1 2018-02-19 Sebastian Perta <sebastian.perta@renesas.com>
3 * config/rl78/rl78.md (movdf): New define expand.
5 2018-02-19 Martin Liska <mliska@suse.cz>
8 * doc/invoke.texi: Fix typo.
9 * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
11 2018-02-18 Segher Boessenkool <segher@kernel.crashing.org>
13 * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
14 handle rs6000_single_float and rs6000_double_float specially for
17 2018-02-16 Jeff Law <law@redhat.com>
19 * config/rx/rx.c (add_pop_cfi_notes): New function.;
22 2018-02-16 Jakub Jelinek <jakub@redhat.com>
25 * ipa-inline.c (inline_small_functions): Fix a typo.
27 2018-02-16 Nathan Sidwell <nathan@acm.org>
29 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
31 2018-02-16 Carl Love <cel@us.ibm.com>
33 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
34 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
35 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
36 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
37 expansion to P8V_BUILTIN_VEC_FLOAT2.
39 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
41 PR rtl-optimization/70023
42 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
43 src_regno into account.
45 2018-02-16 Carl Love <cel@us.ibm.com>
47 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
48 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
49 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
50 * config/rs6000/rs6000.c: Remove case statements for
51 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
52 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
53 and P9V_BUILTIN_VEC_VINSERT4B.
54 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
55 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
56 * config/rs6000/vsx.md:
57 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
60 2018-02-16 Carl Love <cel@us.ibm.com>
62 * config/rs6000/altivec.h: Add builtin names vec_extract4b
64 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
66 * config/rs6000/rs6000-c.c: Add the definitions for
67 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
68 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
69 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
70 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
71 definition for insert4b and define insn *insert3b_internal.
72 * doc/extend.texi: Add documentation for vec_extract4b.
74 2018-02-16 Nathan Sidwell <nathan@acm.org>
76 * doc/extend.texi (Backwards Compatibility): Mention friend
77 injection. Note for-scope is deprecated.
78 * doc/invoke.texi (-ffriend-injection): Deprecate.
80 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
82 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
83 that moved to I2, also allow destinations that are a paradoxical
84 subreg (instead of a normal reg).
86 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
89 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
92 2018-02-16 Richard Biener <rguenther@suse.de>
94 PR tree-optimization/84037
95 PR tree-optimization/84016
97 * config/i386/i386.c (ix86_builtin_vectorization_cost):
98 Adjust vec_construct for the fact we need additional higher latency
99 128bit inserts for AVX256 and AVX512 vector builds.
100 (ix86_add_stmt_cost): Scale vector construction cost for
103 2018-02-16 Richard Biener <rguenther@suse.de>
105 PR tree-optimization/84417
106 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
107 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
108 (non_rewritable_lvalue_p): Likewise, use poly-ints.
110 2018-02-16 Martin Liska <mliska@suse.cz>
113 * internal-fn.def (ASAN_CHECK): Set proper flags.
114 (ASAN_MARK): Likewise.
116 2018-02-16 Julia Koval <julia.koval@intel.com>
118 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
121 2018-02-16 Jakub Jelinek <jakub@redhat.com>
124 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
125 Use ++iter rather than iter++ for std::list iterators.
126 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
127 defer deleting them until all nodes in the forest are processed. Do
128 free even leaf nodes. Change to_process into auto_vec.
131 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
132 * vec.h (vec_default_construct): Use memset instead of placement new
133 if BROKEN_VALUE_INITIALIZATION is defined.
134 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
135 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
138 PR rtl-optimization/83723
139 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
140 * lra.c (lra_substitute_pseudo): Likewise. If true, use
141 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
143 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
145 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
147 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
149 PR rtl-optimization/81443
150 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
151 from inner REGs to paradoxical SUBREGs.
153 2018-02-16 Richard Biener <rguenther@suse.de>
155 PR tree-optimization/84399
156 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
157 For operands we can analyze at their definition make sure we can
158 analyze them at each use as well.
160 2018-02-16 Richard Biener <rguenther@suse.de>
162 PR tree-optimization/84190
163 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
164 volatile accesses if the decl isn't volatile.
166 2018-02-15 Jason Merrill <jason@redhat.com>
168 PR c++/84314 - ICE with templates and fastcall attribute.
169 * attribs.c (build_type_attribute_qual_variant): Don't clobber
170 TYPE_CANONICAL on an existing type.
172 2018-02-15 Jakub Jelinek <jakub@redhat.com>
174 PR tree-optimization/84383
175 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
176 dstoff nor call operand_equal_p if dstbase is NULL.
178 PR tree-optimization/84334
179 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
180 also a CONSTANT_CLASS_P, punt.
182 2018-02-14 Jim Wilson <jimw@sifive.com>
184 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
185 first SMALL_OPERAND check. New local min_second_step. Move assert
186 to where locals are set. Add TARGET_RVC support.
187 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
189 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
191 * doc/invoke.texi: Correct -Wformat-overflow code sample.
193 2018-02-14 Martin Sebor <msebor@redhat.com>
195 PR tree-optimization/83698
196 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
197 arrays constrain the offset range to their bounds.
198 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
199 (builtin_access::overlap): Avoid setting the size of overlap if it's
201 (maybe_diag_overlap): Also consider arrays when deciding what values
202 of offsets to include in diagnostics.
204 2018-02-14 Martin Sebor <msebor@redhat.com>
207 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
208 that correspond to the kind of a declaration.
210 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
213 * config/pa/pa.md: Load address of PIC label using the linkage table
214 if the label is nonlocal.
216 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
218 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
219 warning message if user requests -maltivec=be.
220 * doc/invoke.texi: Document deprecation of -maltivec=be.
222 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
225 * config/rs6000/rs6000-c.c: Update definitions for
226 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
227 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
229 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
232 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
233 add _get_ssp intrinsics. Remove argument from
234 __builtin_ia32_rdssp[d|q].
235 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
236 * config/i386/i386-builtin.def: Remove argument from
237 __builtin_ia32_rdssp[d|q].
238 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
239 ix86_expand_special_args_builtin for _rdssp[d|q].
240 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
241 Clear register before usage.
242 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
243 Add documentation for new _get_ssp and _inc_ssp intrinsics.
245 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
247 PR tree-optimization/84357
248 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
249 operand 1 of an ARRAY_REF too.
251 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
254 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
255 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
257 (set_of_reg): New struct.
258 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
259 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
260 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
262 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
263 Split into bitclr, bitset, bitinvert patterns if appropriate.
264 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
265 use rx_fuse_in_memory_bitop.
266 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
267 to named insn, correct maximum insn length.
269 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
272 * machmode.def: Define a complex mode for PARTIAL_INT.
273 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
275 * doc/rtl.texi: Document CSPImode.
276 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
278 (msp430_hard_regno_nregs_with_padding): Likewise.
280 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
283 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
285 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
287 PR rtl-optimization/84169
288 * combine.c (try_combine): New variable split_i2i3. Set it to true if
289 we generated a parallel as new i3 and we split that to new i2 and i3
290 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
291 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
292 those to i2, not i1. Partially rewrite this scan code.
294 2018-02-13 Jakub Jelinek <jakub@redhat.com>
297 * stor-layout.c (place_field): For variable length fields, adjust
298 offset_align afterwards not just based on the field's alignment,
299 but also on the size.
302 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
303 of exps and logs in the use_exp2 case.
305 2018-02-13 Jeff Law <law@redhat.com>
307 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
310 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
313 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
317 * common.opt (gas-loc-support, gas-locview-support): New.
318 (ginline-points, ginternal-reset-location-views): New.
319 * doc/invoke.texi: Document them. Use @itemx where intended.
320 (gvariable-location-views): Adjust.
321 * target.def (reset_location_view): New.
322 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
323 (TARGET_RESET_LOCATION_VIEW): New.
324 * doc/tm.texi: Rebuilt.
325 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
326 (dwarf2out_default_as_locview_support): New.
327 (output_asm_line_debug_info): Use option variables.
328 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
329 (output_loc_list): Likewise.
330 (add_high_low_attributes): Check option variables.
331 Don't output entry view attribute in strict mode.
332 (gen_inlined_subroutine_die): Check option variables.
333 (dwarf2out_inline_entry): Likewise.
334 (init_sections_and_labels): Likewise.
335 (dwarf2out_early_finish): Likewise.
336 (maybe_reset_location_view): New, from...
337 (dwarf2out_var_location): ... here. Call it.
338 * debug.h (dwarf2out_default_as_loc_support): Declare.
339 (dwarf2out_default_as_locview_support): Declare.
340 * hooks.c (hook_int_rtx_insn_0): New.
341 * hooks.h (hook_int_rtx_insn_0): Declare.
342 * toplev.c (process_options): Take -gas-loc-support and
343 -gas-locview-support from dwarf2out. Enable
344 -gvariable-location-views by default only with locview
345 assembler support. Enable -ginternal-reset-location-views by
346 default only if the target defines the corresponding hook.
347 Enable -ginline-points by default if location views are
348 enabled; force it disabled if statement frontiers are
350 * tree-inline.c (expand_call_inline): Check option variables.
351 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
353 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
355 PR tree-optimization/84321
356 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
357 handling. Also check whether the anti-range contains any values
358 that satisfy the mask; switch to a VR_RANGE if not.
360 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
363 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
365 2018-02-13 Martin Jambor <mjambor@suse.cz>
368 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
369 of call statements, also set location of a load to a temporary.
371 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
373 * config/rl78/rl78.c (add_vector_labels): New function.
374 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
375 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
376 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
377 which checks that no arguments are passed.
378 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
379 * doc/extend.texi: Documentation for the new attribute.
381 2018-02-13 Andreas Schwab <schwab@suse.de>
383 * config/riscv/linux.h (CPP_SPEC): Define.
385 2018-02-13 Jakub Jelinek <jakub@redhat.com>
388 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
389 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
390 OPTION_MASK_ISA_AES as first argument to def_builtin_const
391 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
392 instead of OPTION_MASK_ISA_PCLMUL as first argument to
393 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
394 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
395 temporarily for AES and PCLMUL builtins.
397 PR tree-optimization/84339
398 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
399 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
403 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
404 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
405 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
407 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
409 * omp-simd-clone.h: New file.
410 * omp-simd-clone.c: Include omp-simd-clone.h.
411 (expand_simd_clones): No longer static.
412 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
413 cgraph.h and omp-simd-clone.h.
414 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
415 (vect_recog_widen_shift_pattern): Formatting fix.
416 (vect_pattern_recog_1): Don't check optab for calls.
419 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
420 operands[2] into a REG before using gen_lowpart on it.
422 2018-02-12 Jeff Law <law@redhat.com>
425 * config/sh/sh.c (find_barrier): Consider a sibling call
428 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
429 successfully back substituting a reg.
431 2018-02-12 Richard Biener <rguenther@suse.de>
433 PR tree-optimization/84037
434 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
435 parameter, move visited init to caller.
436 (vect_slp_analyze_operations): Separate cost from validity
437 check, initialize visited once for all instances.
438 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
440 * tree-vect-stmts.c (vect_model_simple_cost): Make early
442 (vect_model_promotion_demotion_cost): Likewise.
443 (vectorizable_bswap): Guard cost modeling with !slp_node
444 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
446 (vectorizable_call): Likewise.
447 (vectorizable_conversion): Likewise.
448 (vectorizable_assignment): Likewise.
449 (vectorizable_shift): Likewise.
450 (vectorizable_operation): Likewise.
451 (vectorizable_store): Likewise.
452 (vectorizable_load): Likewise.
453 (vectorizable_condition): Likewise.
454 (vectorizable_comparison): Likewise.
456 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
459 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
460 (ASAN_MARK): Fix fnspec to account for return value, change pointer
461 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
463 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
466 * params.def (inline-min-speedup): Increase from 8 to 15.
467 (max-inline-insns-auto): Decrease from 40 to 30.
468 * ipa-split.c (consider_split): Add some buffer for function to
469 be considered inlining candidate.
470 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
473 2018-02-12 Richard Biener <rguenther@suse.de>
475 PR tree-optimization/84037
476 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
477 matched stmts if we cannot swap the non-matched ones.
479 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
481 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
482 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
483 _mm_maskz_scalef_round_ss): New intrinsics.
484 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
485 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
486 __builtin_ia32_scalefss_round): Remove.
487 (__builtin_ia32_scalefsd_mask_round,
488 __builtin_ia32_scalefss_mask_round): New intrinsics.
489 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
490 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
491 ((match_operand:VF_128 2 "<round_nimm_predicate>"
492 "<round_constraint>")): Changed to ...
493 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
494 "<round_scalar_constraint>")): ... this.
495 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
496 %0, %1, %2<round_op3>}"): Changed to ...
497 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
498 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
499 %2<round_scalar_mask_op3>}"): ... this.
500 * config/i386/subst.md (round_scalar_nimm_predicate): New.
502 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
504 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
505 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
506 (_mm_maskz_sqrt_round_ss): New intrinsics.
507 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
508 (__builtin_ia32_sqrtsd_mask_round)
509 (__builtin_ia32_sqrtss_mask_round): New builtins.
510 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
511 (__builtin_ia32_sqrtss_round): Remove.
512 (__builtin_ia32_sqrtsd_mask_round)
513 (__builtin_ia32_sqrtss_mask_round): New builtins.
514 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
515 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
516 ((match_operand:VF_128 1 "vector_operand"
517 "xBm,<round_constraint>")): Changed to ...
518 ((match_operand:VF_128 1 "vector_operand"
519 "xBm,<round_scalar_constraint>")): ... this.
520 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
521 %0, %2, %<iptr>1<round_op3>}): Changed to ...
522 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
523 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
524 %<iptr>1<round_scalar_mask_op3>}): ... this.
525 ((set_attr "prefix" "<round_prefix>")): Changed to ...
526 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
528 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
531 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
532 Cast vec_cmpeq result to correct type.
533 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
534 Cast vec_cmpgt result to correct type.
536 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
538 * final.c (final_scan_insn_1): Renamed from...
539 (final_scan_insn): ... this. New wrapper, to recover
540 seen from the outermost call in recursive ones.
541 * config/sparc/sparc.c (output_return): Drop seen from call.
542 (output_sibcall): Likewise.
543 * config/visium/visium.c (output_branch): Likewise.
545 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
547 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
550 2018-02-10 Alan Modra <amodra@gmail.com>
553 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
554 Specify LR as an input.
556 2018-02-10 Jakub Jelinek <jakub@redhat.com>
559 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
560 remove_member_access_dummy_vars): New functions.
561 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
562 lower_omp_1, execute_lower_omp): Use them.
564 PR rtl-optimization/84308
565 * shrink-wrap.c (spread_components): Release todo vector.
567 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
569 PR rtl-optimization/57193
570 * ira-color.c (struct allocno_color_data): Add member
571 conflict_allocno_hard_prefs.
572 (update_conflict_allocno_hard_prefs): New.
573 (bucket_allocno_compare_func): Add a preference based on
574 conflict_allocno_hard_prefs.
575 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
576 (color_allocnos): Remove a dead code. Initiate
577 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
579 2018-02-09 Jakub Jelinek <jakub@redhat.com>
582 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
583 constraint from =wa to wa. Avoid a subreg on the output operand,
584 instead use a pseudo and subreg it in a move.
585 (p9_xxbrd_<mode>): Changed to ...
586 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
587 (p9_xxbrd_v2df): New expander.
588 (p9_xxbrw_<mode>): Changed to ...
589 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
590 (p9_xxbrw_v4sf): New expander.
592 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
594 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
596 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
599 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
600 multiply in 32-bit mode.
601 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
602 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
605 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
607 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
608 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
609 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
610 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
612 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
615 * dwarf2out.c (is_trivial_indirect_ref): New function.
616 (dwarf2out_late_global_decl): Do not generate a location
617 attribute for variables that have a non-trivial DECL_VALUE_EXPR
618 and that are not defined in the current unit.
620 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
622 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
623 instead of a libcall for UNORDERED.
625 2018-02-09 Tamar Christina <tamar.christina@arm.com>
628 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
629 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
631 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
634 * config/s390/s390.c (s390_set_current_function): Invoke
635 s390_indirect_branch_settings also if fndecl didn't change.
637 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
639 * config/rs6000/rs6000.md (blockage): Set length to zero.
