal frequencies.
5288 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5290 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5291 punt for user-aligned variables.
5293 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5295 * tree-chrec.c (chrec_contains_symbols): Return true for
5298 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5301 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5302 of (x|y) == x for BICS pattern.
5304 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5306 PR tree-optimization/83605
5307 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5308 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5311 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5313 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5314 * config/epiphany/rtems.h: New file.
5316 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5317 Uros Bizjak <ubizjak@gmail.com>
5320 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5321 QIreg_operand instead of register_operand predicate.
5322 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5323 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5324 comments instead of -fmitigate[-_]rop.
5326 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5329 * cgraphunit.c (symbol_table::compile): Switch to text_section
5330 before calling assembly_start debug hook.
5331 * run-rtl-passes.c (run_rtl_passes): Likewise.
5334 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5336 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5337 range_int_cst_p rather than !symbolic_range_p before calling
5338 extract_range_from_multiplicative_op_1.
5340 2018-01-04 Jeff Law <law@redhat.com>
5342 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5343 redundant test in assertion.
5345 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5347 * doc/rtl.texi: Document machine_mode wrapper classes.
5349 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5351 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5354 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5356 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5357 the VEC_PERM_EXPR fold to fail.
5359 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5362 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5363 to switched_sections.
5365 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5368 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5371 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5374 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5375 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5377 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5380 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5381 is BLKmode and bitpos not zero or mode change is needed.
5383 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5386 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5389 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5392 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5393 instead of MULT rtx. Update all corresponding splitters.
5395 (*ssub<modesuffix>): Ditto.
5397 (*cmp_sadd_di): Update split patterns.
5398 (*cmp_sadd_si): Ditto.
5399 (*cmp_sadd_sidi): Ditto.
5400 (*cmp_ssub_di): Ditto.
5401 (*cmp_ssub_si): Ditto.
5402 (*cmp_ssub_sidi): Ditto.
5403 * config/alpha/predicates.md (const23_operand): New predicate.
5404 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5405 Look for ASHIFT, not MULT inner operand.
5406 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5408 2018-01-04 Martin Liska <mliska@suse.cz>
5410 PR gcov-profile/83669
5411 * gcov.c (output_intermediate_file): Add version to intermediate
5413 * doc/gcov.texi: Document new field 'version' in intermediate
5414 file format. Fix location of '-k' option of gcov command.
5416 2018-01-04 Martin Liska <mliska@suse.cz>
5419 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5421 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5423 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5425 2018-01-03 Martin Sebor <msebor@redhat.com>
5427 PR tree-optimization/83655
5428 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5429 checking calls with invalid arguments.
5431 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5433 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5434 (vectorizable_mask_load_store): Delete.
5435 (vectorizable_call): Return false for masked loads and stores.
5436 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5437 instead of gimple_assign_rhs1.
5438 (vectorizable_load): Handle IFN_MASK_LOAD.
5439 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5441 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5443 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5445 (vectorizable_mask_load_store): ...here.
5446 (vectorizable_load): ...and here.
5448 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5450 * tree-vect-stmts.c (vect_build_all_ones_mask)
5451 (vect_build_zero_merge_argument): New functions, split out from...
5452 (vectorizable_load): ...here.
5454 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5456 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5458 (vectorizable_mask_load_store): ...here.
5459 (vectorizable_store): ...and here.
5461 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5463 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5465 (vectorizable_mask_load_store): ...here.
5467 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5469 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5470 (vect_model_store_cost): Take a vec_load_store_type instead of a
5472 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5473 (vect_model_store_cost): Take a vec_load_store_type instead of a
5475 (vectorizable_mask_load_store): Update accordingly.
5476 (vectorizable_store): Likewise.
5477 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5479 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5481 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5482 IFN_MASK_LOAD calls here rather than...
5483 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5485 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5486 Alan Hayward <alan.hayward@arm.com>
5487 David Sherwood <david.sherwood@arm.com>
5489 * expmed.c (extract_bit_field_1): For vector extracts,
5490 fall back to extract_bit_field_as_subreg if vec_extract
5493 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5494 Alan Hayward <alan.hayward@arm.com>
5495 David Sherwood <david.sherwood@arm.com>
5497 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5498 they are variable or constant sized.
5499 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5500 slots for constant-sized data.
5502 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5503 Alan Hayward <alan.hayward@arm.com>
5504 David Sherwood <david.sherwood@arm.com>
5506 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5507 handling COND_EXPRs with boolean comparisons, try to find a better
5508 basis for the mask type than the boolean itself.
5510 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5512 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5513 is calculated and how it can be overridden.
5514 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5515 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5517 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5520 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5521 Alan Hayward <alan.hayward@arm.com>
5522 David Sherwood <david.sherwood@arm.com>
5524 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5525 Remove the mode argument.
5526 (aarch64_simd_valid_immediate): Remove the mode and inverse
5528 * config/aarch64/iterators.md (bitsize): New iterator.
5529 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5530 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5531 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5532 aarch64_simd_valid_immediate.
5533 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5534 (aarch64_reg_or_bic_imm): Likewise.
5535 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5536 with an insn_type enum and msl with a modifier_type enum.
5537 Replace element_width with a scalar_mode. Change the shift
5538 to unsigned int. Add constructors for scalar_float_mode and
5539 scalar_int_mode elements.
5540 (aarch64_vect_float_const_representable_p): Delete.
5541 (aarch64_can_const_movi_rtx_p)
5542 (aarch64_simd_scalar_immediate_valid_for_move)
5543 (aarch64_simd_make_constant): Update call to
5544 aarch64_simd_valid_immediate.
5545 (aarch64_advsimd_valid_immediate_hs): New function.
5546 (aarch64_advsimd_valid_immediate): Likewise.
5547 (aarch64_simd_valid_immediate): Remove mode and inverse
5548 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5549 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5550 and aarch64_float_const_representable_p on the result.
5551 (aarch64_output_simd_mov_immediate): Remove mode argument.
5552 Update call to aarch64_simd_valid_immediate and use of
5553 simd_immediate_info.
5554 (aarch64_output_scalar_simd_mov_immediate): Update call
5557 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5558 Alan Hayward <alan.hayward@arm.com>
5559 David Sherwood <david.sherwood@arm.com>
5561 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5562 (mode_nunits): Likewise CONST_MODE_NUNITS.
5563 * machmode.def (ADJUST_NUNITS): Document.
5564 * genmodes.c (mode_data::need_nunits_adj): New field.
5565 (blank_mode): Update accordingly.
5566 (adj_nunits): New variable.
5567 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5569 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5570 listed in adj_nunits.
5571 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5572 listed in adj_nunits. Don't emit case statements for such modes.
5573 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5574 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5575 nothing if adj_nunits is nonnull.