641 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
643 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
645 2018-02-09 Jakub Jelinek <jakub@redhat.com>
648 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
649 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
653 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
654 PARALLEL incoming that failed vt_get_decl_and_offset check.
657 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
658 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
660 (get_variable_section): For decls in named .bss* sections pass true as
661 second argument to bss_initializer_p.
663 2018-02-09 Marek Polacek <polacek@redhat.com>
664 Jakub Jelinek <jakub@redhat.com>
667 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
668 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
669 Sync some changes from cxx_fold_indirect_ref.
671 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
673 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
675 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
676 (BLOCK_INLINE_ENTRY_LABEL): New.
677 (dwarf2out_var_location): Disregard inline entry markers.
678 (inline_entry_data): New struct.
679 (inline_entry_data_hasher): New hashtable type.
680 (inline_entry_data_hasher::hash): New.
681 (inline_entry_data_hasher::equal): New.
682 (inline_entry_data_table): New variable.
683 (add_high_low_attributes): Add DW_AT_entry_pc and
684 DW_AT_GNU_entry_view attributes if a pending entry is found
685 in inline_entry_data_table. Add old entry_pc attribute only
686 if debug nonbinding markers are disabled.
687 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
689 (block_within_block_p, dwarf2out_inline_entry): New.
690 (dwarf2out_finish): Check that no entries remained in
691 inline_entry_data_table.
692 * final.c (reemit_insn_block_notes): Handle inline entry notes.
693 (final_scan_insn, notice_source_line): Likewise.
694 (rest_of_clean_state): Skip inline entry markers.
695 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
697 * gimple.c (gimple_build_debug_inline_entry): New.
698 * gimple.h (enum gimple_debug_subcode): Add
699 GIMPLE_DEBUG_INLINE_ENTRY.
700 (gimple_build_debug_inline_entry): Declare.
701 (gimple_debug_inline_entry_p): New.
702 (gimple_debug_nonbind_marker_p): Adjust.
703 * insn-notes.def (INLINE_ENTRY): New.
704 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
705 inline entry marker notes.
706 (print_insn): Likewise.
707 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
708 (INSN_DEBUG_MARKER_KIND): Likewise.
709 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
710 * tree-inline.c (expand_call_inline): Build and insert
711 debug_inline_entry stmt.
712 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
713 inline entry blocks early, if nonbind markers are enabled.
714 (dump_scope_block): Dump fragment info.
715 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
716 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
717 (gimple_build_debug_inline_entry): New.
718 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
719 Enable/disable inline entry points too.
720 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
721 (DEBUG_INSN): Describe inline entry markers.
723 * common.opt (gvariable-location-views): New.
724 (gvariable-location-views=incompat5): New.
725 * config.in: Rebuilt.
726 * configure: Rebuilt.
727 * configure.ac: Test assembler for view support.
728 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
729 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
730 * dwarf2out.c (var_loc_view): New typedef.
731 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
732 (dwarf2out_locviews_in_attribute): New.
733 (dwarf2out_locviews_in_loclist): New.
734 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
735 (enum dw_line_info_opcode): Add LI_adv_address.
736 (struct dw_line_info_table): Add view.
737 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
738 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
739 (zero_view_p): New variable.
740 (ZERO_VIEW_P): New macro.
741 (output_asm_line_debug_info): New.
742 (struct var_loc_node): Add view.
743 (add_AT_view_list, AT_loc_list): New.
744 (add_var_loc_to_decl): Add view param. Test it against last.
745 (new_loc_list): Add view params. Record them.
746 (AT_loc_list_ptr): Handle loc and view lists.
747 (view_list_to_loc_list_val_node): New.
748 (print_dw_val): Handle dw_val_class_view_list.
749 (size_of_die): Likewise.
750 (value_format): Likewise.
751 (loc_list_has_views): New.
752 (gen_llsym): Set vl_symbol too.
753 (maybe_gen_llsym, skip_loc_list_entry): New.
754 (dwarf2out_maybe_output_loclist_view_pair): New.
755 (output_loc_list): Output view list or entries too.
756 (output_view_list_offset): New.
757 (output_die): Handle dw_val_class_view_list.
758 (output_dwarf_version): New.
759 (output_compilation_unit_header): Use it.
760 (output_skeleton_debug_sections): Likewise.
761 (output_rnglists, output_line_info): Likewise.
762 (output_pubnames, output_aranges): Update version comments.
763 (output_one_line_info_table): Output view numbers in asm comments.
764 (dw_loc_list): Determine current endview, pass it to new_loc_list.
765 Call maybe_gen_llsym.
766 (loc_list_from_tree_1): Adjust.
767 (add_AT_location_description): Create view list attribute if
768 needed, check it's absent otherwise.
769 (convert_cfa_to_fb_loc_list): Adjust.
770 (maybe_emit_file): Call output_asm_line_debug_info for test.
771 (dwarf2out_var_location): Reset views as needed. Precompute
772 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
774 (new_line_info_table): Reset next view.
775 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
776 (dwarf2out_source_line): Likewise. Output view resets and labels to
777 the assembler, or select appropriate line info opcodes.
778 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
779 (optimize_string_length): Catch it. Adjust.
780 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
781 dw_val_class_view_list, and remove it if no longer needed.
782 (hash_loc_list): Hash view numbers.
783 (loc_list_hasher::equal): Compare them.
784 (optimize_location_lists): Check whether a view list symbol is
785 needed, and whether the locview attribute is present, and
786 whether they match. Remove the locview attribute if no longer
788 (index_location_lists): Call skip_loc_list_entry for test.
789 (dwarf2out_finish): Call output_asm_line_debug_info for test.
790 Use output_dwarf_version.
791 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
792 (struct dw_val_node): Add val_view_list.
793 * final.c (SEEN_NEXT_VIEW): New.
794 (set_next_view_needed): New.
795 (clear_next_view_needed): New.
796 (maybe_output_next_view): New.
797 (final_start_function): Rename to...
798 (final_start_function_1): ... this. Take pointer to FIRST,
799 add SEEN parameter. Emit param bindings in the initial view.
800 (final_start_function): Reintroduce SEEN-less interface.
801 (final): Rename to...
802 (final_1): ... this. Take SEEN parameter. Output final pending
803 next view at the end.
804 (final): Reintroduce seen-less interface.
805 (final_scan_insn): Output pending next view before switching
806 sections or ending a block. Mark the next view as needed when
807 outputting variable locations. Notify debug backend of section
808 changes, and of location view changes.
809 (rest_of_handle_final): Adjust.
810 * toplev.c (process_options): Autodetect value for debug variable
811 location views option. Warn on incompat5 without -gdwarf-5.
812 * doc/invoke.texi (gvariable-location-views): New.
813 (gvariable-location-views=incompat5): New.
814 (gno-variable-location-views): New.
816 2018-02-08 David Malcolm <dmalcolm@redhat.com>
818 PR tree-optimization/84136
819 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
820 that the result of find_edge is non-NULL.
822 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
825 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
826 storing integer register in SImode. Fix cost of 256 and 512
827 byte aligned SSE register store.
829 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
831 * config/i386/i386.c (ix86_multiplication_cost): Fix
832 multiplication cost for TARGET_AVX512DQ.
834 2018-02-08 Marek Polacek <polacek@redhat.com>
836 PR tree-optimization/84238
837 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
840 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
842 PR tree-optimization/84265
843 * tree-vect-stmts.c (vectorizable_store): Don't treat
844 VMAT_CONTIGUOUS accesses as grouped.
845 (vectorizable_load): Likewise.
847 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
849 PR tree-optimization/81635
850 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
851 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
852 (test_round_for_mask): New functions.
853 (wide_int_cc_tests): Call test_round_for_mask.
854 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
855 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
856 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
857 range returned by get_range_info.
859 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
862 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
863 * symtab.c: Include builtins.h
864 (symtab_node::output_to_lto_symbol_table_p): Move here
865 from lto-streamer-out.c:output_symbol_p.
866 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
867 (output_symbol_p): Move all logic to symtab.c
868 (produce_symtab): Update.
870 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
872 * config/s390/s390-opts.h (enum indirect_branch): Define.
873 * config/s390/s390-protos.h (s390_return_addr_from_memory)
874 (s390_indirect_branch_via_thunk)
875 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
876 (enum s390_indirect_branch_type): Define.
877 * config/s390/s390.c (struct s390_frame_layout, struct
878 machine_function): Remove.
879 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
880 (indirect_branch_table_label_no, indirect_branch_table_name):
882 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
883 (enum s390_indirect_branch_option): Define.
884 (s390_return_addr_from_memory): New function.
885 (s390_handle_string_attribute): New function.
886 (s390_attribute_table): Add new attribute handler.
887 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
888 (s390_indirect_branch_via_thunk): New function.
889 (s390_indirect_branch_via_inline_thunk): New function.
890 (s390_function_ok_for_sibcall): When jumping via thunk disallow
891 sibling call optimization for non z10 compiles.
892 (s390_emit_call): Force indirect branch target to be a single
893 register. Add r1 clobber for non-z10 compiles.
894 (s390_emit_epilogue): Emit return jump via return_use expander.
895 (s390_reorg): Handle JUMP_INSNs as execute targets.
896 (s390_option_override_internal): Perform validity checks for the
897 new command line options.
898 (s390_indirect_branch_attrvalue): New function.
899 (s390_indirect_branch_settings): New function.
900 (s390_set_current_function): Invoke s390_indirect_branch_settings.
901 (s390_output_indirect_thunk_function): New function.
902 (s390_code_end): Implement target hook.
903 (s390_case_values_threshold): Implement target hook.
904 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
906 * config/s390/s390.h (struct s390_frame_layout)
907 (struct machine_function): Move here from s390.c.
908 (TARGET_INDIRECT_BRANCH_NOBP_RET)
909 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
910 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
911 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
912 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
913 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
914 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
915 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
916 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
917 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
918 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
919 (mnemonic attribute): Add values which aren't recognized
921 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
922 pattern for branch conversion. Fix mnemonic attribute.
923 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
924 indirect branch via thunk if requested.
925 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
926 ("*indirect_jump"): Disable for branch conversion using out of
928 ("indirect_jump_via_thunk<mode>_z10")
929 ("indirect_jump_via_thunk<mode>")
930 ("indirect_jump_via_inlinethunk<mode>_z10")
931 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
932 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
933 ("casesi_jump_via_inlinethunk<mode>_z10")
934 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
935 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
936 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
937 ("*indirect2_jump"): Disable for branch conversion.
938 ("casesi_jump"): Turn into expander and expand patterns for branch
940 ("return_use"): New expander.
941 ("*return"): Emit return via thunk and rename it to ...
942 ("*return<mode>"): ... this one.
943 * config/s390/s390.opt: Add new options and and enum for the
946 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
948 * lra-constraints.c (match_reload): Unconditionally use
949 gen_lowpart_SUBREG, rather than selecting between that
950 and equivalent gen_rtx_SUBREG code.
952 2018-02-08 Richard Biener <rguenther@suse.de>
954 PR tree-optimization/84233
955 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
956 changed flag instead of boguously re-using phi_inserted.
958 2018-02-08 Martin Jambor <mjambor@suse.cz>
960 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
961 static local variables.
963 2018-02-08 Richard Biener <rguenther@suse.de>
965 PR tree-optimization/84278
966 * tree-vect-stmts.c (vectorizable_store): When looking for
967 smaller vector types to perform grouped strided loads/stores
968 make sure the mode is supported by the target.
969 (vectorizable_load): Likewise.
971 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
973 * config/aarch64/aarch64.c (aarch64_components_for_bb):
974 Increase LDP/STP opportunities by adding adjacent callee-saves.
976 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
978 PR rtl-optimization/84068
979 PR rtl-optimization/83459
980 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
982 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
984 PR tree-optimization/84224
985 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
986 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
989 2018-02-07 Iain Sandoe <iain@codesourcery.com>
992 * config/rs6000/altivec.md (*restore_world): Remove LR use.
993 * config/rs6000/predicates.md (restore_world_operation): Adjust op
994 count, remove one USE.
996 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
998 * doc/install.texi (Configuration): Document the
999 --with-long-double-format={ibm,ieee} PowerPC configuration
1003 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
1004 Convert from define_expand to be define_insn_and_split. Rework
1005 float/double/_Float128 conversions to QI/HI/SImode to work with
1006 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
1007 conversions to QI/HImode types did a store and then a load to
1008 truncate the value. For conversions to VSX registers, don't split
1009 the insn, instead emit the code directly. Use the code iterator
1010 any_fix to combine signed and unsigned conversions.
1011 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
1012 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
1013 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1014 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
1015 (fix_<mode>di2_hw): Likewise.
1016 (fixuns_<mode>di2_hw): Likewise.
1017 (fix_<mode>si2_hw): Likewise.
1018 (fixuns_<mode>si2_hw): Likewise.
1019 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
1020 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1021 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
1022 fix<uns>_trunc<SFDF:mode>si2_p8.
1023 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
1025 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1026 (fix<uns>_<mode>_mem): Likewise.
1027 (fctiw<u>z_<mode>_mem): Likewise.
1028 (fix<uns>_<mode>_mem): Likewise.
1029 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1030 the register allocator from doing a direct move to the GPRs to do
1031 a store, and instead use the ISA 3.0 store byte/half-word from
1032 vector register instruction. For IEEE 128-bit floating point,
1033 also optimize stores of 32-bit ints.
1034 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1036 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1038 * genextract.c (push_pathstr_operand): New function to support
1040 (walk_rtx): Call push_pathstr_operand.
1041 (print_path): Support [a-zA-Z].
1043 2018-02-07 Richard Biener <rguenther@suse.de>
1045 PR tree-optimization/84037
1046 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1047 (cse_and_gimplify_to_preheader): Declare.
1048 (vect_get_place_in_interleaving_chain): Likewise.
1049 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1051 (_loop_vec_info::~_loop_vec_info): Delete it.
1052 (cse_and_gimplify_to_preheader): New function.
1053 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1054 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1055 (vectorizable_load): Likewise. For grouped stores always base
1056 the IV on the first element.
1057 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1058 condition before gimplifying.
1060 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1062 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1063 *DIV_EXPR and *MOD_EXPR.
1065 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1068 * config/i386/i386.c (ix86_option_override_internal): Mask out
1069 the CF_SET bit when checking -fcf-protection.
1071 2018-02-07 Tom de Vries <tom@codesourcery.com>
1074 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1077 2018-02-07 Richard Biener <rguenther@suse.de>
1079 PR tree-optimization/84204
1080 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1083 PR tree-optimization/84205
1084 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1085 special-case isl_ast_op_zdiv_r.
1087 PR tree-optimization/84223
1088 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1089 Only add conditions from within the region.
1090 (gather_bbs::after_dom_children): Adjust.
1092 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1095 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1096 * config/avr/avr.md: Only post-reload split REG-REG moves if
1097 either register is REGERAL_REG_P.
1099 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1101 PR tree-optimization/84235
1102 * tree-ssa-scopedtables.c
1103 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1104 if the subtraction is performed in floating point type where NaNs are
1105 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1106 build 1. Formatting fix.
1108 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1111 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1112 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1113 and skip it regardless of bb boundaries. Use CALL_P macro,
1114 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1115 unnecessarily, formatting fix.
1117 2018-02-06 Michael Collison <michael.collison@arm.com>
1119 * config/arm/thumb2.md:
1120 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1121 (*thumb_mov_notscc): Ditto.
1123 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1126 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1127 unsigned_fix, not "s".
1129 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1131 * configure.ac (gcc_fn_eh_frame_ro): New function.
1132 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1133 correct .eh_frame permissions.
1134 * configure: Regenerate.
1136 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1138 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1141 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1143 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1144 Display warning message for -mno-speculate-indirect-jumps.
1146 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1148 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1150 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1152 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1154 PR tree-optimization/84225
1155 * tree-eh.c (find_trapping_overflow): Only call
1156 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1158 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1161 * config/i386/i386.c: Reimplement the check of possible options
1162 -mibt/-mshstk conbination. Change error messages.
1163 * doc/invoke.texi: Fix a typo: remove extra '='.
1165 2018-02-06 Marek Polacek <polacek@redhat.com>
1167 PR tree-optimization/84228
1168 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1170 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1173 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1174 emitted arch directives.
1175 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1176 __ARM_FEATURE_COPROC before changing architectures.
1178 2018-02-06 Richard Biener <rguenther@suse.de>
1180 * config/i386/i386.c (print_reg): Fix typo.
1181 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1183 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1185 * configure: Regenerate.
1187 2018-02-05 Martin Sebor <msebor@redhat.com>
1189 PR tree-optimization/83369
1190 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1193 2018-02-05 Martin Liska <mliska@suse.cz>
1195 * doc/invoke.texi: Cherry-pick upstream r323995.
1197 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1199 * ira.c (ira_init_register_move_cost): Adjust comment.
1201 2018-02-05 Martin Liska <mliska@suse.cz>
1203 PR gcov-profile/84137
1204 * doc/gcov.texi: Fix typo in documentation.
1206 2018-02-05 Martin Liska <mliska@suse.cz>
1208 PR gcov-profile/83879
1209 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1210 using dlopen functionality.
1212 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1214 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1215 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1216 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1217 _mm_maskz_range_round_ss): New intrinsics.
1218 (__builtin_ia32_rangesd128_round)
1219 (__builtin_ia32_rangess128_round): Remove.
1220 (__builtin_ia32_rangesd128_mask_round,
1221 __builtin_ia32_rangess128_mask_round): New builtins.
1222 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1223 __builtin_ia32_rangess128_round): Remove.
1224 (__builtin_ia32_rangesd128_mask_round,
1225 __builtin_ia32_rangess128_mask_round): New builtins.
1226 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1227 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1228 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1229 "<round_saeonly_constraint>")): Changed to ...
1230 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1231 "<round_saeonly_scalar_constraint>")): ... this.
1232 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1233 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1234 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1235 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1236 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1238 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1240 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1242 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1243 Remove all values except native, 8540 and 8548.
1245 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1247 * config/i386/i386.c (ix86_output_function_return): Pass
1248 INVALID_REGNUM, instead of -1, as invalid register number to
1249 indirect_thunk_name and output_indirect_thunk.
1251 2018-02-02 Julia Koval <julia.koval@intel.com>
1253 * config.gcc: Add -march=icelake.
1254 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1255 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1256 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1257 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1258 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1259 (processor_target_table): Add icelake.
1260 (ix86_option_override_internal): Handle new PTAs.
1261 (get_builtin_code_for_version): Handle icelake.
1262 (M_INTEL_COREI7_ICELAKE): New.
1263 (fold_builtin_cpu): Handle icelake.
1264 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1265 * doc/invoke.texi: Add -march=icelake.
1267 2018-02-02 Julia Koval <julia.koval@intel.com>
1269 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1270 to wide_int_bitmask.
1271 * wide-int-bitmask.h: New.
1273 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1276 * config/i386/i386.md: Replace Pmode with word_mode in
1277 builtin_setjmp_setup and builtin_longjmp to support x32.
1279 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1283 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1285 (rs6000_supported_cpu_names): New static variable.
1286 (linux_cpu_translation_table): Likewise.
1287 (elf_platform) <cpu>: Define new static variable and use it.
1288 Translate kernel AT_PLATFORM name to canonical name if needed.
1289 Error if platform name is unknown.
1291 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1294 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1296 2018-02-01 Jeff Law <law@redhat.com>
1299 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1300 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1301 the scratch if RELEASE_VIA_POP is false.
1302 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1303 If we have to save a temporary register, decrement SIZE appropriately.
1304 Pass new arguments to release_scratch_register_on_entry.
1305 (ix86_adjust_stack_and_probe): Likewise.
1306 (ix86_emit_probe_stack_range): Pass new arguments to
1307 release_scratch_register_on_entry.
1309 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1311 PR rtl-optimization/84157
1312 * combine.c (change_zero_ext): Use REG_P predicate in
1313 front of HARD_REGISTER_P predicate.
1315 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1317 * config/avr/avr.c (avr_option_override): Move disabling of
1318 -fdelete-null-pointer-checks to...
1319 * common/config/avr/avr-common.c (avr_option_optimization_table):
1322 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1324 PR tree-optimization/81635
1325 * tree-data-ref.c (split_constant_offset_1): For types that
1326 wrap on overflow, try to use range info to prove that wrapping
1329 2018-02-01 Renlin Li <renlin.li@arm.com>
1332 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1334 (aarch64_register_move_cost): Likewise.
1335 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1337 (REG_CLASS_NAMES): Likewise.
1338 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1339 TAILCALL_ADDR_REGS. Remove IP registers.
1340 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1342 2018-02-01 Richard Biener <rguenther@suse.de>
1344 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1345 for specifying RPO order and allow NULL for that.
1346 * domwalk.c (dom_walker::dom_walker): Likewise.
1347 (dom_walker::walk): Handle NULL RPO order.
1348 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1350 (rewrite_update_dom_walker): Likewise.
1351 (mark_def_dom_walker): Likewise.
1353 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1355 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1356 (aarch64_maybe_expand_sve_subreg_move): Declare.
1357 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1358 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1360 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1361 that are semantically a reverse operation.
1362 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1363 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1364 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1366 (aarch64_can_change_mode_class): For big-endian, forbid changes
1367 between two SVE modes if they have different element sizes.
1369 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1371 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1372 the TImode handling for big-endian targets.
1374 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1376 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1377 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1379 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1380 Remove BSWAP handing for big-endian targets and use the form of
1381 LD1RQ appropariate for the mode.
1383 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1385 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1386 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1389 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1392 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1393 check for operands that need to go through aarch64_sve_reload_be.
1395 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1397 PR tree-optimization/81661
1398 PR tree-optimization/84117
1399 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1400 * tree-eh.c: Include gimplify.h.
1401 (find_trapping_overflow, replace_trapping_overflow,
1402 rewrite_to_non_trapping_overflow): New functions.
1403 * tree-vect-loop.c: Include tree-eh.h.
1404 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1405 * tree-data-ref.c: Include tree-eh.h.
1406 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1408 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1410 PR rtl-optimization/84123
1411 * combine.c (change_zero_ext): Check if hard register satisfies
1412 can_change_dest_mode before calling gen_lowpart_SUBREG.
1414 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1417 * ira.c (ira_init_register_move_cost): Remove assert.
1419 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1421 PR rtl-optimization/84071
1422 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1423 * doc/tm.texi: Regenerate.
1425 2018-01-31 Richard Biener <rguenther@suse.de>
1427 PR tree-optimization/84132
1428 * tree-data-ref.c (analyze_miv_subscript): Properly
1429 check whether evolution_function_is_affine_multivariate_p
1430 before calling gcd_of_steps_may_divide_p.
1432 2018-01-31 Julia Koval <julia.koval@intel.com>
1435 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1436 * config/i386/i386.md (rdpid_rex64) New.
1437 (rdpid): Make 32bit only.
1439 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1442 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1443 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1445 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1448 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1450 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1452 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1454 PR rtl-optimization/84071
1455 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1456 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1458 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1460 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1461 (arc_attribute_table): Add 'aux' attribute.
1462 (arc_in_small_data_p): Consider aux like variables.
1463 (arc_is_aux_reg_p): New function.
1464 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1465 (arc_get_aux_arg): New function.
1466 (prepare_move_operands): Handle aux-register access.
1467 (arc_handle_aux_attribute): New function.
1468 * doc/extend.texi (ARC Variable attributes): Add subsection.
1470 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1472 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1473 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1474 (arc_attribute_table): Add 'uncached' attribute.
1475 (arc_print_operand): Print '.di' flag for uncached memory
1477 (arc_in_small_data_p): Do not consider for small data the uncached
1479 (arc_is_uncached_mem_p): New function.
1480 * config/arc/predicates.md (compact_store_memory_operand): Check
1481 for uncached memory accesses.
1482 (nonvol_nonimm_operand): Likewise.
1483 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1485 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1488 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1489 falign-loops=): Add Optimization flag.
1491 2018-01-30 Jeff Law <law@redhat.com>
1494 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1495 INT_REGISTERS_SAVED. Check it prior to calling
1496 get_scratch_register_on_entry.
1497 (ix86_adjust_stack_and_probe): Similarly.
1498 (ix86_emit_probe_stack_range): Similarly.
1499 (ix86_expand_prologue): Corresponding changes.
1501 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1504 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1505 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1507 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1510 * lra-constraints.c (curr_insn_transform): Process AND in the
1513 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1515 PR rtl-optimization/83986
1516 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1517 dependence against last_pending_memory_flush in addition to
1520 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1522 PR tree-optimization/81611
1523 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1526 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1529 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1532 2018-01-30 Richard Biener <rguenther@suse.de>
1533 Jakub Jelinek <jakub@redhat.com>
1535 PR tree-optimization/84111
1536 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1537 inner loops added during recursion, as they don't have up-to-date
1540 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1543 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1544 (can_inline_edge_by_limits_p): ... here.
1545 (can_early_inline_edge_p, check_callers,
1546 update_caller_keys, update_callee_keys, recursive_inlining,
1547 add_new_edges_to_heap, speculation_useful_p,
1548 inline_small_functions,
1549 inline_small_functions, flatten_function,
1550 inline_to_all_callers_1): Update.
1552 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1554 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1557 2018-01-30 Richard Biener <rguenther@suse.de>
1559 PR tree-optimization/83008
1560 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1561 invariant and constant vector uses in stmts when they need
1564 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1567 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1568 * configure: Regenerate.
1570 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1572 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1574 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1575 Use gen_rtx_REG rather than gen_lowpart.
1577 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1579 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1580 rather than 0 when creating partial subregs.
1582 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1584 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1587 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1590 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1591 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1592 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1593 flags. This restores the settings used before the 2017-07-24.
1594 Turning off pre increment/decrement/modify allows IVOPTS to
1595 optimize DF/SF loops where the index is an int.
1597 2018-01-29 Richard Biener <rguenther@suse.de>
1598 Kelvin Nilsen <kelvin@gcc.gnu.org>
1601 * tree-vect-stmts.c (vectorizable_call): Don't call
1602 targetm.vectorize_builtin_md_vectorized_function if callee is
1605 2018-01-22 Carl Love <cel@us.ibm.com>
1607 * doc/extend.tex: Fix typo in second arg in
1608 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1610 2018-01-29 Richard Biener <rguenther@suse.de>
1612 PR tree-optimization/84086
1613 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1614 (flush_ssaname_freelist): When SSA names were released reset
1615 the SCEV hash table.
1617 2018-01-29 Richard Biener <rguenther@suse.de>
1619 PR tree-optimization/84057
1620 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1621 removed paths when removing edges.
1623 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1625 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1626 -mfunction-return=@var{choice}.
1628 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1631 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1633 (layout::print_source_line): Likewise.
1634 (test_get_line_width_without_trailing_whitespace): Add test cases.
1636 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1639 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1642 2018-01-26 Jim Wilson <jimw@sifive.com>
1644 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1646 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1649 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1651 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1652 and CMP + SUB-immediate -> SUBS.
1654 2018-01-26 Martin Sebor <msebor@redhat.com>
1656 PR tree-optimization/83896
1657 * tree-ssa-strlen.c (get_string_len): Rename...
1658 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1659 Avoid assuming length is constant.
1660 (handle_char_store): Use HOST_WIDE_INT for string length.
1662 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1665 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1666 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1668 2018-01-26 Richard Biener <rguenther@suse.de>
1670 PR rtl-optimization/84003
1671 * dse.c (record_store): Only record redundant stores when
1672 the earlier store aliases at least all accesses the later one does.
1674 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1676 PR rtl-optimization/83985
1677 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1678 REG_CFA_RESTORE insns.
1679 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1680 REG_CFA_RESTORE insns here.
1683 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1684 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1686 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1688 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1689 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1690 (arc_init): Likewise.
1691 (arc_override_options): Likewise.
1692 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1694 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1696 * config/arc/arc.h (TARGET_DBNZ): Define.
1697 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1698 properly set the tune attribute.
1699 (dbnz): Use TARGET_DBNZ guard.
1700 * config/arc/arc.opt (mtune): Add core3 option.
1702 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1704 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1705 recognize new pic like addresses.
1706 (arc_delegitimize_address): Clean up.
1708 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1710 * config/arc/arc-arches.def: Option mrf16 valid for all
1712 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1713 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1714 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1715 * config/arc/arc-tables.opt: Regenerate.
1716 * config/arc/arc.c (arc_conditional_register_usage): Handle
1717 reduced register file case.
1718 (arc_file_start): Set must have build attributes.
1719 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1721 * config/arc/arc.opt (mrf16): Add new option.
1722 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1723 * config/arc/genmultilib.awk: Handle new mrf16 option.
1724 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1725 * config/arc/t-multilib: Regenerate.
1726 * doc/invoke.texi (ARC Options): Document mrf16 option.
1728 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1730 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1731 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1732 (arc_attribute_table): Add 'secure_call' attribute.
1733 (arc_print_operand): Print secure call operand.
1734 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1736 (arc_is_secure_call_p): New function. * config/arc/arc.md
1737 (call_i): Add support for sjli instruction.
1738 (call_value_i): Likewise.
1739 * config/arc/constraints.md (Csc): New constraint.
1741 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1742 John Eric Martin <John.Martin@emmicro-us.com>
1744 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1745 * config/arc/arc.c (_arc_jli_section): New struct.
1746 (arc_jli_section): New type.
1747 (rc_jli_sections): New static variable.
1748 (arc_handle_jli_attribute): New function.
1749 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1750 (arc_file_end): New function.
1751 (TARGET_ASM_FILE_END): Define.
1752 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1753 (arc_add_jli_section): New function.
1754 (jli_call_scan): Likewise.
1755 (arc_reorg): Call jli_call_scan.
1756 (arc_output_addsi): Remove 'S' from printing asm operand.
1757 (arc_is_jli_call_p): New function.
1758 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1760 (movhi_insn): Likewise.
1761 (movsi_insn): Likewise.
1762 (movsi_set_cc_insn): Likewise.
1763 (loadqi_update): Likewise.
1764 (load_zeroextendqisi_update): Likewise.
1765 (load_signextendqisi_update): Likewise.
1766 (loadhi_update): Likewise.
1767 (load_zeroextendhisi_update): Likewise.
1768 (load_signextendhisi_update): Likewise.
1769 (loadsi_update): Likewise.
1770 (loadsf_update): Likewise.
1771 (movsicc_insn): Likewise.
1772 (bset_insn): Likewise.
1773 (bxor_insn): Likewise.
1774 (bclr_insn): Likewise.
1775 (bmsk_insn): Likewise.
1776 (bicsi3_insn): Likewise.
1777 (cmpsi_cc_c_insn): Likewise.
1778 (movsi_ne): Likewise.
1779 (movsi_cond_exec): Likewise.
1780 (clrsbsi2): Likewise.
1791 (call_i): Remove 'S' asm letter, add jli instruction.
1792 (call_value_i): Likewise.
1793 * config/arc/arc.op (mjli-always): New option.
1794 * config/arc/constraints.md (Cji): New constraint.
1795 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1797 (subsf3_fpx): Likewise.
1798 (mulsf3_fpx): Likewise.
1799 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1801 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1803 * doc/invoke.texi (ARC): Document mjli-always option.
1805 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1807 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
1808 avoid addition with 0 and use incw and decw where possible.
1810 2018-01-26 Richard Biener <rguenther@suse.de>
1812 PR tree-optimization/81082
1813 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1814 association if it requires casting to unsigned.
1815 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1816 from fold_plusminus_mult_expr to catch important cases late when
1817 range info is available.
1819 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1821 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1822 * configure.ac (hidden_linkonce): New test.
1823 * configure: Regenerate.
1824 * config.in: Regenerate.
1826 2018-01-26 Julia Koval <julia.koval@intel.com>
1828 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1829 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1830 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1831 _mm_mask_bitshuffle_epi64_mask): Fix type.
1832 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1833 USI_FTYPE_V4DI_V4DI_USI): Remove.
1834 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1835 __builtin_ia32_vpshufbitqmb256_mask,
1836 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1837 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1838 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1840 2018-01-26 Alan Modra <amodra@gmail.com>
1843 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1844 UNSPEC_VBPERMQ. Sort other unspecs.