5576 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5577 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5578 (emit_mode_fbit): Update use of print_maybe_const_decl.
5579 (emit_move_size): Likewise. Treat the array as non-const
5581 (emit_mode_adjustments): Handle adj_nunits.
5583 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5585 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5586 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5587 (VECTOR_MODES): Use it.
5588 (make_vector_modes): Take the prefix as an argument.
5590 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5591 Alan Hayward <alan.hayward@arm.com>
5592 David Sherwood <david.sherwood@arm.com>
5594 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5595 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5596 for MODE_VECTOR_BOOL.
5597 * machmode.def (VECTOR_BOOL_MODE): Document.
5598 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5599 (make_vector_bool_mode): New function.
5600 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5602 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5603 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5605 * stor-layout.c (int_mode_for_mode): Likewise.
5606 * tree.c (build_vector_type_for_mode): Likewise.
5607 * varasm.c (output_constant_pool_2): Likewise.
5608 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5609 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5610 for MODE_VECTOR_BOOL.
5611 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5612 of mode class checks.
5613 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5614 instead of a list of mode class checks.
5615 (expand_vector_scalar_condition): Likewise.
5616 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5618 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5619 Alan Hayward <alan.hayward@arm.com>
5620 David Sherwood <david.sherwood@arm.com>
5622 * machmode.h (mode_size): Change from unsigned short to
5624 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5625 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5626 or if measurement_type is not polynomial.
5627 (fixed_size_mode::includes_p): Check for constant-sized modes.
5628 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5629 return a poly_uint16 rather than an unsigned short.
5630 (emit_mode_size): Change the type of mode_size from unsigned short
5631 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5632 (emit_mode_adjustments): Cope with polynomial vector sizes.
5633 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5635 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5637 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5638 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5639 * caller-save.c (setup_save_areas): Likewise.
5640 (replace_reg_with_saved_mem): Likewise.
5641 * calls.c (emit_library_call_value_1): Likewise.
5642 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5643 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5644 (gen_lowpart_for_combine): Likewise.
5645 * convert.c (convert_to_integer_1): Likewise.
5646 * cse.c (equiv_constant, cse_insn): Likewise.
5647 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5648 (cselib_subst_to_values): Likewise.
5649 * dce.c (word_dce_process_block): Likewise.
5650 * df-problems.c (df_word_lr_mark_ref): Likewise.
5651 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5652 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5653 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5654 (rtl_for_decl_location): Likewise.
5655 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5656 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5657 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5658 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5659 (expand_expr_real_1): Likewise.
5660 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5661 (pad_below): Likewise.
5662 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5663 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5664 * ira.c (get_subreg_tracking_sizes): Likewise.
5665 * ira-build.c (ira_create_allocno_objects): Likewise.
5666 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5667 (ira_sort_regnos_for_alter_reg): Likewise.
5668 * ira-costs.c (record_operand_costs): Likewise.
5669 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5670 (resolve_simple_move): Likewise.
5671 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5672 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5673 (lra_constraints): Likewise.
5674 (CONST_POOL_OK_P): Reject variable-sized modes.
5675 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5676 (add_pseudo_to_slot, lra_spill): Likewise.
5677 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5678 * optabs-query.c (get_best_extraction_insn): Likewise.
5679 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5680 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5681 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5682 * recog.c (offsettable_address_addr_space_p): Likewise.
5683 * regcprop.c (maybe_mode_change): Likewise.
5684 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5685 * regrename.c (build_def_use): Likewise.
5686 * regstat.c (dump_reg_info): Likewise.
5687 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5688 (find_reloads, find_reloads_subreg_address): Likewise.
5689 * reload1.c (eliminate_regs_1): Likewise.
5690 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5691 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5692 (simplify_binary_operation_1, simplify_subreg): Likewise.
5693 * targhooks.c (default_function_arg_padding): Likewise.
5694 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5695 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5696 (verify_gimple_assign_ternary): Likewise.
5697 * tree-inline.c (estimate_move_cost): Likewise.
5698 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5699 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5700 (get_address_cost_ainc): Likewise.
5701 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5702 (vect_supportable_dr_alignment): Likewise.
5703 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5704 (vectorizable_reduction): Likewise.
5705 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5706 (vectorizable_operation, vectorizable_load): Likewise.
5707 * tree.c (build_same_sized_truth_vector_type): Likewise.
5708 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5709 * var-tracking.c (emit_note_insn_var_location): Likewise.
5710 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5711 (ADDR_VEC_ALIGN): Likewise.
5713 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5714 Alan Hayward <alan.hayward@arm.com>
5715 David Sherwood <david.sherwood@arm.com>
5717 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5719 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5720 or if measurement_type is polynomial.
5721 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5722 * combine.c (make_extraction): Likewise.
5723 * dse.c (find_shift_sequence): Likewise.
5724 * dwarf2out.c (mem_loc_descriptor): Likewise.
5725 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5726 (extract_bit_field, extract_low_bits): Likewise.
5727 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5728 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5729 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5730 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5731 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5732 * reload.c (find_reloads): Likewise.
5733 * reload1.c (alter_reg): Likewise.
5734 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5735 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5736 * tree-if-conv.c (predicate_mem_writes): Likewise.
5737 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5738 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5739 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5740 * valtrack.c (dead_debug_insert_temp): Likewise.
5741 * varasm.c (mergeable_constant_section): Likewise.
5742 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5744 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5745 Alan Hayward <alan.hayward@arm.com>
5746 David Sherwood <david.sherwood@arm.com>
5748 * expr.c (expand_assignment): Cope with polynomial mode sizes
5749 when assigning to a CONCAT.
5751 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5752 Alan Hayward <alan.hayward@arm.com>
5753 David Sherwood <david.sherwood@arm.com>
5755 * machmode.h (mode_precision): Change from unsigned short to
5757 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5759 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5760 or if measurement_type is not polynomial.
5761 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5762 in which the mode is already known to be a scalar_int_mode.
5763 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5764 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5766 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5767 for GET_MODE_PRECISION.
5768 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5769 for GET_MODE_PRECISION.
5770 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5772 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5773 (expand_field_assignment, make_extraction): Likewise.
5774 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5775 (get_last_value): Likewise.
5776 * convert.c (convert_to_integer_1): Likewise.
5777 * cse.c (cse_insn): Likewise.
5778 * expr.c (expand_expr_real_1): Likewise.
5779 * lra-constraints.c (simplify_operand_subreg): Likewise.
5780 * optabs-query.c (can_atomic_load_p): Likewise.
5781 * optabs.c (expand_atomic_load): Likewise.
5782 (expand_atomic_store): Likewise.
5783 * ree.c (combine_reaching_defs): Likewise.
5784 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5785 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5786 * tree.h (type_has_mode_precision_p): Likewise.
5787 * ubsan.c (instrument_si_overflow): Likewise.
5789 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5790 Alan Hayward <alan.hayward@arm.com>
5791 David Sherwood <david.sherwood@arm.com>
5793 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5794 polynomial numbers of units.
5795 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5796 (valid_vector_subparts_p): New function.
5797 (build_vector_type): Remove temporary shim and take the number
5798 of units as a poly_uint64 rather than an int.
5799 (build_opaque_vector_type): Take the number of units as a
5800 poly_uint64 rather than an int.
5801 * tree.c (build_vector_from_ctor): Handle polynomial
5802 TYPE_VECTOR_SUBPARTS.
5803 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5804 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5805 (build_vector_from_val): If the number of units is variable,
5806 use build_vec_duplicate_cst for constant operands and
5807 VEC_DUPLICATE_EXPR otherwise.
5808 (make_vector_type): Remove temporary is_constant ().
5809 (build_vector_type, build_opaque_vector_type): Take the number of
5810 units as a poly_uint64 rather than an int.
5811 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5813 * cfgexpand.c (expand_debug_expr): Likewise.
5814 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5815 (store_constructor, expand_expr_real_1): Likewise.
5816 (const_scalar_mask_from_tree): Likewise.
5817 * fold-const-call.c (fold_const_reduction): Likewise.
5818 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5819 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5820 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5821 (fold_relational_const): Likewise.
5822 (native_interpret_vector): Likewise. Change the size from an
5823 int to an unsigned int.
5824 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5825 TYPE_VECTOR_SUBPARTS.
5826 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5827 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5828 duplicating a non-constant operand into a variable-length vector.
5829 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5830 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5831 * ipa-icf.c (sem_variable::equals): Likewise.
5832 * match.pd: Likewise.
5833 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5834 * print-tree.c (print_node): Likewise.
5835 * stor-layout.c (layout_type): Likewise.
5836 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5837 * tree-cfg.c (verify_gimple_comparison): Likewise.
5838 (verify_gimple_assign_binary): Likewise.
5839 (verify_gimple_assign_ternary): Likewise.
5840 (verify_gimple_assign_single): Likewise.
5841 * tree-pretty-print.c (dump_generic_node): Likewise.
5842 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5843 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5844 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5845 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5846 (vect_shift_permute_load_chain): Likewise.
5847 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5848 (expand_vector_condition, optimize_vector_constructor): Likewise.
5849 (lower_vec_perm, get_compute_type): Likewise.
5850 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5851 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5852 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5853 (vect_recog_mask_conversion_pattern): Likewise.
5854 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5855 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5856 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5857 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5858 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5859 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5860 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5861 (supportable_widening_operation): Likewise.
5862 (supportable_narrowing_operation): Likewise.
5863 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5865 * varasm.c (output_constant): Likewise.
5867 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5868 Alan Hayward <alan.hayward@arm.com>
5869 David Sherwood <david.sherwood@arm.com>
5871 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5872 so that both the length == 3 and length != 3 cases set up their
5873 own permute vectors. Add comments explaining why we know the
5874 number of elements is constant.
5875 (vect_permute_load_chain): Likewise.
5877 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5878 Alan Hayward <alan.hayward@arm.com>
5879 David Sherwood <david.sherwood@arm.com>
5881 * machmode.h (mode_nunits): Change from unsigned char to
5883 (ONLY_FIXED_SIZE_MODES): New macro.
5884 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5885 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5886 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5888 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5889 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5890 or if measurement_type is not polynomial.
5891 * genmodes.c (ZERO_COEFFS): New macro.
5892 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5894 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5895 Use ZERO_COEFFS when emitting initializers.
5896 * data-streamer.h (bp_pack_poly_value): New function.
5897 (bp_unpack_poly_value): Likewise.
5898 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5899 for GET_MODE_NUNITS.
5900 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5901 for GET_MODE_NUNITS.
5902 * tree.c (make_vector_type): Remove temporary shim and make
5903 the real function take the number of units as a poly_uint64
5905 (build_vector_type_for_mode): Handle polynomial nunits.
5906 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5907 * emit-rtl.c (const_vec_series_p_1): Likewise.
5908 (gen_rtx_CONST_VECTOR): Likewise.
5909 * fold-const.c (test_vec_duplicate_folding): Likewise.
5910 * genrecog.c (validate_pattern): Likewise.
5911 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5912 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5913 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5914 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5915 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5916 * rtlanal.c (subreg_get_info): Likewise.
5917 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5918 (vect_grouped_load_supported): Likewise.
5919 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5920 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5921 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5922 (simplify_const_unary_operation, simplify_binary_operation_1)
5923 (simplify_const_binary_operation, simplify_ternary_operation)
5924 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5925 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5926 instead of CONST_VECTOR_NUNITS.
5927 * varasm.c (output_constant_pool_2): Likewise.
5928 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5929 explicit-encoded elements in the XVEC for variable-length vectors.
5931 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5933 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5935 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5936 Alan Hayward <alan.hayward@arm.com>
5937 David Sherwood <david.sherwood@arm.com>
5939 * coretypes.h (fixed_size_mode): Declare.
5940 (fixed_size_mode_pod): New typedef.
5941 * builtins.h (target_builtins::x_apply_args_mode)
5942 (target_builtins::x_apply_result_mode): Change type to
5943 fixed_size_mode_pod.
5944 * builtins.c (apply_args_size, apply_result_size, result_vector)
5945 (expand_builtin_apply_args_1, expand_builtin_apply)
5946 (expand_builtin_return): Update accordingly.
5948 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5950 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5951 * cselib.c (cselib_hash_rtx): Likewise.
5952 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5953 CONST_VECTOR encoding.
5955 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5956 Jeff Law <law@redhat.com>
5959 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5960 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5961 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5962 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5965 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5966 explicitly probe *sp in a noreturn function if there were any callee
5967 register saves or frame pointer is needed.
5969 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5972 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5973 BLKmode for ternary, binary or unary expressions.
5976 * var-tracking.c (delete_vta_debug_insn): New inline function.
5977 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5978 insns from get_insns () to NULL instead of each bb separately.
5979 Use delete_vta_debug_insn. No longer static.
5980 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5981 delete_vta_debug_insns callers.
5982 * rtl.h (delete_vta_debug_insns): Declare.
5983 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5984 instead of variable_tracking_main.
5986 2018-01-03 Martin Sebor <msebor@redhat.com>
5988 PR tree-optimization/83603
5989 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5990 arguments past the endof the argument list in functions declared
5991 without a prototype.
5992 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5993 Avoid checking when arguments are null.
5995 2018-01-03 Martin Sebor <msebor@redhat.com>
5998 * doc/extend.texi (attribute const): Fix a typo.
5999 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6000 issuing -Wsuggest-attribute for void functions.