1846 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1848 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1850 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1853 * predict.c (drop_profile): Do not push/pop cfun; update also
1855 (handle_missing_profiles): Fix logic looking for zero profiles.
1857 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1860 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1861 on functions with #pragma omp declare simd or functions with simd
1863 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1864 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1865 Remove trailing \n from warning_at calls.
1867 2018-01-25 Tom de Vries <tom@codesourcery.com>
1870 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1871 for neutered workers.
1873 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1876 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1877 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1879 2018-01-24 Jeff Law <law@redhat.com>
1882 * i386.c (get_probe_interval): Move to earlier point.
1883 (ix86_compute_frame_layout): If -fstack-clash-protection and
1884 the frame is larger than the probe interval, then use pushes
1885 to save registers rather than reg->mem moves.
1886 (ix86_expand_prologue): Remove conditional for int_registers_saved
1889 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1892 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1893 min/max for never referenced object.
1895 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1898 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1900 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1901 attributes from DECL_ATTRIBUTES (decl) without affecting
1902 DECL_ATTRIBUTES (current_function_decl).
1903 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1904 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1906 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1908 PR tree-optimization/83979
1909 * fold-const.c (fold_comparison): Use constant_boolean_node
1910 instead of boolean_{true,false}_node.
1912 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1914 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1917 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1919 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1920 Simplify the clause that sets the length attribute.
1921 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1922 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1923 clause that sets the length attribute.
1924 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1926 2018-01-24 Tom de Vries <tom@codesourcery.com>
1929 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1930 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1931 Add strict parameter.
1932 (prevent_branch_around_nothing): Insert dummy insn between branch to
1933 label and label with no ptx insn inbetween.
1934 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1936 2018-01-24 Tom de Vries <tom@codesourcery.com>
1939 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1940 for neutered threads in warp.
1941 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1943 2018-01-24 Richard Biener <rguenther@suse.de>
1945 PR tree-optimization/83176
1946 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1949 2018-01-24 Richard Biener <rguenther@suse.de>
1951 PR tree-optimization/82819
1952 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1953 code generating pluses that are no-ops in the target precision.
1955 2018-01-24 Richard Biener <rguenther@suse.de>
1958 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1960 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1962 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1963 to merge probabilities.
1964 * predict.c (probably_never_executed): Also mark as cold functions
1965 with global 0 profile and guessed local profile.
1966 * profile-count.c (profile_probability::combine_with_count): New
1968 * profile-count.h (profile_probability::operator*,
1969 profile_probability::operator*=, profile_probability::operator/,
1970 profile_probability::operator/=): Reduce precision to adjusted
1971 and set value to guessed on contradictory divisions.
1972 (profile_probability::combine_with_freq): Remove.
1973 (profile_probability::combine_wiht_count): Declare.
1974 (profile_count::force_nonzero):: Set to adjusted.
1975 (profile_count::probability_in):: Set quality to adjusted.
1976 * tree-ssa-tail-merge.c (replace_block_by): Use
1979 2018-01-23 Andrew Waterman <andrew@sifive.com>
1980 Jim Wilson <jimw@sifive.com>
1982 * config/riscv/riscv.c (riscv_stack_boundary): New.
1983 (riscv_option_override): Set riscv_stack_boundary. Handle
1984 riscv_preferred_stack_boundary_arg.
1985 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1986 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1987 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1988 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1989 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1990 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1992 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1995 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1996 of struct ix86_frame.
1997 (ix86_expand_epilogue): Likewise. Add a local variable for
1998 the reg_save_offset field in struct ix86_frame.
2000 2018-01-23 Bin Cheng <bin.cheng@arm.com>
2002 PR tree-optimization/82604
2003 * tree-loop-distribution.c (enum partition_kind): New enum item
2004 PKIND_PARTIAL_MEMSET.
2005 (partition_builtin_p): Support above new enum item.
2006 (generate_code_for_partition): Ditto.
2007 (compute_access_range): Differentiate cases that equality can be
2008 proven at all loops, the innermost loops or no loops.
2009 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
2010 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
2011 (finalize_partitions, distribute_loop): Don't fuse partition of
2012 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
2013 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
2016 2018-01-23 Martin Liska <mliska@suse.cz>
2018 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
2019 order to ignore the predictor.
2020 (PRED_POLYMORPHIC_CALL): Likewise.
2021 (PRED_RECURSIVE_CALL): Likewise.
2023 2018-01-23 Martin Liska <mliska@suse.cz>
2025 * tree-profile.c (tree_profiling): Print function header to
2026 aware reader which function we are working on.
2027 * value-prof.c (gimple_find_values_to_profile): Do not print
2028 not interesting value histograms.
2030 2018-01-23 Martin Liska <mliska@suse.cz>
2032 * profile-count.h (enum profile_quality): Add
2033 profile_uninitialized as the first value. Do not number values
2034 as they are zero based.
2035 (profile_count::verify): Update sanity check.
2036 (profile_probability::verify): Likewise.
2038 2018-01-23 Nathan Sidwell <nathan@acm.org>
2040 * doc/invoke.texi (ffor-scope): Deprecate.
2042 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2044 PR tree-optimization/83510
2045 * domwalk.c (set_all_edges_as_executable): New function.
2046 (dom_walker::dom_walker): Convert bool param
2047 "skip_unreachable_blocks" to enum reachability. Move setup of
2048 edge flags to set_all_edges_as_executable and only do it when
2049 reachability is REACHABLE_BLOCKS.
2050 * domwalk.h (enum dom_walker::reachability): New enum.
2051 (dom_walker::dom_walker): Convert bool param
2052 "skip_unreachable_blocks" to enum reachability.
2053 (set_all_edges_as_executable): New decl.
2054 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2055 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2057 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2058 but converting true to REACHABLE_BLOCKS.
2059 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2061 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2062 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2063 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2065 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2066 if check_all_array_refs will be called.
2068 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2070 * tree.c (selftest::test_location_wrappers): Add more test
2073 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2075 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2076 (selftest::test_bit_in_range): Likewise.
2078 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2081 * doc/sourcebuild.texi (vect_float): Say that the selector
2082 only describes the situation when -funsafe-math-optimizations is on.
2083 (vect_float_strict): Document.
2085 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2087 PR tree-optimization/83965
2088 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2089 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2090 instead of checking only for a reduction.
2091 (vect_recog_widen_sum_pattern): Likewise.
2093 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2095 * predict.c (probably_never_executed): Only use precise profile info.
2096 (compute_function_frequency): Skip after inlining hack since we now
2097 have quality checking.
2099 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2101 * profile-count.h (profile_probability::very_unlikely,
2102 profile_probability::unlikely, profile_probability::even): Set
2103 precision to guessed.
2105 2018-01-23 Richard Biener <rguenther@suse.de>
2107 PR tree-optimization/83963
2108 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2109 Properly terminate dominator walk when crossing the exit edge not
2110 when visiting its source block.
2112 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2115 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2116 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2118 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2120 PR tree-optimization/83957
2121 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2122 semicolon after for body surrounded by braces.
2124 PR tree-optimization/83081
2125 * profile-count.h (profile_probability::split): New method.
2126 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2127 Use profile_probability::split.
2128 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2129 when splitting a single conditional jump into 2.
2131 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2133 PR tree-optimization/69452
2134 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2137 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2139 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2140 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2141 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2143 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2145 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2146 * config/rl78/rl78.md (movdi): New define_expand.
2147 * config/rl78/rl78.c (rl78_split_movdi): New function.
2149 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2152 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2154 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2155 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2156 128-bit to produce an UNSPEC move to get the double word with the
2157 signbit and then a shift directly to do signbit.
2158 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2159 implementation with a new version that just does either a direct
2160 move or a regular move. Move memory interface to separate insns.
2161 Move insns so they are next to the expander.
2162 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2163 with signbit move. Split big and little endian case.
2164 (signbit<mode>2_dm_mem_le): Likewise.
2165 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2166 (signbit<mode>2_dm2): Likewise.
2168 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2170 * config/rl78/rl78.md (anddi3): New define_expand.
2172 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2174 * config/rl78/rl78.md (umindi3): New define_expand.
2176 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2178 * config/rl78/rl78.md (smindi3): New define_expand.
2180 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2182 * config/rl78/rl78.md (smaxdi3): New define_expand.
2184 2018-01-22 Carl Love <cel@us.ibm.com>
2186 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2187 LVX_V1TI): Add macro expansion.
2188 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2189 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2190 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2191 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2192 Change check to determine if the instruction is a byte reversing
2193 entry. Fix typo in comment.
2194 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2195 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2196 Add def_builtin calls for new builtins.
2197 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2198 Add define_insn expansion.
2200 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2202 * config/rl78/rl78.md (umaxdi3): New define_expand.
2204 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2206 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2207 for non-QImode registers.
2209 2018-01-22 Richard Biener <rguenther@suse.de>
2211 PR tree-optimization/83963
2212 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2213 including the loop exit block.
2214 (scop_detection::merge_sese): Likewise.
2215 (scop_detection::add_scop): Do it here instead.
2217 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2219 * doc/sourcebuild.texi (arm_softfloat): Document.
2221 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2224 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2225 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2226 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2228 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2229 David Edelsohn <dje.gcc@gmail.com>
2232 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2233 Change "crset eq" to "crset 2".
2234 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2235 (*call_indirect_aix<mode>_nospec): Likewise.
2236 (*call_value_indirect_aix<mode>_nospec): Likewise.
2237 (*call_indirect_elfv2<mode>_nospec): Likewise.
2238 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2239 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2240 change assembly output from . to $.
2241 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2242 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2243 (*tablejump<mode>_internal1_nospec): Likewise.
2245 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2248 * config/sh/sh_optimize_sett_clrt.cc:
2249 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2251 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2253 PR tree-optimization/83940
2254 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2255 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2256 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2257 use it to pass back the definition type.
2258 (vect_check_store_rhs): Likewise rhs_dt_out.
2259 (vect_build_gather_load_calls): Add a mask_dt argument and use
2260 it instead of a call to vect_is_simple_use.
2261 (vectorizable_store): Update calls to vect_check_load_store_mask
2262 and vect_check_store_rhs. Use the dt returned by the latter instead
2263 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2264 instead of calls to vect_is_simple_use. Pass the scalar rather
2265 than the vector operand to vect_is_simple_use when handling
2266 second and subsequent copies of an rhs value.
2267 (vectorizable_load): Update calls to vect_check_load_store_mask
2268 and vect_build_gather_load_calls. Use the cached mask_dt and
2269 gs_info.offset_dt instead of calls to vect_is_simple_use.
2271 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2274 * tree-emutls.c: Include gimplify.h.
2275 (lower_emutls_2): New function.
2276 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2277 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2278 it before further processing.
2281 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2282 UINTVAL (trueop1) instead of INTVAL (op1).
2284 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2288 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2289 INCOMING_FRAME_SP_OFFSET if not defined.
2290 (scan_trace): Add ENTRY argument. If true and
2291 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2292 emit a note to adjust the CFA offset.
2293 (create_cfi_notes): Adjust scan_trace callers.
2294 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2295 INCOMING_FRAME_SP_OFFSET in the CIE.
2296 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2297 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2299 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2300 * doc/tm.texi: Regenerated.
2302 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2304 PR rtl-optimization/83147
2305 * lra-constraints.c (remove_inheritance_pseudos): Use
2306 lra_substitute_pseudo_within_insn.
2308 2018-01-19 Tom de Vries <tom@codesourcery.com>
2309 Cesar Philippidis <cesar@codesourcery.com>
2312 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2314 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2317 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2318 spaces for function labels.
2320 2018-01-19 Martin Liska <mliska@suse.cz>
2322 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2323 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2324 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2325 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2326 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2327 (PRED_CONST_RETURN): Change from 69 to 65.
2328 (PRED_NULL_RETURN): Change from 91 to 71.
2329 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2330 (PRED_LOOP_GUARD): Change from 66 to 73.
2332 2018-01-19 Martin Liska <mliska@suse.cz>
2334 * predict.c (predict_insn_def): Add new assert.
2335 (struct branch_predictor): Change type to signed integer.
2336 (test_prediction_value_range): Amend test to cover
2338 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2339 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2340 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2341 (PRED_LOOP_IV_COMPARE): Likewise.
2342 * predict.h (PROB_UNINITIALIZED): Define new constant.
2344 2018-01-19 Martin Liska <mliska@suse.cz>
2346 * predict.c (dump_prediction): Add new format for
2347 analyze_brprob.py script which is enabled with -details
2349 * profile-count.h (precise_p): New function.
2351 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2353 PR tree-optimization/83922
2354 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2355 there are no statements that need masking.
2356 (vect_active_double_reduction_p): New function.
2357 (vect_analyze_loop_operations): Use it when handling phis that
2358 are not in the loop header.
2360 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2362 PR tree-optimization/83914
2363 * tree-vect-loop.c (vectorizable_induction): Don't convert
2364 init_expr or apply the peeling adjustment for inductions
2365 that are nested within the vectorized loop.
2367 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2369 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2372 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2376 * function.h (gimplify_parameters): Add gimple_seq * argument.
2377 * function.c: Include gimple.h and options.h.
2378 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2379 for the added local temporaries if needed.
2380 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2381 if there are any parameter cleanups, wrap whole body into a
2382 try/finally with the cleanups.
2384 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2387 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2388 Use GET_MODE_CLASS for scalar floating point.
2390 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2394 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2395 Fix call of call_cgraph_insertion_hooks.
2397 2018-01-18 Martin Sebor <msebor@redhat.com>
2399 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2401 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2404 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2407 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2410 * common.opt: (-ffile-prefix-map): New option.
2411 * opts.c (common_handle_option): Defer it.
2412 * opts-global.c (handle_common_deferred_options): Handle it.
2413 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2414 * file-prefix-map.h: New file.
2415 (remap_debug_filename, add_debug_prefix_map): ...here.
2416 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2417 * final.c (debug_prefix_map, add_debug_prefix_map
2418 remap_debug_filename): Move to...
2419 * file-prefix-map.c: New file.
2420 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2421 generalize, get rid of alloca(), use strrchr() instead of strchr().
2422 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2423 Implement in terms of add_prefix_map().
2424 (remap_macro_filename, remap_debug_filename): Implement in term of
2426 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2427 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2428 * dbxout.c: Include file-prefix-map.h.
2429 * varasm.c: Likewise.
2430 * vmsdbgout.c: Likewise.
2431 * xcoffout.c: Likewise.
2432 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2433 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2434 * doc/invoke.texi (-ffile-prefix-map): Document.
2435 (-fdebug-prefix-map): Update description.
2437 2018-01-18 Martin Liska <mliska@suse.cz>
2439 * config/i386/i386.c (indirect_thunk_name): Document that also
2441 (output_indirect_thunk): Document why both instructions
2442 (pause and lfence) are generated.
2444 2018-01-18 Richard Biener <rguenther@suse.de>
2446 PR tree-optimization/83887
2447 * graphite-scop-detection.c
2448 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2449 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2450 (scop_detection::merge_sese): Re-implement with a flood-fill
2451 algorithm that properly finds a SESE region if it exists.
2453 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2456 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2457 pointer_diff optimizations use view_convert instead of convert.
2459 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2461 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2462 Generate different code for -mno-speculate-indirect-jumps.
2463 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2464 (*call_indirect_aix<mode>): Disable for
2465 -mno-speculate-indirect-jumps.
2466 (*call_indirect_aix<mode>_nospec): New define_insn.
2467 (*call_value_indirect_aix<mode>): Disable for
2468 -mno-speculate-indirect-jumps.
2469 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2470 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2471 -mno-speculate-indirect-jumps.
2472 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2474 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2476 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2477 long double type, set the flags for noting the default long double
2478 type, even if we don't pass or return a long double type.
2480 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2483 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2486 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2488 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2489 support for merge[hl].
2490 (fold_mergehl_helper): New helper function.
2491 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2492 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2493 (altivec_vmrglw_direct): Add xxmrglw insn.
2495 2018-01-17 Andrew Waterman <andrew@sifive.com>
2497 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2498 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2500 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2503 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2504 call the lto_location_cache before reading the
2505 DECL_SOURCE_LOCATION of the types.
2507 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2508 Richard Sandiford <richard.sandiford@linaro.org>
2510 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2511 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2512 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2513 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2514 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2516 * config/aarch64/constraints.md (aarch64_movti_operand):
2518 * config/aarch64/predicates.md (Uti): Add new constraint.