6002 2018-01-03 Martin Sebor <msebor@redhat.com>
6004 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6005 offset_int::from instead of wide_int::to_shwi.
6006 (maybe_diag_overlap): Remove assertion.
6007 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6008 * gimple-ssa-sprintf.c (format_directive): Same.
6009 (parse_directive): Same.
6010 (sprintf_dom_walker::compute_format_length): Same.
6011 (try_substitute_return_value): Same.
6013 2018-01-03 Jeff Law <law@redhat.com>
6016 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6017 non-constant residual for zero at runtime and avoid probing in
6018 that case. Reorganize code for trailing problem to mirror handling
6021 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6023 PR tree-optimization/83501
6024 * tree-ssa-strlen.c (get_string_cst): New.
6025 (handle_char_store): Call get_string_cst.
6027 2018-01-03 Martin Liska <mliska@suse.cz>
6029 PR tree-optimization/83593
6030 * tree-ssa-strlen.c: Include tree-cfg.h.
6031 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6032 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6033 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6035 (strlen_dom_walker::before_dom_children): Call
6036 gimple_purge_dead_eh_edges. Dump tranformation with details
6038 (strlen_dom_walker::before_dom_children): Update call by adding
6039 new argument cleanup_eh.
6040 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6042 2018-01-03 Martin Liska <mliska@suse.cz>
6045 * cif-code.def (VARIADIC_THUNK): New enum value.
6046 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6049 2018-01-03 Jan Beulich <jbeulich@suse.com>
6051 * sse.md (mov<mode>_internal): Tighten condition for when to use
6052 vmovdqu<ssescalarsize> for TI and OI modes.
6054 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6056 Update copyright years.
6058 2018-01-03 Martin Liska <mliska@suse.cz>
6061 * ipa-visibility.c (function_and_variable_visibility): Skip
6062 functions with noipa attribure.
6064 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6066 * gcc.c (process_command): Update copyright notice dates.
6067 * gcov-dump.c (print_version): Ditto.
6068 * gcov.c (print_version): Ditto.
6069 * gcov-tool.c (print_version): Ditto.
6070 * gengtype.c (create_file): Ditto.
6071 * doc/cpp.texi: Bump @copying's copyright year.
6072 * doc/cppinternals.texi: Ditto.
6073 * doc/gcc.texi: Ditto.
6074 * doc/gccint.texi: Ditto.
6075 * doc/gcov.texi: Ditto.
6076 * doc/install.texi: Ditto.
6077 * doc/invoke.texi: Ditto.
6079 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6081 * vector-builder.h (vector_builder::m_full_nelts): Change from
6082 unsigned int to poly_uint64.
6083 (vector_builder::full_nelts): Update prototype accordingly.
6084 (vector_builder::new_vector): Likewise.
6085 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6086 (vector_builder::operator ==): Likewise.
6087 (vector_builder::finalize): Likewise.
6088 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6089 Take the number of elements as a poly_uint64 rather than an
6091 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6092 from unsigned int to poly_uint64.
6093 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6094 (vec_perm_indices::new_vector): Likewise.
6095 (vec_perm_indices::length): Likewise.
6096 (vec_perm_indices::nelts_per_input): Likewise.
6097 (vec_perm_indices::input_nelts): Likewise.
6098 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6099 number of elements per input as a poly_uint64 rather than an
6100 unsigned int. Use the original encoding for variable-length
6101 vectors, rather than clamping each individual element.
6102 For the second and subsequent elements in each pattern,
6103 clamp the step and base before clamping their sum.
6104 (vec_perm_indices::series_p): Handle polynomial element counts.
6105 (vec_perm_indices::all_in_range_p): Likewise.
6106 (vec_perm_indices_to_tree): Likewise.
6107 (vec_perm_indices_to_rtx): Likewise.
6108 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6109 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6110 (tree_vector_builder::new_binary_operation): Handle polynomial
6111 element counts. Return false if we need to know the number
6112 of elements at compile time.
6113 * fold-const.c (fold_vec_perm): Punt if the number of elements
6114 isn't known at compile time.
6116 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6118 * vec-perm-indices.h (vec_perm_builder): Change element type
6119 from HOST_WIDE_INT to poly_int64.
6120 (vec_perm_indices::element_type): Update accordingly.
6121 (vec_perm_indices::clamp): Handle polynomial element_types.
6122 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6123 (vec_perm_indices::all_in_range_p): Likewise.
6124 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6126 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6127 polynomial vec_perm_indices element types.
6128 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6129 * fold-const.c (fold_vec_perm): Likewise.
6130 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6131 * tree-vect-generic.c (lower_vec_perm): Likewise.
6132 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6133 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6134 element type to HOST_WIDE_INT.
6136 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6137 Alan Hayward <alan.hayward@arm.com>
6138 David Sherwood <david.sherwood@arm.com>
6140 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6141 rather than an int. Use plus_constant.
6142 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6143 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6145 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6146 Alan Hayward <alan.hayward@arm.com>
6147 David Sherwood <david.sherwood@arm.com>
6149 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6150 a HOST_WIDE_INT to a poly_int64.
6152 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6153 Alan Hayward <alan.hayward@arm.com>
6154 David Sherwood <david.sherwood@arm.com>
6156 * calls.c (load_register_parameters): Cope with polynomial
6157 mode sizes. Require a constant size for BLKmode parameters
6158 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6159 forces a parameter to be padded at the lsb end in order to
6160 fill a complete number of words, require the parameter size
6161 to be ordered wrt UNITS_PER_WORD.
6163 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6164 Alan Hayward <alan.hayward@arm.com>
6165 David Sherwood <david.sherwood@arm.com>
6167 * reload1.c (spill_stack_slot_width): Change element type
6168 from unsigned int to poly_uint64_pod.
6169 (alter_reg): Treat mode sizes as polynomial.
6171 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6172 Alan Hayward <alan.hayward@arm.com>
6173 David Sherwood <david.sherwood@arm.com>
6175 * reload.c (complex_word_subreg_p): New function.
6176 (reload_inner_reg_of_subreg, push_reload): Use it.
6178 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6179 Alan Hayward <alan.hayward@arm.com>
6180 David Sherwood <david.sherwood@arm.com>
6182 * lra-constraints.c (process_alt_operands): Reject matched
6183 operands whose sizes aren't ordered.
6184 (match_reload): Refer to this check here.
6186 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6187 Alan Hayward <alan.hayward@arm.com>
6188 David Sherwood <david.sherwood@arm.com>
6190 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6191 that the mode size is in the set {1, 2, 4, 8, 16}.
6193 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6194 Alan Hayward <alan.hayward@arm.com>
6195 David Sherwood <david.sherwood@arm.com>
6197 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6198 Use plus_constant instead of gen_rtx_PLUS.