2520 2018-01-17 Carl Love <cel@us.ibm.com>
2522 * config/rs6000/vsx.md (define_expand xl_len_r,
2523 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2524 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2526 (define_expand, define_insn): Move the shift left from the
2527 define_insn to the define_expand for lxvl and stxvl instructions.
2528 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2529 and XL_LEN_R definitions to PURE.
2531 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2533 * config/i386/i386.c (indirect_thunk_name): Declare regno
2534 as unsigned int. Compare regno with INVALID_REGNUM.
2535 (output_indirect_thunk): Ditto.
2536 (output_indirect_thunk_function): Ditto.
2537 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2538 in the call to output_indirect_thunk_function.
2540 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2543 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2544 rather than the size of inner_type to determine the stack slot size
2545 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2547 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2550 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2553 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2555 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2556 endian Linux systems to optionally enable multilibs for selecting
2557 the long double type if the user configured an explicit type.
2558 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2559 have no long double multilibs if not defined.
2560 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2561 warn if the user used -mabi={ieee,ibm}longdouble and we built
2562 multilibs for long double.
2563 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2564 appropriate multilib option.
2565 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2567 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2568 for building long double multilibs.
2569 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2571 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2573 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2576 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2578 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2581 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2584 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2587 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2589 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2590 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2592 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2594 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2595 different rtl trees depending on TARGET_64BIT.
2596 (rs6000_gen_lvx): Likewise.
2598 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2600 * config/visium/visium.md (nop): Tweak comment.
2601 (hazard_nop): Likewise.
2603 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2605 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2606 -mspeculate-indirect-jumps.
2607 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2608 for -mno-speculate-indirect-jumps.
2609 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2610 (*call_value_indirect_elfv2<mode>): Disable for
2611 -mno-speculate-indirect-jumps.
2612 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2613 (indirect_jump): Emit different RTL for
2614 -mno-speculate-indirect-jumps.
2615 (*indirect_jump<mode>): Disable for
2616 -mno-speculate-indirect-jumps.
2617 (*indirect_jump<mode>_nospec): New define_insn.
2618 (tablejump): Emit different RTL for
2619 -mno-speculate-indirect-jumps.
2620 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2621 (tablejumpsi_nospec): New define_expand.
2622 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2623 (tablejumpdi_nospec): New define_expand.
2624 (*tablejump<mode>_internal1): Disable for
2625 -mno-speculate-indirect-jumps.
2626 (*tablejump<mode>_internal1_nospec): New define_insn.
2627 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2630 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2632 * caller-save.c (insert_save): Drop unnecessary parameter. All
2635 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2636 Richard Biener <rguenth@suse.de>
2639 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2640 return early, inline manually is_gimple_sizepos. Make sure if we
2641 call gimplify_expr we don't end up with a gimple constant.
2642 * tree.c (variably_modified_type_p): Don't return true for
2643 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2644 * gimplify.h (is_gimple_sizepos): Remove.
2646 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2648 PR tree-optimization/83857
2649 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2650 vectorizable_live_operation for pure SLP statements.
2651 (vectorizable_live_operation): Handle PHIs.
2653 2018-01-16 Richard Biener <rguenther@suse.de>
2655 PR tree-optimization/83867
2656 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2657 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2659 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2662 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2663 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2664 If off is not INTEGER_CST, issue a may not be aligned warning
2665 rather than isn't aligned. Use isn%'t rather than isn't.
2666 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2668 <case MULT_EXPR>: Improve the case when bottom and one of the
2669 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2670 operand, in that case check if the other operand is multiple of
2671 bottom divided by the INTEGER_CST operand.
2673 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2676 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2677 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2678 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2679 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2680 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2681 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2682 (pa_function_arg_size): New function.
2684 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2686 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2687 in a separate statement.
2689 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2691 PR tree-optimization/83847
2692 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2693 group gathers and scatters.
2695 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2697 PR rtl-optimization/86620
2698 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2700 PR rtl-optimization/83213
2701 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2702 to last if both are JUMP_INSNs.
2704 PR tree-optimization/83843
2705 * gimple-ssa-store-merging.c
2706 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2707 store_immediate_info for bswap/nop orig_stores.
2709 2018-01-15 Andrew Waterman <andrew@sifive.com>
2711 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2713 <UDIV>: Increase cost if !TARGET_DIV.
2715 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2717 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2718 (define_attr "cr_logical_3op"): New.
2719 (cceq_ior_compare): Adjust.
2720 (cceq_ior_compare_complement): Adjust.
2721 (*cceq_rev_compare): Adjust.
2722 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2723 (is_cracked_insn): Adjust.
2724 (insn_must_be_first_in_group): Adjust.
2725 * config/rs6000/40x.md: Adjust.
2726 * config/rs6000/440.md: Adjust.
2727 * config/rs6000/476.md: Adjust.
2728 * config/rs6000/601.md: Adjust.
2729 * config/rs6000/603.md: Adjust.
2730 * config/rs6000/6xx.md: Adjust.
2731 * config/rs6000/7450.md: Adjust.
2732 * config/rs6000/7xx.md: Adjust.
2733 * config/rs6000/8540.md: Adjust.
2734 * config/rs6000/cell.md: Adjust.
2735 * config/rs6000/e300c2c3.md: Adjust.
2736 * config/rs6000/e500mc.md: Adjust.
2737 * config/rs6000/e500mc64.md: Adjust.
2738 * config/rs6000/e5500.md: Adjust.
2739 * config/rs6000/e6500.md: Adjust.
2740 * config/rs6000/mpc.md: Adjust.
2741 * config/rs6000/power4.md: Adjust.
2742 * config/rs6000/power5.md: Adjust.
2743 * config/rs6000/power6.md: Adjust.
2744 * config/rs6000/power7.md: Adjust.
2745 * config/rs6000/power8.md: Adjust.
2746 * config/rs6000/power9.md: Adjust.
2747 * config/rs6000/rs64.md: Adjust.
2748 * config/rs6000/titan.md: Adjust.
2750 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2752 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2753 ix86_indirect_branch_register logic.
2755 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2757 * config/i386/constraints.md (Bs): Update
2758 ix86_indirect_branch_register check. Don't check
2759 ix86_indirect_branch_register with GOT_memory_operand.
2761 * config/i386/predicates.md (GOT_memory_operand): Don't check
2762 ix86_indirect_branch_register here.
2763 (GOT32_symbol_operand): Likewise.
2765 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2767 * config/i386/predicates.md (constant_call_address_operand):
2768 Rewrite ix86_indirect_branch_register logic.
2769 (sibcall_insn_operand): Likewise.
2771 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2773 * config/i386/constraints.md (Bs): Replace
2774 ix86_indirect_branch_thunk_register with
2775 ix86_indirect_branch_register.
2777 * config/i386/i386.md (indirect_jump): Likewise.
2778 (tablejump): Likewise.
2779 (*sibcall_memory): Likewise.
2780 (*sibcall_value_memory): Likewise.
2781 Peepholes of indirect call and jump via memory: Likewise.
2782 * config/i386/i386.opt: Likewise.
2783 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2784 (GOT_memory_operand): Likewise.
2785 (call_insn_operand): Likewise.
2786 (sibcall_insn_operand): Likewise.
2787 (GOT32_symbol_operand): Likewise.
2789 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2792 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2793 type rather than type addr's type points to.
2794 (expand_omp_atomic_mutex): Likewise.
2795 (expand_omp_atomic): Likewise.
2797 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2800 * config/i386/i386.c (output_indirect_thunk_function): Use
2801 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2802 for __x86_return_thunk.
2804 2018-01-15 Richard Biener <rguenther@suse.de>
2807 * expmed.c (extract_bit_field_1): Fix typo.
2809 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2812 * config/arm/iterators.md (VF): New mode iterator.
2813 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2814 Remove integer-related logic from pattern.
2815 (neon_vabd<mode>_3): Likewise.
2817 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2820 * common.opt (fstrict-overflow): No longer an alias.
2821 (fwrapv-pointer): New option.
2822 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2823 also for pointer types based on flag_wrapv_pointer.
2824 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2825 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2826 opts->x_flag_wrapv got set.
2827 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2828 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2829 POINTER_TYPE_OVERFLOW_UNDEFINED.
2830 * match.pd: Likewise in address comparison pattern.
2831 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2833 2018-01-15 Richard Biener <rguenther@suse.de>
2836 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2837 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2838 Reset type names to their identifier if their TYPE_DECL doesn't
2839 have linkage (and thus is used for ODR and devirt).
2840 (save_debug_info_for_decl): Remove.
2841 (save_debug_info_for_type): Likewise.
2842 (add_tree_to_fld_list): Adjust.
2843 * tree-pretty-print.c (dump_generic_node): Make dumping of
2844 type names more robust.
2846 2018-01-15 Richard Biener <rguenther@suse.de>
2848 * BASE-VER: Bump to 8.0.1.
2850 2018-01-14 Martin Sebor <msebor@redhat.com>
2853 * builtins.c (check_access): Avoid warning when the no-warning bit
2856 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2858 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2859 * ira-color (allocno_hard_regs_compare): Likewise.
2861 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2864 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2865 Use .pushsection/.popsection.
2867 2018-01-14 Martin Sebor <msebor@redhat.com>
2870 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2872 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2874 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2875 entry from extra_headers.
2876 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2877 extra_headers, make the list bitwise identical to the i?86-*-* one.
2879 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2881 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2882 -mcmodel=large with -mindirect-branch=thunk,
2883 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2884 -mfunction-return=thunk-extern.
2885 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2886 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2887 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2889 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2891 * config/i386/i386.c (print_reg): Print the name of the full
2892 integer register without '%'.
2893 (ix86_print_operand): Handle 'V'.
2894 * doc/extend.texi: Document 'V' modifier.
2896 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2898 * config/i386/constraints.md (Bs): Disallow memory operand for
2899 -mindirect-branch-register.
2901 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2902 (GOT_memory_operand): Likewise.
2903 (call_insn_operand): Likewise.
2904 (sibcall_insn_operand): Likewise.
2905 (GOT32_symbol_operand): Likewise.
2906 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2907 for -mindirect-branch-register.
2908 (tablejump): Likewise.
2909 (*sibcall_memory): Likewise.
2910 (*sibcall_value_memory): Likewise.
2911 Disallow peepholes of indirect call and jump via memory for
2912 -mindirect-branch-register.
2913 (*call_pop): Replace m with Bw.
2914 (*call_value_pop): Likewise.
2915 (*sibcall_pop_memory): Replace m with Bs.
2916 * config/i386/i386.opt (mindirect-branch-register): New option.
2917 * doc/invoke.texi: Document -mindirect-branch-register option.
2919 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2921 * config/i386/i386-protos.h (ix86_output_function_return): New.
2922 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2923 set function_return_type.
2924 (indirect_thunk_name): Add ret_p to indicate thunk for function
2926 (output_indirect_thunk_function): Pass false to
2927 indirect_thunk_name.
2928 (ix86_output_indirect_branch_via_reg): Likewise.
2929 (ix86_output_indirect_branch_via_push): Likewise.
2930 (output_indirect_thunk_function): Create alias for function
2931 return thunk if regno < 0.
2932 (ix86_output_function_return): New function.
2933 (ix86_handle_fndecl_attribute): Handle function_return.
2934 (ix86_attribute_table): Add function_return.
2935 * config/i386/i386.h (machine_function): Add
2936 function_return_type.
2937 * config/i386/i386.md (simple_return_internal): Use
2938 ix86_output_function_return.
2939 (simple_return_internal_long): Likewise.
2940 * config/i386/i386.opt (mfunction-return=): New option.
2941 (indirect_branch): Mention -mfunction-return=.
2942 * doc/extend.texi: Document function_return function attribute.
2943 * doc/invoke.texi: Document -mfunction-return= option.
2945 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2947 * config/i386/i386-opts.h (indirect_branch): New.
2948 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2949 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2950 with local indirect jump when converting indirect call and jump.
2951 (ix86_set_indirect_branch_type): New.
2952 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2953 (indirectlabelno): New.
2954 (indirect_thunk_needed): Likewise.
2955 (indirect_thunk_bnd_needed): Likewise.
2956 (indirect_thunks_used): Likewise.
2957 (indirect_thunks_bnd_used): Likewise.
2958 (INDIRECT_LABEL): Likewise.
2959 (indirect_thunk_name): Likewise.
2960 (output_indirect_thunk): Likewise.
2961 (output_indirect_thunk_function): Likewise.
2962 (ix86_output_indirect_branch_via_reg): Likewise.
2963 (ix86_output_indirect_branch_via_push): Likewise.
2964 (ix86_output_indirect_branch): Likewise.
2965 (ix86_output_indirect_jmp): Likewise.
2966 (ix86_code_end): Call output_indirect_thunk_function if needed.
2967 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2969 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2970 (ix86_attribute_table): Add indirect_branch.
2971 * config/i386/i386.h (machine_function): Add indirect_branch_type
2972 and has_local_indirect_jump.
2973 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2975 (tablejump): Likewise.
2976 (*indirect_jump): Use ix86_output_indirect_jmp.
2977 (*tablejump_1): Likewise.
2978 (simple_return_indirect_internal): Likewise.
2979 * config/i386/i386.opt (mindirect-branch=): New option.
2980 (indirect_branch): New.
2983 (thunk-inline): Likewise.
2984 (thunk-extern): Likewise.
2985 * doc/extend.texi: Document indirect_branch function attribute.
2986 * doc/invoke.texi: Document -mindirect-branch= option.
2988 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2991 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2993 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2995 * ipa-inline.c (want_inline_small_function_p): Return false if
2996 inlining has already failed with CIF_FINAL_ERROR.
2997 (update_caller_keys): Call want_inline_small_function_p before
2999 (update_callee_keys): Likewise.
3001 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
3003 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
3005 (rs6000_quadword_masked_address_p): Likewise.
3006 (quad_aligned_load_p): Likewise.
3007 (quad_aligned_store_p): Likewise.
3008 (const_load_sequence_p): Add comment to describe the outer-most loop.
3009 (mimic_memory_attributes_and_flags): New function.
3010 (rs6000_gen_stvx): Likewise.
3011 (replace_swapped_aligned_store): Likewise.
3012 (rs6000_gen_lvx): Likewise.
3013 (replace_swapped_aligned_load): Likewise.
3014 (replace_swapped_load_constant): Capitalize argument name in
3015 comment describing this function.
3016 (rs6000_analyze_swaps): Add a third pass to search for vector loads
3017 and stores that access quad-word aligned addresses and replace
3018 with stvx or lvx instructions when appropriate.
3019 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
3020 New function prototype.
3021 (rs6000_quadword_masked_address_p): Likewise.
3022 (rs6000_gen_lvx): Likewise.
3023 (rs6000_gen_stvx): Likewise.
3024 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
3025 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3026 when memory address is aligned.
3027 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3028 this split to select lvx instruction when memory address is aligned.
3029 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3030 instruction when memory address is aligned.
3031 (*vsx_le_perm_load_v16qi): Likewise.
3032 (four unnamed splitters): Modify to select the stvx instruction
3033 when memory is aligned.
3035 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3037 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3038 which appears in the queue multiple times.
3040 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3041 Alan Hayward <alan.hayward@arm.com>
3042 David Sherwood <david.sherwood@arm.com>
3044 * tree-vectorizer.h (vec_lower_bound): New structure.
3045 (_loop_vec_info): Add check_nonzero and lower_bounds.
3046 (LOOP_VINFO_CHECK_NONZERO): New macro.
3047 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3048 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3049 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3050 fields. Make seg_len the distance travelled, not including the
3052 (dr_direction_indicator): Declare.
3053 (dr_zero_step_indicator): Likewise.
3054 (dr_known_forward_stride_p): Likewise.
3055 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3057 (runtime_alias_check_p): Allow runtime alias checks with
3059 (operator ==): Compare access_size and align.
3060 (prune_runtime_alias_test_list): Rework for new distinction between
3061 the access_size and seg_len.
3062 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3064 (get_segment_min_max): New function.
3065 (create_intersect_range_checks): Use it.
3066 (dr_step_indicator): New function.
3067 (dr_direction_indicator): Likewise.