6200 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6201 Alan Hayward <alan.hayward@arm.com>
6202 David Sherwood <david.sherwood@arm.com>
6204 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6205 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6206 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6207 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6208 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6209 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6210 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6211 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6212 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6213 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6215 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6216 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6217 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6218 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6219 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6220 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6221 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6222 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6223 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6224 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6226 * expr.c (emit_move_resolve_push): Treat the input and result
6227 of PUSH_ROUNDING as a poly_int64.
6228 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6229 (emit_push_insn): Likewise.
6230 * lra-eliminations.c (mark_not_eliminable): Likewise.
6231 * recog.c (push_operand): Likewise.
6232 * reload1.c (elimination_effects): Likewise.
6233 * rtlanal.c (nonzero_bits1): Likewise.
6234 * calls.c (store_one_arg): Likewise. Require the padding to be
6235 known at compile time.
6237 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6238 Alan Hayward <alan.hayward@arm.com>
6239 David Sherwood <david.sherwood@arm.com>
6241 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6242 Use plus_constant instead of gen_rtx_PLUS.
6244 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6245 Alan Hayward <alan.hayward@arm.com>
6246 David Sherwood <david.sherwood@arm.com>
6248 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6251 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6252 Alan Hayward <alan.hayward@arm.com>
6253 David Sherwood <david.sherwood@arm.com>
6255 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6256 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6257 via stack temporaries. Treat the mode size as polynomial too.
6259 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6260 Alan Hayward <alan.hayward@arm.com>
6261 David Sherwood <david.sherwood@arm.com>
6263 * expr.c (expand_expr_real_2): When handling conversions involving
6264 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6265 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6266 as a poly_uint64 too.
6268 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6269 Alan Hayward <alan.hayward@arm.com>
6270 David Sherwood <david.sherwood@arm.com>
6272 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6274 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6275 Alan Hayward <alan.hayward@arm.com>
6276 David Sherwood <david.sherwood@arm.com>
6278 * combine.c (can_change_dest_mode): Handle polynomial
6279 REGMODE_NATURAL_SIZE.
6280 * expmed.c (store_bit_field_1): Likewise.
6281 * expr.c (store_constructor): Likewise.
6282 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6283 and polynomial REGMODE_NATURAL_SIZE.
6284 (gen_lowpart_common): Likewise.
6285 * reginfo.c (record_subregs_of_mode): Likewise.
6286 * rtlanal.c (read_modify_subreg_p): Likewise.
6288 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6289 Alan Hayward <alan.hayward@arm.com>
6290 David Sherwood <david.sherwood@arm.com>
6292 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6293 numbers of elements.
6295 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6296 Alan Hayward <alan.hayward@arm.com>
6297 David Sherwood <david.sherwood@arm.com>
6299 * match.pd: Cope with polynomial numbers of vector elements.
6301 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6302 Alan Hayward <alan.hayward@arm.com>
6303 David Sherwood <david.sherwood@arm.com>
6305 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6306 in a POINTER_PLUS_EXPR.
6308 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6309 Alan Hayward <alan.hayward@arm.com>
6310 David Sherwood <david.sherwood@arm.com>
6312 * omp-simd-clone.c (simd_clone_subparts): New function.
6313 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6314 (ipa_simd_modify_function_body): Likewise.
6316 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6317 Alan Hayward <alan.hayward@arm.com>
6318 David Sherwood <david.sherwood@arm.com>
6320 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6321 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6322 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6323 (expand_vector_condition, vector_element): Likewise.
6324 (subparts_gt): New function.
6325 (get_compute_type): Use subparts_gt.
6326 (count_type_subparts): Delete.
6327 (expand_vector_operations_1): Use subparts_gt instead of
6328 count_type_subparts.
6330 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6331 Alan Hayward <alan.hayward@arm.com>
6332 David Sherwood <david.sherwood@arm.com>
6334 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6335 (vect_compile_time_alias): ...this new function. Do the calculation
6336 on poly_ints rather than trees.
6337 (vect_prune_runtime_alias_test_list): Update call accordingly.
6339 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6340 Alan Hayward <alan.hayward@arm.com>
6341 David Sherwood <david.sherwood@arm.com>
6343 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6345 (vect_schedule_slp_instance): Likewise.
6347 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6348 Alan Hayward <alan.hayward@arm.com>
6349 David Sherwood <david.sherwood@arm.com>
6351 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6352 constant and extern definitions for variable-length vectors.
6353 (vect_get_constant_vectors): Note that the number of units
6354 is known to be constant.
6356 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6357 Alan Hayward <alan.hayward@arm.com>
6358 David Sherwood <david.sherwood@arm.com>
6360 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6361 of units as polynomial. Choose between WIDE and NARROW based
6364 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6365 Alan Hayward <alan.hayward@arm.com>
6366 David Sherwood <david.sherwood@arm.com>
6368 * tree-vect-stmts.c (simd_clone_subparts): New function.
6369 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6371 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6372 Alan Hayward <alan.hayward@arm.com>
6373 David Sherwood <david.sherwood@arm.com>
6375 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6376 vectors as polynomial. Use build_index_vector for
6379 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6380 Alan Hayward <alan.hayward@arm.com>
6381 David Sherwood <david.sherwood@arm.com>
6383 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6384 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6385 for variable-length vectors.
6386 (vectorizable_mask_load_store): Treat the number of units as
6387 polynomial, asserting that it is constant if the condition has
6388 already been enforced.
6389 (vectorizable_store, vectorizable_load): Likewise.
6391 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6392 Alan Hayward <alan.hayward@arm.com>
6393 David Sherwood <david.sherwood@arm.com>
6395 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6396 of units as polynomial. Punt if we can't tell at compile time
6397 which vector contains the final result.
6399 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6400 Alan Hayward <alan.hayward@arm.com>
6401 David Sherwood <david.sherwood@arm.com>
6403 * tree-vect-loop.c (vectorizable_induction): Treat the number
6404 of units as polynomial. Punt on SLP inductions. Use an integer
6405 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6406 cast of such a series for variable-length floating-point
6409 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6410 Alan Hayward <alan.hayward@arm.com>
6411 David Sherwood <david.sherwood@arm.com>
6413 * tree.h (build_index_vector): Declare.
6414 * tree.c (build_index_vector): New function.
6415 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6416 of units as polynomial, forcibly converting it to a constant if
6417 vectorizable_reduction has already enforced the condition.
6418 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6419 to create a {1,2,3,...} vector.
6420 (vectorizable_reduction): Treat the number of units as polynomial.
6421 Choose vectype_in based on the largest scalar element size rather
6422 than the smallest number of units. Enforce the restrictions
6425 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6426 Alan Hayward <alan.hayward@arm.com>
6427 David Sherwood <david.sherwood@arm.com>
6429 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6430 number of units as polynomial.
6432 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6433 Alan Hayward <alan.hayward@arm.com>
6434 David Sherwood <david.sherwood@arm.com>
6436 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6437 * target.def (autovectorize_vector_sizes): Return the vector sizes
6438 by pointer, using vector_sizes rather than a bitmask.
6439 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6440 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6441 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6443 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6444 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6445 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6446 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6447 * omp-general.c (omp_max_vf): Likewise.
6448 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6449 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6450 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6451 * tree-vect-slp.c (vect_slp_bb): Likewise.
6452 * doc/tm.texi: Regenerate.
6453 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6455 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6456 the vector size as a poly_uint64 rather than an unsigned int.
6457 (current_vector_size): Change from an unsigned int to a poly_uint64.
6458 (get_vectype_for_scalar_type): Update accordingly.
6459 * tree.h (build_truth_vector_type): Take the size and number of
6460 units as a poly_uint64 rather than an unsigned int.
6461 (build_vector_type): Add a temporary overload that takes
6462 the number of units as a poly_uint64 rather than an unsigned int.
6463 * tree.c (make_vector_type): Likewise.
6464 (build_truth_vector_type): Take the number of units as a poly_uint64
6465 rather than an unsigned int.
6467 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6468 Alan Hayward <alan.hayward@arm.com>
6469 David Sherwood <david.sherwood@arm.com>
6471 * target.def (get_mask_mode): Take the number of units and length
6472 as poly_uint64s rather than unsigned ints.
6473 * targhooks.h (default_get_mask_mode): Update accordingly.
6474 * targhooks.c (default_get_mask_mode): Likewise.
6475 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6476 * doc/tm.texi: Regenerate.
6478 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6479 Alan Hayward <alan.hayward@arm.com>
6480 David Sherwood <david.sherwood@arm.com>
6482 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6483 * omp-general.c (omp_max_vf): Likewise.
6484 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6485 (expand_omp_simd): Handle polynomial safelen.
6486 * omp-low.c (omplow_simd_context): Add a default constructor.
6487 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6488 (lower_rec_simd_input_clauses): Update accordingly.
6489 (lower_rec_input_clauses): Likewise.
6491 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6492 Alan Hayward <alan.hayward@arm.com>
6493 David Sherwood <david.sherwood@arm.com>
6495 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6496 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6497 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6498 (vect_analyze_slp_cost): Likewise.
6499 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6500 (vect_model_load_cost): Likewise.
6502 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6503 Alan Hayward <alan.hayward@arm.com>
6504 David Sherwood <david.sherwood@arm.com>
6506 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6507 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6508 from an unsigned int * to a poly_uint64_pod *.
6509 (calculate_unrolling_factor): New function.
6510 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6512 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6513 Alan Hayward <alan.hayward@arm.com>
6514 David Sherwood <david.sherwood@arm.com>
6516 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6517 from an unsigned int to a poly_uint64.
6518 (_loop_vec_info::slp_unrolling_factor): Likewise.
6519 (_loop_vec_info::vectorization_factor): Change from an int
6521 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6522 (vect_get_num_vectors): New function.
6523 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6524 (vect_get_num_copies): Use vect_get_num_vectors.
6525 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6526 to an unsigned int *.
6527 (vect_analyze_data_refs): Change min_vf from an int * to a
6529 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6530 than an unsigned HOST_WIDE_INT.
6531 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6532 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6533 to an unsigned int *.
6534 (vect_analyze_data_ref_dependences): Likewise.
6535 (vect_compute_data_ref_alignment): Handle polynomial vf.
6536 (vect_enhance_data_refs_alignment): Likewise.
6537 (vect_prune_runtime_alias_test_list): Likewise.
6538 (vect_shift_permute_load_chain): Likewise.
6539 (vect_supportable_dr_alignment): Likewise.
6540 (dependence_distance_ge_vf): Take the vectorization factor as a
6541 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6542 (vect_analyze_data_refs): Change min_vf from an int * to a
6544 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6545 vfm1 as a poly_uint64 rather than an int. Make the same change
6546 for the returned bound_scalar.
6547 (vect_gen_vector_loop_niters): Handle polynomial vf.
6548 (vect_do_peeling): Likewise. Update call to
6549 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6550 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6552 * tree-vect-loop.c (vect_determine_vectorization_factor)
6553 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6554 (vect_get_known_peeling_cost): Likewise.
6555 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6556 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6557 (vect_transform_loop): Likewise. Use the lowest possible VF when
6558 updating the upper bounds of the loop.
6559 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6561 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6562 polynomial unroll factors.
6563 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6564 (vect_make_slp_decision): Likewise.
6565 (vect_supported_load_permutation_p): Likewise, and polynomial
6567 (vect_analyze_slp_cost): Handle polynomial vf.
6568 (vect_slp_analyze_node_operations): Likewise.
6569 (vect_slp_analyze_bb_1): Likewise.
6570 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6571 than an unsigned HOST_WIDE_INT.
6572 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6573 (vectorizable_load): Handle polynomial vf.
6574 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6576 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6578 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6579 Alan Hayward <alan.hayward@arm.com>
6580 David Sherwood <david.sherwood@arm.com>
6582 * match.pd: Handle bit operations involving three constants
6583 and try to fold one pair.
6585 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6587 * tree-vect-loop-manip.c: Include gimple-fold.h.
6588 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6589 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6590 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6591 Add a path that uses a step of VF instead of 1, but disable it
6593 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6594 and niters_no_overflow parameters. Update calls to
6595 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6596 Create a new SSA name if the latter choses to use a ste other
6597 than zero, and return it via niters_vector_mult_vf_var.
6598 * tree-vect-loop.c (vect_transform_loop): Update calls to
6599 vect_do_peeling, vect_gen_vector_loop_niters and
6600 slpeel_make_loop_iterate_ntimes.
6601 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6602 (vect_gen_vector_loop_niters): Update declarations after above changes.
6604 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6606 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6607 128-bit round to integer instructions.
6608 (ceil<mode>2): Likewise.
6609 (btrunc<mode>2): Likewise.
6610 (round<mode>2): Likewise.
6612 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6614 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6615 unaligned VSX load/store on P8/P9.
6616 (expand_block_clear): Allow the use of unaligned VSX
6617 load/store on P8/P9.
6619 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6621 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6623 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6624 swap associated with both a load and a store.
6626 2018-01-02 Andrew Waterman <andrew@sifive.com>
6628 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6629 * config/riscv/riscv.md (clear_cache): Use it.
6631 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6633 * web.c: Remove out-of-date comment.
6635 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6637 * expr.c (fixup_args_size_notes): Check that any existing
6638 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6639 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6640 (emit_single_push_insn): ...here.
6642 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6644 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6645 (const_vector_encoded_nelts): New function.