3068 (dr_zero_step_indicator): Likewise.
3069 (dr_known_forward_stride_p): Likewise.
3070 * tree-loop-distribution.c (data_ref_segment_size): Return
3071 DR_STEP * (niters - 1).
3072 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3074 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3075 (vect_preserves_scalar_order_p): New function, split out from...
3076 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3077 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3078 (vect_vfa_access_size): New function.
3079 (vect_vfa_align): Likewise.
3080 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3081 (dump_lower_bound): New function.
3082 (vect_check_lower_bound): Likewise.
3083 (vect_small_gap_p): Likewise.
3084 (vectorizable_with_step_bound_p): Likewise.
3085 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3086 depencies if the vectorization factor is 1. Convert the checks
3087 for nonzero steps into checks on the bounds of DR_STEP. Try using
3088 a bunds check for variable steps if the minimum required step is
3089 relatively small. Update calls to the dr_with_seg_len
3090 constructor and to vect_compile_time_alias.
3091 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3093 (vect_loop_versioning): Call it.
3094 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3096 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3098 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3099 Alan Hayward <alan.hayward@arm.com>
3100 David Sherwood <david.sherwood@arm.com>
3102 * doc/sourcebuild.texi (vect_scatter_store): Document.
3103 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3105 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3107 * genopinit.c (main): Add supports_vec_scatter_store and
3108 supports_vec_scatter_store_cached to target_optabs.
3109 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3110 IFN_MASK_SCATTER_STORE.
3111 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3113 * internal-fn.h (internal_store_fn_p): Declare.
3114 (internal_fn_stored_value_index): Likewise.
3115 * internal-fn.c (scatter_store_direct): New macro.
3116 (expand_scatter_store_optab_fn): New function.
3117 (direct_scatter_store_optab_supported_p): New macro.
3118 (internal_store_fn_p): New function.
3119 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3120 IFN_MASK_SCATTER_STORE.
3121 (internal_fn_mask_index): Likewise.
3122 (internal_fn_stored_value_index): New function.
3123 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3125 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3126 * optabs-query.c (supports_vec_scatter_store_p): New function.
3127 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3128 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3129 true for scatter stores.
3130 (vect_gather_scatter_fn_p): Handle scatter stores too.
3131 (vect_check_gather_scatter): Consider using scatter stores if
3132 supports_vec_scatter_store_p.
3133 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3135 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3136 internal_fn_stored_value_index.
3137 (check_load_store_masking): Handle scatter stores too.
3138 (vect_get_store_rhs): Make public.
3139 (vectorizable_call): Use internal_store_fn_p.
3140 (vectorizable_store): Handle scatter store internal functions.
3141 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3142 when deciding whether the end of the group has been reached.
3143 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3144 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3145 (mask_scatter_store<mode>): New insns.
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-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3152 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3153 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3155 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3156 Use vect_truncate_gather_scatter_offset if we can't treat the
3157 operation as a normal gather load or scatter store.
3158 (get_group_load_store_type): Take the gather_scatter_info
3159 as argument. Try using a gather load or scatter store for
3160 single-element groups.
3161 (get_load_store_type): Update calls to get_group_load_store_type
3162 and vect_use_strided_gather_scatters_p.
3164 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3165 Alan Hayward <alan.hayward@arm.com>
3166 David Sherwood <david.sherwood@arm.com>
3168 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3169 optional tree argument.
3170 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3172 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3173 but continue to use the current value as a fallback.
3174 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3175 to compare the updates.
3176 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3177 (get_load_store_type): Use it when handling a strided access.
3178 (vect_get_strided_load_store_ops): New function.
3179 (vect_get_data_ptr_increment): Likewise.
3180 (vectorizable_load): Handle strided gather loads. Always pass
3181 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3183 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3184 Alan Hayward <alan.hayward@arm.com>
3185 David Sherwood <david.sherwood@arm.com>
3187 * doc/md.texi (gather_load@var{m}): Document.
3188 (mask_gather_load@var{m}): Likewise.
3189 * genopinit.c (main): Add supports_vec_gather_load and
3190 supports_vec_gather_load_cached to target_optabs.
3191 * optabs-tree.c (init_tree_optimization_optabs): Use
3192 ggc_cleared_alloc to allocate target_optabs.
3193 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3194 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3196 * internal-fn.h (internal_load_fn_p): Declare.
3197 (internal_gather_scatter_fn_p): Likewise.
3198 (internal_fn_mask_index): Likewise.
3199 (internal_gather_scatter_fn_supported_p): Likewise.
3200 * internal-fn.c (gather_load_direct): New macro.
3201 (expand_gather_load_optab_fn): New function.
3202 (direct_gather_load_optab_supported_p): New macro.
3203 (direct_internal_fn_optab): New function.
3204 (internal_load_fn_p): Likewise.
3205 (internal_gather_scatter_fn_p): Likewise.
3206 (internal_fn_mask_index): Likewise.
3207 (internal_gather_scatter_fn_supported_p): Likewise.
3208 * optabs-query.c (supports_at_least_one_mode_p): New function.
3209 (supports_vec_gather_load_p): Likewise.
3210 * optabs-query.h (supports_vec_gather_load_p): Declare.
3211 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3212 and memory_type field.
3213 (NUM_PATTERNS): Bump to 15.
3214 * tree-vect-data-refs.c: Include internal-fn.h.
3215 (vect_gather_scatter_fn_p): New function.
3216 (vect_describe_gather_scatter_call): Likewise.
3217 (vect_check_gather_scatter): Try using internal functions for
3218 gather loads. Recognize existing calls to a gather load function.
3219 (vect_analyze_data_refs): Consider using gather loads if
3220 supports_vec_gather_load_p.
3221 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3222 (vect_get_gather_scatter_offset_type): Likewise.
3223 (vect_convert_mask_for_vectype): Likewise.
3224 (vect_add_conversion_to_patterm): Likewise.
3225 (vect_try_gather_scatter_pattern): Likewise.
3226 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3227 (vect_vect_recog_func_ptrs): Add it.
3228 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3229 internal_fn_mask_index and internal_gather_scatter_fn_p.
3230 (check_load_store_masking): Take the gather_scatter_info as an
3231 argument and handle gather loads.
3232 (vect_get_gather_scatter_ops): New function.
3233 (vectorizable_call): Check internal_load_fn_p.
3234 (vectorizable_load): Likewise. Handle gather load internal
3236 (vectorizable_store): Update call to check_load_store_masking.
3237 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3238 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3239 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3240 (aarch64_gather_scale_operand_d): New predicates.
3241 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3242 (mask_gather_load<mode>): New insns.
3244 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3245 Alan Hayward <alan.hayward@arm.com>
3246 David Sherwood <david.sherwood@arm.com>
3248 * optabs.def (fold_left_plus_optab): New optab.
3249 * doc/md.texi (fold_left_plus_@var{m}): Document.
3250 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3251 * internal-fn.c (fold_left_direct): Define.
3252 (expand_fold_left_optab_fn): Likewise.
3253 (direct_fold_left_optab_supported_p): Likewise.
3254 * fold-const-call.c (fold_const_fold_left): New function.
3255 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3256 * tree-parloops.c (valid_reduction_p): New function.
3257 (gather_scalar_reductions): Use it.
3258 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3259 (vect_finish_replace_stmt): Declare.
3260 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3261 (needs_fold_left_reduction_p): New function, split out from...
3262 (vect_is_simple_reduction): ...here. Accept reductions that
3263 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3264 (vect_force_simple_reduction): Also store the reduction type in
3265 the assignment's STMT_VINFO_REDUC_TYPE.
3266 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3267 (merge_with_identity): New function.
3268 (vect_expand_fold_left): Likewise.
3269 (vectorize_fold_left_reduction): Likewise.
3270 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3271 scalar phi in place for it. Check for target support and reject
3272 cases that would reassociate the operation. Defer the transform
3273 phase to vectorize_fold_left_reduction.
3274 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3275 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3276 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3278 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3280 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3281 call to ifc_temp_var.
3283 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3284 Alan Hayward <alan.hayward@arm.com>
3285 David Sherwood <david.sherwood@arm.com>
3287 * target.def (legitimize_address_displacement): Take the original
3288 offset as a poly_int.
3289 * targhooks.h (default_legitimize_address_displacement): Update
3291 * targhooks.c (default_legitimize_address_displacement): Likewise.
3292 * doc/tm.texi: Regenerate.
3293 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3294 as an argument, moving assert of ad->disp == ad->disp_term to...
3295 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3296 Try calling targetm.legitimize_address_displacement before expanding
3297 the address rather than afterwards, and adjust for the new interface.
3298 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3299 Match the new hook interface. Handle SVE addresses.
3300 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3303 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3305 * Makefile.in (OBJS): Add early-remat.o.
3306 * target.def (select_early_remat_modes): New hook.
3307 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3308 * doc/tm.texi: Regenerate.
3309 * targhooks.h (default_select_early_remat_modes): Declare.
3310 * targhooks.c (default_select_early_remat_modes): New function.
3311 * timevar.def (TV_EARLY_REMAT): New timevar.
3312 * passes.def (pass_early_remat): New pass.
3313 * tree-pass.h (make_pass_early_remat): Declare.
3314 * early-remat.c: New file.
3315 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3317 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3319 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3320 Alan Hayward <alan.hayward@arm.com>
3321 David Sherwood <david.sherwood@arm.com>
3323 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3324 vfm1 with a bound_epilog parameter.
3325 (vect_do_peeling): Update calls accordingly, and move the prologue
3326 call earlier in the function. Treat the base bound_epilog as 0 for
3327 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3328 this base when peeling for gaps.
3329 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3330 with fully-masked loops.
3331 (vect_estimate_min_profitable_iters): Handle the single peeled
3332 iteration in that case.
3334 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3335 Alan Hayward <alan.hayward@arm.com>
3336 David Sherwood <david.sherwood@arm.com>
3338 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3339 single-element interleaving even if the size is not a power of 2.
3340 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3341 accesses for single-element interleaving if the group size is
3344 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3345 Alan Hayward <alan.hayward@arm.com>
3346 David Sherwood <david.sherwood@arm.com>
3348 * doc/md.texi (fold_extract_last_@var{m}): Document.
3349 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3350 * optabs.def (fold_extract_last_optab): New optab.
3351 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3352 * internal-fn.c (fold_extract_direct): New macro.
3353 (expand_fold_extract_optab_fn): Likewise.
3354 (direct_fold_extract_optab_supported_p): Likewise.
3355 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3356 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3357 EXTRACT_LAST_REDUCTION.
3358 (get_initial_def_for_reduction): Do not create an initial vector
3359 for EXTRACT_LAST_REDUCTION reductions.
3360 (vectorizable_reduction): Leave the scalar phi in place for
3361 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3362 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3363 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3364 transform phase to vectorizable_condition.
3365 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3367 (vect_finish_stmt_generation): ...here.
3368 (vect_finish_replace_stmt): New function.
3369 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3370 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3372 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3374 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3375 Alan Hayward <alan.hayward@arm.com>
3376 David Sherwood <david.sherwood@arm.com>
3378 * doc/md.texi (extract_last_@var{m}): Document.
3379 * optabs.def (extract_last_optab): New optab.
3380 * internal-fn.def (EXTRACT_LAST): New internal function.
3381 * internal-fn.c (cond_unary_direct): New macro.
3382 (expand_cond_unary_optab_fn): Likewise.
3383 (direct_cond_unary_optab_supported_p): Likewise.
3384 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3385 loops using EXTRACT_LAST.
3386 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3387 (extract_last_<mode>): ...this optab.
3388 (vec_extract<mode><Vel>): Update accordingly.
3390 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3391 Alan Hayward <alan.hayward@arm.com>
3392 David Sherwood <david.sherwood@arm.com>
3394 * target.def (empty_mask_is_expensive): New hook.
3395 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3396 * doc/tm.texi: Regenerate.
3397 * targhooks.h (default_empty_mask_is_expensive): Declare.
3398 * targhooks.c (default_empty_mask_is_expensive): New function.
3399 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3400 if the target says that empty masks are expensive.
3401 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3403 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3405 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3406 Alan Hayward <alan.hayward@arm.com>
3407 David Sherwood <david.sherwood@arm.com>
3409 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3410 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3411 (vect_use_loop_mask_for_alignment_p): New function.
3412 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3413 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3414 niters_skip argument. Make sure that the first niters_skip elements
3415 of the first iteration are inactive.
3416 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3417 Update call to vect_set_loop_masks_directly.
3418 (get_misalign_in_elems): New function, split out from...
3419 (vect_gen_prolog_loop_niters): ...here.
3420 (vect_update_init_of_dr): Take a code argument that specifies whether
3421 the adjustment should be added or subtracted.
3422 (vect_update_init_of_drs): Likewise.
3423 (vect_prepare_for_masked_peels): New function.
3424 (vect_do_peeling): Skip prologue peeling if we're using a mask
3425 instead. Update call to vect_update_inits_of_drs.
3426 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3428 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3429 alignment. Do not include the number of peeled iterations in
3430 the minimum threshold in that case.
3431 (vectorizable_induction): Adjust the start value down by
3432 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3433 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3434 Take the number of skipped iterations into account when calculating
3436 * tree-vect-stmts.c (vect_gen_while_not): New function.
3438 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3439 Alan Hayward <alan.hayward@arm.com>
3440 David Sherwood <david.sherwood@arm.com>
3442 * doc/sourcebuild.texi (vect_fully_masked): Document.
3443 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3445 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3447 (vect_analyze_loop_2): ...here. Don't check the vectorization
3448 factor against the number of loop iterations if the loop is
3451 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3452 Alan Hayward <alan.hayward@arm.com>
3453 David Sherwood <david.sherwood@arm.com>
3455 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3456 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3457 (dump_groups): Update accordingly.
3458 (iv_use::mem_type): New member variable.
3459 (address_p): New function.
3460 (record_use): Add a mem_type argument and initialize the new
3462 (record_group_use): Add a mem_type argument. Use address_p.
3463 Remove obsolete null checks of base_object. Update call to record_use.
3464 (find_interesting_uses_op): Update call to record_group_use.
3465 (find_interesting_uses_cond): Likewise.
3466 (find_interesting_uses_address): Likewise.
3467 (get_mem_type_for_internal_fn): New function.
3468 (find_address_like_use): Likewise.
3469 (find_interesting_uses_stmt): Try find_address_like_use before
3470 calling find_interesting_uses_op.
3471 (addr_offset_valid_p): Use the iv mem_type field as the type
3472 of the addressed memory.
3473 (add_autoinc_candidates): Likewise.
3474 (get_address_cost): Likewise.
3475 (split_small_address_groups_p): Use address_p.
3476 (split_address_groups): Likewise.
3477 (add_iv_candidate_for_use): Likewise.
3478 (autoinc_possible_for_pair): Likewise.
3479 (rewrite_groups): Likewise.
3480 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3481 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3482 (get_alias_ptr_type_for_ptr_address): New function.
3483 (rewrite_use_address): Rewrite address uses in calls that were
3484 identified by find_address_like_use.
3486 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3487 Alan Hayward <alan.hayward@arm.com>
3488 David Sherwood <david.sherwood@arm.com>
3490 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3492 * gimple-expr.h (is_gimple_addressable: Likewise.
3493 * gimple-expr.c (is_gimple_address): Likewise.
3494 * internal-fn.c (expand_call_mem_ref): New function.
3495 (expand_mask_load_optab_fn): Use it.
3496 (expand_mask_store_optab_fn): Likewise.
3498 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3499 Alan Hayward <alan.hayward@arm.com>
3500 David Sherwood <david.sherwood@arm.com>
3502 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3503 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3504 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3505 (cond_umax@var{mode}): Document.
3506 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3507 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3508 (cond_umin_optab, cond_umax_optab): New optabs.
3509 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3510 (COND_IOR, COND_XOR): New internal functions.
3511 * internal-fn.h (get_conditional_internal_fn): Declare.
3512 * internal-fn.c (cond_binary_direct): New macro.
3513 (expand_cond_binary_optab_fn): Likewise.
3514 (direct_cond_binary_optab_supported_p): Likewise.
3515 (get_conditional_internal_fn): New function.
3516 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3517 Cope with reduction statements that are vectorized as calls rather
3519 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3520 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3521 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3522 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3523 (UNSPEC_COND_EOR): New unspecs.