6646 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6647 (const_vector_int_elt, const_vector_elt): Declare.
6648 * emit-rtl.c (const_vector_int_elt_1): New function.
6649 (const_vector_elt): Likewise.
6650 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6651 of CONST_VECTOR_ELT.
6653 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6655 * expr.c: Include rtx-vector-builder.h.
6656 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6657 directly on the tree encoding.
6658 (const_vector_from_tree): Likewise.
6659 * optabs.c: Include rtx-vector-builder.h.
6660 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6661 sequence of "u" values.
6662 * vec-perm-indices.c: Include rtx-vector-builder.h.
6663 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6664 directly on the vec_perm_indices encoding.
6666 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6668 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6669 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6670 * rtx-vector-builder.h: New file.
6671 * rtx-vector-builder.c: Likewise.
6672 * rtl.h (rtx_def::u2): Add a const_vector field.
6673 (CONST_VECTOR_NPATTERNS): New macro.
6674 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6675 (CONST_VECTOR_DUPLICATE_P): Likewise.
6676 (CONST_VECTOR_STEPPED_P): Likewise.
6677 (CONST_VECTOR_ENCODED_ELT): Likewise.
6678 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6679 (unwrap_const_vec_duplicate): Likewise.
6680 (const_vec_series_p): Check for a non-duplicated vector encoding.
6681 Say that the function only returns true for integer vectors.
6682 * emit-rtl.c: Include rtx-vector-builder.h.
6683 (gen_const_vec_duplicate_1): Delete.
6684 (gen_const_vector): Call gen_const_vec_duplicate instead of
6685 gen_const_vec_duplicate_1.
6686 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6687 (gen_const_vec_duplicate): Use rtx_vector_builder.
6688 (gen_const_vec_series): Likewise.
6689 (gen_rtx_CONST_VECTOR): Likewise.
6690 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6691 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6692 Build a new vector rather than modifying a CONST_VECTOR in-place.
6693 (handle_special_swappables): Update call accordingly.
6694 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6695 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6696 Build a new vector rather than modifying a CONST_VECTOR in-place.
6697 (handle_special_swappables): Update call accordingly.
6699 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6701 * simplify-rtx.c (simplify_const_binary_operation): Use
6702 CONST_VECTOR_ELT instead of XVECEXP.
6704 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6706 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6707 the selector elements to be different from the data elements
6708 if the selector is a VECTOR_CST.
6709 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6710 ssizetype for the selector.
6712 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6714 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6715 before testing each element individually.
6716 * tree-vect-generic.c (lower_vec_perm): Likewise.
6718 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6720 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6721 * selftest-run-tests.c (selftest::run_tests): Call it.
6722 * vector-builder.h (vector_builder::operator ==): New function.
6723 (vector_builder::operator !=): Likewise.
6724 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6725 (vec_perm_indices::all_from_input_p): New function.
6726 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6727 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6728 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6729 instead of reading the VECTOR_CST directly. Detect whether both
6730 vector inputs are the same before constructing the vec_perm_indices,
6731 and update the number of inputs argument accordingly. Use the
6732 utility functions added above. Only construct sel2 if we need to.
6734 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6736 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6737 the broadcast of the low byte.
6738 (expand_mult_highpart): Use an explicit encoding for the permutes.
6739 * optabs-query.c (can_mult_highpart_p): Likewise.
6740 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6741 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6742 (vectorizable_bswap): Likewise.
6743 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6744 explicit encoding for the power-of-2 permutes.
6745 (vect_permute_store_chain): Likewise.
6746 (vect_grouped_load_supported): Likewise.
6747 (vect_permute_load_chain): Likewise.
6749 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6751 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6752 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6753 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6754 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6755 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6756 (vect_gen_perm_mask_any): Likewise.
6758 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6760 * int-vector-builder.h: New file.
6761 * vec-perm-indices.h: Include int-vector-builder.h.
6762 (vec_perm_indices): Redefine as an int_vector_builder.
6763 (auto_vec_perm_indices): Delete.
6764 (vec_perm_builder): Redefine as a stand-alone class.
6765 (vec_perm_indices::vec_perm_indices): New function.
6766 (vec_perm_indices::clamp): Likewise.
6767 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6768 (vec_perm_indices::new_vector): New function.
6769 (vec_perm_indices::new_expanded_vector): Update for new
6770 vec_perm_indices class.
6771 (vec_perm_indices::rotate_inputs): New function.
6772 (vec_perm_indices::all_in_range_p): Operate directly on the
6773 encoded form, without computing elided elements.
6774 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6775 encoding. Update for new vec_perm_indices class.
6776 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6777 the given vec_perm_builder.
6778 (expand_vec_perm_var): Update vec_perm_builder constructor.
6779 (expand_mult_highpart): Use vec_perm_builder instead of
6780 auto_vec_perm_indices.
6781 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6782 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6783 or double series encoding as appropriate.
6784 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6785 vec_perm_indices instead of auto_vec_perm_indices.
6786 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6787 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6788 (vect_permute_store_chain): Likewise.
6789 (vect_grouped_load_supported): Likewise.
6790 (vect_permute_load_chain): Likewise.
6791 (vect_shift_permute_load_chain): Likewise.
6792 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6793 (vect_transform_slp_perm_load): Likewise.
6794 (vect_schedule_slp_instance): Likewise.
6795 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6796 (vectorizable_mask_load_store): Likewise.
6797 (vectorizable_bswap): Likewise.
6798 (vectorizable_store): Likewise.
6799 (vectorizable_load): Likewise.
6800 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6801 vec_perm_indices instead of auto_vec_perm_indices. Use
6802 tree_to_vec_perm_builder to read the vector from a tree.
6803 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6804 vec_perm_builder instead of a vec_perm_indices.
6805 (have_whole_vector_shift): Use vec_perm_builder and
6806 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6807 truncation to calc_vec_perm_mask_for_shift.
6808 (vect_create_epilog_for_reduction): Likewise.
6809 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6810 from auto_vec_perm_indices to vec_perm_indices.
6811 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6812 instead of changing individual elements.
6813 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6814 the vector in d.perm.
6815 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6816 from auto_vec_perm_indices to vec_perm_indices.
6817 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6818 instead of changing individual elements.
6819 (arm_vectorize_vec_perm_const): Use new_vector to install
6820 the vector in d.perm.
6821 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6822 Update vec_perm_builder constructor.
6823 (rs6000_expand_interleave): Likewise.
6824 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6825 (rs6000_expand_interleave): Likewise.
6827 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6829 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6830 to qimode could truncate the indices.
6831 * optabs.c (expand_vec_perm_var): Likewise.
6833 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6835 * Makefile.in (OBJS): Add vec-perm-indices.o.
6836 * vec-perm-indices.h: New file.
6837 * vec-perm-indices.c: Likewise.
6838 * target.h (vec_perm_indices): Replace with a forward class
6840 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6841 * optabs.h: Include vec-perm-indices.h.
6842 (expand_vec_perm): Delete.
6843 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6844 (expand_vec_perm_const): Declare.
6845 * target.def (vec_perm_const_ok): Replace with...
6846 (vec_perm_const): ...this new hook.
6847 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6848 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6849 * doc/tm.texi: Regenerate.
6850 * optabs.def (vec_perm_const): Delete.
6851 * doc/md.texi (vec_perm_const): Likewise.
6852 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6853 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6854 expand_vec_perm for constant permutation vectors. Assert that
6855 the mode of variable permutation vectors is the integer equivalent
6856 of the mode that is being permuted.
6857 * optabs-query.h (selector_fits_mode_p): Declare.
6858 * optabs-query.c: Include vec-perm-indices.h.
6859 (selector_fits_mode_p): New function.
6860 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6861 is defined, instead of checking whether the vec_perm_const_optab
6862 exists. Use targetm.vectorize.vec_perm_const instead of
6863 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6864 fit in the vector mode before using a variable permute.
6865 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6866 vec_perm_indices instead of an rtx.
6867 (expand_vec_perm): Replace with...
6868 (expand_vec_perm_const): ...this new function. Take the selector
6869 as a vec_perm_indices rather than an rtx. Also take the mode of
6870 the selector. Update call to shift_amt_for_vec_perm_mask.
6871 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6872 Use vec_perm_indices::new_expanded_vector to expand the original
6873 selector into bytes. Check whether the indices fit in the vector
6874 mode before using a variable permute.
6875 (expand_vec_perm_var): Make global.
6876 (expand_mult_highpart): Use expand_vec_perm_const.
6877 * fold-const.c: Includes vec-perm-indices.h.
6878 * tree-ssa-forwprop.c: Likewise.
6879 * tree-vect-data-refs.c: Likewise.
6880 * tree-vect-generic.c: Likewise.
6881 * tree-vect-loop.c: Likewise.
6882 * tree-vect-slp.c: Likewise.
6883 * tree-vect-stmts.c: Likewise.
6884 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6886 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6887 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6888 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6889 (aarch64_vectorize_vec_perm_const): ...this new function.
6890 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6891 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6892 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6893 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6894 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6895 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6896 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6898 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6899 check for NEON modes.
6900 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6901 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6902 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6903 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6905 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6906 the old VEC_PERM_CONST conditions.
6907 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6908 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6909 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6910 (ia64_vectorize_vec_perm_const_ok): Merge into...
6911 (ia64_vectorize_vec_perm_const): ...this new function.
6912 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6913 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6914 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6915 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6916 * config/mips/mips.c (mips_expand_vec_perm_const)
6917 (mips_vectorize_vec_perm_const_ok): Merge into...
6918 (mips_vectorize_vec_perm_const): ...this new function.
6919 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6920 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6921 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6922 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6923 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6924 (rs6000_expand_vec_perm_const): Delete.
6925 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6927 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6928 (altivec_expand_vec_perm_const_le): Take each operand individually.
6929 Operate on constant selectors rather than rtxes.
6930 (altivec_expand_vec_perm_const): Likewise. Update call to
6931 altivec_expand_vec_perm_const_le.
6932 (rs6000_expand_vec_perm_const): Delete.
6933 (rs6000_vectorize_vec_perm_const_ok): Delete.
6934 (rs6000_vectorize_vec_perm_const): New function.
6935 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6936 an element count and rtx array.
6937 (rs6000_expand_extract_even): Update call accordingly.
6938 (rs6000_expand_interleave): Likewise.
6939 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6940 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6941 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6942 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6943 (rs6000_expand_vec_perm_const): Delete.
6944 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6945 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6946 (altivec_expand_vec_perm_const_le): Take each operand individually.
6947 Operate on constant selectors rather than rtxes.
6948 (altivec_expand_vec_perm_const): Likewise. Update call to
6949 altivec_expand_vec_perm_const_le.
6950 (rs6000_expand_vec_perm_const): Delete.
6951 (rs6000_vectorize_vec_perm_const_ok): Delete.
6952 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6953 reference to the SPE evmerge intructions.
6954 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6955 an element count and rtx array.
6956 (rs6000_expand_extract_even): Update call accordingly.
6957 (rs6000_expand_interleave): Likewise.
6958 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6959 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6961 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6963 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6965 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6966 vector mode and that that mode matches the mode of the data
6968 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6969 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6970 directly using expand_vec_perm_1 when forcing selectors into
6972 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6974 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6976 * optabs-query.h (can_vec_perm_p): Delete.
6977 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6978 * optabs-query.c (can_vec_perm_p): Split into...
6979 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6980 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6981 particular selector is valid.
6982 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6983 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6984 (vect_grouped_load_supported): Likewise.
6985 (vect_shift_permute_load_chain): Likewise.
6986 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6987 (vect_transform_slp_perm_load): Likewise.
6988 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6989 (vectorizable_bswap): Likewise.
6990 (vect_gen_perm_mask_checked): Likewise.
6991 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6992 implementations of variable permutation vectors into account
6993 when deciding which selector to use.
6994 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6995 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6996 with a false third argument.
6997 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6998 to test whether the constant selector is valid and can_vec_perm_var_p
6999 to test whether a variable selector is valid.
7001 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7003 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7004 * optabs-query.c (can_vec_perm_p): Likewise.
7005 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7006 instead of vec_perm_indices.
7007 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7008 (vect_gen_perm_mask_checked): Likewise,
7009 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7010 (vect_gen_perm_mask_checked): Likewise,
7012 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7014 * optabs-query.h (qimode_for_vec_perm): Declare.
7015 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7016 (qimode_for_vec_perm): ...this new function.
7017 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7019 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7021 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7022 does not have a conditional at the top.
7024 2018-01-02 Richard Biener <rguenther@suse.de>
7026 * ipa-inline.c (big_speedup_p): Fix expression.
7028 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7031 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7034 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7038 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7039 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7040 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7041 cond_taken_branch_cost 3->4.
7043 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7045 PR tree-optimization/83581
7046 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7047 TODO_cleanup_cfg if any changes have been made.
7050 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7051 convert_modes if target mode has the right side, but different mode
7055 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7056 last argument when extracting from CONCAT. If either from_real or
7057 from_imag is NULL, use expansion through memory. If result is not
7058 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7059 the parts directly to inner mode, if even that fails, use expansion
7063 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7064 check for bswap in mode rather than HImode and use that in expand_unop
7067 Copyright (C) 2018 Free Software Foundation, Inc.
7069 Copying and distribution of this file, with or without modification,
7070 are permitted in any medium without royalty provided the copyright
7071 notice and this notice are preserved.