3524 (optab): Add mappings for them.
3525 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3526 (sve_int_op, sve_fp_op): New int attributes.
3528 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3529 Alan Hayward <alan.hayward@arm.com>
3530 David Sherwood <david.sherwood@arm.com>
3532 * optabs.def (while_ult_optab): New optab.
3533 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3534 * internal-fn.def (WHILE_ULT): New internal function.
3535 * internal-fn.h (direct_internal_fn_supported_p): New override
3536 that takes two types as argument.
3537 * internal-fn.c (while_direct): New macro.
3538 (expand_while_optab_fn): New function.
3539 (convert_optab_supported_p): Likewise.
3540 (direct_while_optab_supported_p): New macro.
3541 * wide-int.h (wi::udiv_ceil): New function.
3542 * tree-vectorizer.h (rgroup_masks): New structure.
3543 (vec_loop_masks): New typedef.
3544 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3546 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3547 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3548 (vect_max_vf): New function.
3549 (slpeel_make_loop_iterate_ntimes): Delete.
3550 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3551 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3552 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3553 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3554 internal-fn.h, stor-layout.h and optabs-query.h.
3555 (vect_set_loop_mask): New function.
3556 (add_preheader_seq): Likewise.
3557 (add_header_seq): Likewise.
3558 (interleave_supported_p): Likewise.
3559 (vect_maybe_permute_loop_masks): Likewise.
3560 (vect_set_loop_masks_directly): Likewise.
3561 (vect_set_loop_condition_masked): Likewise.
3562 (vect_set_loop_condition_unmasked): New function, split out from
3563 slpeel_make_loop_iterate_ntimes.
3564 (slpeel_make_loop_iterate_ntimes): Rename to..
3565 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3566 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3567 (vect_do_peeling): Update call accordingly.
3568 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3570 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3571 mask_compare_type, can_fully_mask_p and fully_masked_p.
3572 (release_vec_loop_masks): New function.
3573 (_loop_vec_info): Use it to free the loop masks.
3574 (can_produce_all_loop_masks_p): New function.
3575 (vect_get_max_nscalars_per_iter): Likewise.
3576 (vect_verify_full_masking): Likewise.
3577 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3578 retries, and free the mask rgroups before retrying. Check loop-wide
3579 reasons for disallowing fully-masked loops. Make the final decision
3580 about whether use a fully-masked loop or not.
3581 (vect_estimate_min_profitable_iters): Do not assume that peeling
3582 for the number of iterations will be needed for fully-masked loops.
3583 (vectorizable_reduction): Disable fully-masked loops.
3584 (vectorizable_live_operation): Likewise.
3585 (vect_halve_mask_nunits): New function.
3586 (vect_double_mask_nunits): Likewise.
3587 (vect_record_loop_mask): Likewise.
3588 (vect_get_loop_mask): Likewise.
3589 (vect_transform_loop): Handle the case in which the final loop
3590 iteration might handle a partial vector. Call vect_set_loop_condition
3591 instead of slpeel_make_loop_iterate_ntimes.
3592 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3593 (check_load_store_masking): New function.
3594 (prepare_load_store_mask): Likewise.
3595 (vectorizable_store): Handle fully-masked loops.
3596 (vectorizable_load): Likewise.
3597 (supportable_widening_operation): Use vect_halve_mask_nunits for
3599 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3600 (vect_gen_while): New function.
3601 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3602 (aarch64_uqdec<mode>): New insn.
3604 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3605 Alan Hayward <alan.hayward@arm.com>
3606 David Sherwood <david.sherwood@arm.com>
3608 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3609 (reduc_xor_scal_optab): New optabs.
3610 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3611 (reduc_xor_scal_@var{m}): Document.
3612 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3613 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3615 * fold-const-call.c (fold_const_call): Handle them.
3616 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3617 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3618 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3619 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3620 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3621 (UNSPEC_XORV): New unspecs.
3622 (optab): Add entries for them.
3623 (BITWISEV): New int iterator.
3624 (bit_reduc_op): New int attributes.
3626 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3627 Alan Hayward <alan.hayward@arm.com>
3628 David Sherwood <david.sherwood@arm.com>
3630 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3631 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3632 * optabs.def (vec_shl_insert_optab): New optab.
3633 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3634 (duplicate_and_interleave): Likewise.
3635 * tree-vect-loop.c: Include internal-fn.h.
3636 (neutral_op_for_slp_reduction): New function, split out from
3637 get_initial_defs_for_reduction.
3638 (get_initial_def_for_reduction): Handle option 2 for variable-length
3639 vectors by loading the neutral value into a vector and then shifting
3640 the initial value into element 0.
3641 (get_initial_defs_for_reduction): Replace the code argument with
3642 the neutral value calculated by neutral_op_for_slp_reduction.
3643 Use gimple_build_vector for constant-length vectors.
3644 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3645 but the first group_size elements have a neutral value.
3646 Use duplicate_and_interleave otherwise.
3647 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3648 Update call to get_initial_defs_for_reduction. Handle SLP
3649 reductions for variable-length vectors by creating one vector
3650 result for each scalar result, with the elements associated
3651 with other scalar results stubbed out with the neutral value.
3652 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3653 Require IFN_VEC_SHL_INSERT for double reductions on
3654 variable-length vectors, or SLP reductions that have
3655 a neutral value. Require can_duplicate_and_interleave_p
3656 support for variable-length unchained SLP reductions if there
3657 is no neutral value, such as for MIN/MAX reductions. Also require
3658 the number of vector elements to be a multiple of the number of
3659 SLP statements when doing variable-length unchained SLP reductions.
3660 Update call to vect_create_epilog_for_reduction.
3661 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3662 and remove initial values.
3663 (duplicate_and_interleave): Make public.
3664 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3665 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3667 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3668 Alan Hayward <alan.hayward@arm.com>
3669 David Sherwood <david.sherwood@arm.com>
3671 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3672 (can_duplicate_and_interleave_p): New function.
3673 (vect_get_and_check_slp_defs): Take the vector of statements
3674 rather than just the current one. Remove excess parentheses.
3675 Restriction rejectinon of vect_constant_def and vect_external_def
3676 for variable-length vectors to boolean types, or types for which
3677 can_duplicate_and_interleave_p is false.
3678 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3679 (duplicate_and_interleave): New function.
3680 (vect_get_constant_vectors): Use gimple_build_vector for
3681 constant-length vectors and suitable variable-length constant
3682 vectors. Use duplicate_and_interleave for other variable-length
3683 vectors. Don't defer the update when inserting new statements.
3685 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3686 Alan Hayward <alan.hayward@arm.com>
3687 David Sherwood <david.sherwood@arm.com>
3689 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3690 min_profitable_iters doesn't go negative.
3692 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3693 Alan Hayward <alan.hayward@arm.com>
3694 David Sherwood <david.sherwood@arm.com>
3696 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3697 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3698 * optabs.def (vec_mask_load_lanes_optab): New optab.
3699 (vec_mask_store_lanes_optab): Likewise.
3700 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3701 (MASK_STORE_LANES): Likewise.
3702 * internal-fn.c (mask_load_lanes_direct): New macro.
3703 (mask_store_lanes_direct): Likewise.
3704 (expand_mask_load_optab_fn): Handle masked operations.
3705 (expand_mask_load_lanes_optab_fn): New macro.
3706 (expand_mask_store_optab_fn): Handle masked operations.
3707 (expand_mask_store_lanes_optab_fn): New macro.
3708 (direct_mask_load_lanes_optab_supported_p): Likewise.
3709 (direct_mask_store_lanes_optab_supported_p): Likewise.
3710 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3712 (vect_load_lanes_supported): Likewise.
3713 * tree-vect-data-refs.c (strip_conversion): New function.
3714 (can_group_stmts_p): Likewise.
3715 (vect_analyze_data_ref_accesses): Use it instead of checking
3716 for a pair of assignments.
3717 (vect_store_lanes_supported): Take a masked_p parameter.
3718 (vect_load_lanes_supported): Likewise.
3719 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3720 vect_store_lanes_supported and vect_load_lanes_supported.
3721 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3722 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3723 parameter. Don't allow gaps for masked accesses.
3724 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3725 and vect_load_lanes_supported.
3726 (get_load_store_type): Take a masked_p parameter and update
3727 call to get_group_load_store_type.
3728 (vectorizable_store): Update call to get_load_store_type.
3729 Handle IFN_MASK_STORE_LANES.
3730 (vectorizable_load): Update call to get_load_store_type.
3731 Handle IFN_MASK_LOAD_LANES.
3733 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3734 Alan Hayward <alan.hayward@arm.com>
3735 David Sherwood <david.sherwood@arm.com>
3737 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3739 * config/aarch64/aarch64-protos.h
3740 (aarch64_sve_struct_memory_operand_p): Declare.
3741 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3742 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3743 (VPRED, vpred): Handle SVE structure modes.
3744 * config/aarch64/constraints.md (Utx): New constraint.
3745 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3746 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3747 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3748 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3749 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3750 structure modes. Split into pieces after RA.
3751 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3752 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3754 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3755 SVE structure modes.
3756 (aarch64_classify_address): Likewise.
3757 (sizetochar): Move earlier in file.
3758 (aarch64_print_operand): Handle SVE register lists.
3759 (aarch64_array_mode): New function.
3760 (aarch64_sve_struct_memory_operand_p): Likewise.
3761 (TARGET_ARRAY_MODE): Redefine.
3763 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3764 Alan Hayward <alan.hayward@arm.com>
3765 David Sherwood <david.sherwood@arm.com>
3767 * target.def (array_mode): New target hook.
3768 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3769 * doc/tm.texi: Regenerate.
3770 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3771 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3772 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3774 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3777 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3778 Alan Hayward <alan.hayward@arm.com>
3779 David Sherwood <david.sherwood@arm.com>
3781 * fold-const.c (fold_binary_loc): Check the argument types
3782 rather than the result type when testing for a vector operation.
3784 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3786 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3787 * doc/tm.texi: Regenerate.
3789 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3790 Alan Hayward <alan.hayward@arm.com>
3791 David Sherwood <david.sherwood@arm.com>
3793 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3794 (sve): Document new AArch64 extension.
3795 * doc/md.texi (w): Extend the description of the AArch64
3796 constraint to include SVE vectors.
3797 (Upl, Upa): Document new AArch64 predicate constraints.
3798 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3800 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3801 (msve-vector-bits=): New option.
3802 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3803 SVE when these are disabled.
3804 (sve): New extension.
3805 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3806 modes. Adjust their number of units based on aarch64_sve_vg.
3807 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3808 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3809 aarch64_addr_query_type.
3810 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3811 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3812 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3813 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3814 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3815 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3816 (aarch64_simd_imm_zero_p): Delete.
3817 (aarch64_check_zero_based_sve_index_immediate): Declare.
3818 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3819 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3820 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3821 (aarch64_sve_float_mul_immediate_p): Likewise.
3822 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3824 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3825 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3826 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3827 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3828 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3829 (aarch64_regmode_natural_size): Likewise.
3830 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3831 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3833 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3834 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3835 for VG and the SVE predicate registers.
3836 (V_ALIASES): Add a "z"-prefixed alias.
3837 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3838 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3839 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3840 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3841 (REG_CLASS_NAMES): Add entries for them.
3842 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3843 and the predicate registers.
3844 (aarch64_sve_vg): Declare.
3845 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3846 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3847 (REGMODE_NATURAL_SIZE): Define.
3848 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3850 * config/aarch64/aarch64.c: Include cfgrtl.h.
3851 (simd_immediate_info): Add a constructor for series vectors,
3852 and an associated step field.
3853 (aarch64_sve_vg): New variable.
3854 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3855 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3856 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3857 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3858 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3859 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3860 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3861 (aarch64_get_mask_mode): New functions.
3862 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3863 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3864 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3865 predicate modes and predicate registers. Explicitly restrict
3866 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3867 to store a vector mode if it is recognized by
3868 aarch64_classify_vector_mode.
3869 (aarch64_regmode_natural_size): New function.
3870 (aarch64_hard_regno_caller_save_mode): Return the original mode
3872 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3873 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3874 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3875 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3877 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3878 does not overlap dest if the function is frame-related. Handle
3880 (aarch64_split_add_offset): New function.
3881 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3882 them aarch64_add_offset.
3883 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3884 and update call to aarch64_sub_sp.
3885 (aarch64_add_cfa_expression): New function.
3886 (aarch64_expand_prologue): Pass extra temporary registers to the
3887 functions above. Handle the case in which we need to emit new
3888 DW_CFA_expressions for registers that were originally saved
3889 relative to the stack pointer, but now have to be expressed
3890 relative to the frame pointer.
3891 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3893 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3894 IP0 and IP1 values for SVE frames.
3895 (aarch64_expand_vec_series): New function.
3896 (aarch64_expand_sve_widened_duplicate): Likewise.
3897 (aarch64_expand_sve_const_vector): Likewise.
3898 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3899 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3900 into the register, rather than emitting a SET directly.
3901 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3902 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3903 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3904 (offset_9bit_signed_scaled_p): New functions.
3905 (aarch64_replicate_bitmask_imm): New function.
3906 (aarch64_bitmask_imm): Use it.
3907 (aarch64_cannot_force_const_mem): Reject expressions involving
3908 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3909 (aarch64_classify_index): Handle SVE indices, by requiring
3910 a plain register index with a scale that matches the element size.
3911 (aarch64_classify_address): Handle SVE addresses. Assert that
3912 the mode of the address is VOIDmode or an integer mode.
3913 Update call to aarch64_classify_symbol.
3914 (aarch64_classify_symbolic_expression): Update call to
3915 aarch64_classify_symbol.
3916 (aarch64_const_vec_all_in_range_p): New function.
3917 (aarch64_print_vector_float_operand): Likewise.
3918 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3919 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3920 and the FP immediates 1.0 and 0.5.
3921 (aarch64_print_address_internal): Handle SVE addresses.
3922 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3923 (aarch64_regno_regclass): Handle predicate registers.
3924 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3926 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3927 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3928 (aarch64_convert_sve_vector_bits): New function.
3929 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3930 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3932 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3933 Handle SVE vector and predicate modes. Accept VL-based constants
3934 that need only one temporary register, and VL offsets that require
3935 no temporary registers.
3936 (aarch64_conditional_register_usage): Mark the predicate registers
3937 as fixed if SVE isn't available.
3938 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3939 Return true for SVE vector and predicate modes.
3940 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3941 rather than an unsigned int. Handle SVE modes.
3942 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3944 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3946 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3947 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3948 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3949 (aarch64_sve_float_mul_immediate_p): New functions.
3950 (aarch64_sve_valid_immediate): New function.
3951 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3952 Explicitly reject structure modes. Check for INDEX constants.
3953 Handle PTRUE and PFALSE constants.
3954 (aarch64_check_zero_based_sve_index_immediate): New function.
3955 (aarch64_simd_imm_zero_p): Delete.
3956 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3957 vector modes. Accept constants in the range of CNT[BHWD].
3958 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3959 ask for an Advanced SIMD mode.
3960 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3961 (aarch64_simd_vector_alignment): Handle SVE predicates.
3962 (aarch64_vectorize_preferred_vector_alignment): New function.
3963 (aarch64_simd_vector_alignment_reachable): Use it instead of
3965 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3966 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3968 (MAX_VECT_LEN): Delete.
3969 (expand_vec_perm_d): Add a vec_flags field.
3970 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3971 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3972 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3974 (aarch64_evpc_rev): Rename to...
3975 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3976 (aarch64_evpc_rev_global): New function.
3977 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3978 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3980 (aarch64_evpc_sve_tbl): New function.
3981 (aarch64_expand_vec_perm_const_1): Update after rename of
3982 aarch64_evpc_rev. Handle SVE permutes too, trying
3983 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3984 than aarch64_evpc_tbl.
3985 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3986 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3987 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3988 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3989 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3990 (aarch64_expand_sve_vcond): New functions.
3991 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3992 of aarch64_vector_mode_p.
3993 (aarch64_dwarf_poly_indeterminate_value): New function.
3994 (aarch64_compute_pressure_classes): Likewise.
3995 (aarch64_can_change_mode_class): Likewise.
3996 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3997 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3998 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3999 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
4000 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
4001 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
4002 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
4003 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
4005 (Dn, Dl, Dr): Accept const as well as const_vector.
4006 (Dz): Likewise. Compare against CONST0_RTX.
4007 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
4008 of "vector" where appropriate.
4009 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
4010 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
4011 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
4012 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
4013 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
4014 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
4015 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
4016 (v_int_equiv): Extend to SVE modes.
4017 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
4019 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
4020 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
4021 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
4022 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
4023 (SVE_COND_FP_CMP): New int iterators.
4024 (perm_hilo): Handle the new unpack unspecs.
4025 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4027 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4028 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4029 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4030 (aarch64_equality_operator, aarch64_constant_vector_operand)
4031 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4032 (aarch64_sve_nonimmediate_operand): Likewise.
4033 (aarch64_sve_general_operand): Likewise.
4034 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4035 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4036 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4037 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4038 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4039 (aarch64_sve_float_arith_immediate): Likewise.
4040 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4041 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4042 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4043 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4044 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4045 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4046 (aarch64_sve_float_arith_operand): Likewise.
4047 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4048 (aarch64_sve_float_mul_operand): Likewise.
4049 (aarch64_sve_vec_perm_operand): Likewise.
4050 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4051 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4052 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4053 as well as const_vector.
4054 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4055 in file. Use CONST0_RTX and CONSTM1_RTX.
4056 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4057 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4058 Use aarch64_simd_imm_zero.
4059 * config/aarch64/aarch64-sve.md: New file.
4060 * config/aarch64/aarch64.md: Include it.
4061 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4062 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4063 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4064 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4065 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4066 (sve): New attribute.
4067 (enabled): Disable instructions with the sve attribute unless
4069 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4070 aarch64_expand_mov_immediate.
4071 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4072 CNT[BHSD] immediates.
4073 (movti): Split CONST_POLY_INT moves into two halves.
4074 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4075 Split additions that need a temporary here if the destination
4076 is the stack pointer.
4077 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4078 (*add<mode>3_poly_1): New instruction.
4079 (set_clobber_cc): New expander.
4081 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4083 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4084 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4085 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4086 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4087 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4088 Change innermode from fixed_mode_size to machine_mode.
4089 (simplify_subreg): Update call accordingly. Handle a constant-sized
4090 subreg of a variable-length CONST_VECTOR.
4092 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4093 Alan Hayward <alan.hayward@arm.com>
4094 David Sherwood <david.sherwood@arm.com>
4096 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4097 (add_offset_to_base): New function, split out from...
4098 (create_mem_ref): ...here. When handling a scale other than 1,
4099 check first whether the address is valid without the offset.
4100 Add it into the base if so, leaving the index and scale as-is.
4102 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4105 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4106 fold_for_warn before checking if arg2 is INTEGER_CST.
4108 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4110 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4111 (store_multiple_operation): Delete.
4112 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4113 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4114 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4115 guarded by TARGET_STRING.
4116 (rs6000_output_load_multiple): Delete.
4117 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4118 OPTION_MASK_STRING / TARGET_STRING handling.
4119 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4120 (const rs6000_opt_masks) <"string">: Change mask to 0.
4121 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4122 (MASK_STRING): Delete.
4123 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4125 (load_multiple): Delete.
4132 (store_multiple): Delete.
4139 (movmemsi_8reg): Delete.
4140 (corresponding unnamed define_insn): Delete.
4141 (movmemsi_6reg): Delete.
4142 (corresponding unnamed define_insn): Delete.
4143 (movmemsi_4reg): Delete.
4144 (corresponding unnamed define_insn): Delete.
4145 (movmemsi_2reg): Delete.
4146 (corresponding unnamed define_insn): Delete.
4147 (movmemsi_1reg): Delete.
4148 (corresponding unnamed define_insn): Delete.
4149 * config/rs6000/rs6000.opt (mno-string): New.
4150 (mstring): Replace by deprecation warning stub.
4151 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4153 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4155 * regrename.c (regrename_do_replace): If replacing the same
4156 reg multiple times, try to reuse last created gen_raw_REG.
4159 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4160 main to workaround a bug in GDB.
4162 2018-01-12 Tom de Vries <tom@codesourcery.com>
4165 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4167 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4169 PR rtl-optimization/80481
4170 * ira-color.c (get_cap_member): New function.
4171 (allocnos_conflict_by_live_ranges_p): Use it.
4172 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4173 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4175 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4178 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4179 (*saddl_se_1): Ditto.
4181 (*ssubl_se_1): Ditto.
4183 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4185 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4186 rather than wi::to_widest for DR_INITs.
4187 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4188 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4189 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4191 (vect_analyze_group_access_1): Note that here.
4193 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4195 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4196 polynomial type sizes.
4198 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4200 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4201 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4202 (gimple_add_tmp_var): Likewise.
4204 2018-01-12 Martin Liska <mliska@suse.cz>
4206 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4207 (gimple_alloc_sizes): Likewise.
4208 (dump_gimple_statistics): Use PRIu64 in printf format.
4209 * gimple.h: Change uint64_t to int.
4211 2018-01-12 Martin Liska <mliska@suse.cz>
4213 * tree-core.h: Use uint64_t instead of int.
4214 * tree.c (tree_node_counts): Likewise.
4215 (tree_node_sizes): Likewise.
4216 (dump_tree_statistics): Use PRIu64 in printf format.
4218 2018-01-12 Martin Liska <mliska@suse.cz>
4220 * Makefile.in: As qsort_chk is implemented in vec.c, add
4221 vec.o to linkage of gencfn-macros.
4222 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4223 passing the info to record_node_allocation_statistics.
4224 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4226 * ggc-common.c (struct ggc_usage): Add operator== and use
4227 it in operator< and compare function.
4228 * mem-stats.h (struct mem_usage): Likewise.
4229 * vec.c (struct vec_usage): Remove operator< and compare
4230 function. Can be simply inherited.
4232 2018-01-12 Martin Jambor <mjambor@suse.cz>
4235 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4236 * tree-ssa-math-opts.c: Include domwalk.h.
4237 (convert_mult_to_fma_1): New function.
4238 (fma_transformation_info): New type.
4239 (fma_deferring_state): Likewise.
4240 (cancel_fma_deferring): New function.
4241 (result_of_phi): Likewise.
4242 (last_fma_candidate_feeds_initial_phi): Likewise.
4243 (convert_mult_to_fma): Added deferring logic, split actual
4244 transformation to convert_mult_to_fma_1.
4245 (math_opts_dom_walker): New type.
4246 (math_opts_dom_walker::after_dom_children): New method, body moved
4247 here from pass_optimize_widening_mul::execute, added deferring logic
4249 (pass_optimize_widening_mul::execute): Moved most of code to
4250 math_opts_dom_walker::after_dom_children.
4251 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4252 * config/i386/i386.c (ix86_option_override_internal): Added
4253 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4255 2018-01-12 Richard Biener <rguenther@suse.de>
4258 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4259 inline instance vars.
4261 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4264 * config/rx/rx.c (rx_is_restricted_memory_address):
4267 2018-01-12 Richard Biener <rguenther@suse.de>
4269 PR tree-optimization/80846
4270 * target.def (split_reduction): New target hook.
4271 * targhooks.c (default_split_reduction): New function.
4272 * targhooks.h (default_split_reduction): Declare.
4273 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4274 target requests first reduce vectors by combining low and high
4276 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4277 (get_vectype_for_scalar_type_and_size): Export.
4278 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4279 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4280 * doc/tm.texi: Regenerate.
4281 * config/i386/i386.c (ix86_split_reduction): Implement
4282 TARGET_VECTORIZE_SPLIT_REDUCTION.
4284 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4287 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4288 in PIC mode except for TARGET_VXWORKS_RTP.
4289 * config/sparc/sparc.c: Include cfgrtl.h.
4290 (TARGET_INIT_PIC_REG): Define.
4291 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4292 (sparc_pic_register_p): New predicate.
4293 (sparc_legitimate_address_p): Use it.
4294 (sparc_legitimize_pic_address): Likewise.
4295 (sparc_delegitimize_address): Likewise.
4296 (sparc_mode_dependent_address_p): Likewise.
4297 (gen_load_pcrel_sym): Remove 4th parameter.
4298 (load_got_register): Adjust call to above. Remove obsolete stuff.
4299 (sparc_expand_prologue): Do not call load_got_register here.
4300 (sparc_flat_expand_prologue): Likewise.
4301 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4302 (sparc_use_pseudo_pic_reg): New function.
4303 (sparc_init_pic_reg): Likewise.
4304 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4305 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4307 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4309 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4310 Add item for branch_cost.
4312 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4314 PR rtl-optimization/83565
4315 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4316 not extend the result to a larger mode for rotate operations.
4317 (num_sign_bit_copies1): Likewise.
4319 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4322 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4324 Use values-Xc.o for -pedantic.
4325 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4327 2018-01-12 Martin Liska <mliska@suse.cz>
4330 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4332 (possible_polymorphic_call_targets): Use it.
4333 (ipa_devirt): Likewise.
4335 2018-01-12 Martin Liska <mliska@suse.cz>
4337 * profile-count.h (enum profile_quality): Use 0 as invalid
4338 enum value of profile_quality.
4340 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4342 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4343 -mext-string options.
4345 2018-01-12 Richard Biener <rguenther@suse.de>
4347 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4348 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4349 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4351 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4353 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4355 * configure.ac (--with-long-double-format): Add support for the
4356 configuration option to change the default long double format on
4358 * config.gcc (powerpc*-linux*-*): Likewise.
4359 * configure: Regenerate.
4360 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4361 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4362 used without modification.
4364 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4366 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4367 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4368 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4369 MISC_BUILTIN_SPEC_BARRIER.
4370 (rs6000_init_builtins): Likewise.
4371 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4373 (speculation_barrier): New define_insn.
4374 * doc/extend.texi: Document __builtin_speculation_barrier.
4376 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4379 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4380 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4381 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4383 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4384 integral modes instead of "ss" and "sd".
4385 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4386 vectors with 32-bit and 64-bit elements.
4387 (vecdupssescalarmodesuffix): New mode attribute.
4388 (vec_dup<mode>): Use it.
4390 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4393 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4394 frame if argument is passed on stack.
4396 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4399 * ree.c (combine_reaching_defs): Optimize also
4400 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4401 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4403 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4406 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4408 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4411 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4412 after they are computed.
4414 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4416 PR tree-optimization/83695
4417 * gimple-loop-linterchange.cc
4418 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4419 reset cached scev information after interchange.
4420 (pass_linterchange::execute): Remove call to scev_reset_htab.
4422 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4424 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4425 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4426 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4427 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4428 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4429 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4430 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4431 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4432 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4433 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4434 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4435 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4436 (V_lane_reg): Likewise.
4437 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4439 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4440 (vfmal_lane_low<mode>_intrinsic,
4441 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4442 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4443 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4444 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4445 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4446 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4448 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4450 * config/arm/arm-cpus.in (fp16fml): New feature.
4451 (ALL_SIMD): Add fp16fml.
4452 (armv8.2-a): Add fp16fml as an option.
4453 (armv8.3-a): Likewise.
4454 (armv8.4-a): Add fp16fml as part of fp16.
4455 * config/arm/arm.h (TARGET_FP16FML): Define.
4456 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4458 * config/arm/arm-modes.def (V2HF): Define.
4459 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4460 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4461 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4462 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4463 vfmsl_low, vfmsl_high): New set of builtins.
4464 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4465 (vfml_op): New code attribute.
4466 (VFMLHALVES): New int iterator.
4467 (VFML, VFMLSEL): New mode attributes.
4468 (V_reg): Define mapping for V2HF.
4469 (V_hi, V_lo): New mode attributes.
4470 (VF_constraint): Likewise.
4471 (vfml_half, vfml_half_selector): New int attributes.
4472 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4474 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4475 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4477 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4478 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4479 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4480 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4482 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4483 Document new effective target and option set.
4485 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4487 * config/arm/arm-cpus.in (armv8_4): New feature.
4488 (ARMv8_4a): New fgroup.
4489 (armv8.4-a): New arch.
4490 * config/arm/arm-tables.opt: Regenerate.
4491 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4492 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4493 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4494 Add matching rules for -march=armv8.4-a and extensions.
4495 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4497 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4500 * config/rx/rx.md (BW): New mode attribute.
4501 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4503 2018-01-11 Richard Biener <rguenther@suse.de>
4505 PR tree-optimization/83435
4506 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4507 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4508 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4510 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4511 Alan Hayward <alan.hayward@arm.com>
4512 David Sherwood <david.sherwood@arm.com>
4514 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4516 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4517 (aarch64_print_address_internal): Use it to check for a zero offset.
4519 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4520 Alan Hayward <alan.hayward@arm.com>
4521 David Sherwood <david.sherwood@arm.com>
4523 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4524 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4525 Return a poly_int64 rather than a HOST_WIDE_INT.
4526 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4527 rather than a HOST_WIDE_INT.
4528 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4529 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4530 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4531 final_offset from HOST_WIDE_INT to poly_int64.
4532 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4533 to_constant when getting the number of units in an Advanced SIMD
4535 (aarch64_builtin_vectorized_function): Check for a constant number
4537 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4539 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4540 attribute instead of GET_MODE_NUNITS.
4541 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4542 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4543 GET_MODE_SIZE for fixed-size registers.
4544 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4545 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4546 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4547 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4548 (aarch64_print_operand, aarch64_print_address_internal)
4549 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4550 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4551 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4552 Handle polynomial GET_MODE_SIZE.
4553 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4554 wider than SImode without modification.
4555 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4556 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4557 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4558 passing and returning SVE modes.
4559 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4560 rather than GEN_INT.
4561 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4562 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4563 (aarch64_allocate_and_probe_stack_space): Likewise.
4564 (aarch64_layout_frame): Cope with polynomial offsets.
4565 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4566 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4568 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4569 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4570 poly_int64 rather than a HOST_WIDE_INT.
4571 (aarch64_get_separate_components, aarch64_process_components)
4572 (aarch64_expand_prologue, aarch64_expand_epilogue)
4573 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4574 (aarch64_anchor_offset): New function, split out from...
4575 (aarch64_legitimize_address): ...here.
4576 (aarch64_builtin_vectorization_cost): Handle polynomial
4577 TYPE_VECTOR_SUBPARTS.
4578 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4580 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4581 number of elements from the PARALLEL rather than the mode.
4582 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4583 rather than GET_MODE_BITSIZE.
4584 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4585 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4586 (aarch64_expand_vec_perm_const_1): Handle polynomial
4587 d->perm.length () and d->perm elements.
4588 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4589 Apply to_constant to d->perm elements.
4590 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4591 polynomial CONST_VECTOR_NUNITS.
4592 (aarch64_move_pointer): Take amount as a poly_int64 rather
4594 (aarch64_progress_pointer): Avoid temporary variable.
4595 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4596 the mode attribute instead of GET_MODE.
4598 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4599 Alan Hayward <alan.hayward@arm.com>
4600 David Sherwood <david.sherwood@arm.com>
4602 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4603 x exists before using it.
4604 (aarch64_add_constant_internal): Rename to...
4605 (aarch64_add_offset_1): ...this. Replace regnum with separate
4606 src and dest rtxes. Handle the case in which they're different,
4607 including when the offset is zero. Replace scratchreg with an rtx.
4608 Use 2 additions if there is no spare register into which we can
4609 move a 16-bit constant.
4610 (aarch64_add_constant): Delete.
4611 (aarch64_add_offset): Replace reg with separate src and dest
4612 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4613 Use aarch64_add_offset_1.
4614 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4615 an rtx rather than an int. Take the delta as a poly_int64
4616 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4617 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4618 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4619 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4620 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4622 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4623 aarch64_add_constant.
4625 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4627 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4628 Use scalar_float_mode.
4630 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4632 * config/aarch64/aarch64-simd.md
4633 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4634 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4635 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4636 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4637 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4638 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4639 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4640 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4641 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4642 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4644 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4647 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4648 targ_options->x_arm_arch_string is non NULL.
4650 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4652 * config/aarch64/aarch64.h
4653 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4655 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4658 * expmed.c (emit_store_flag_force): Swap if const op0
4659 and change VOIDmode to mode of op0.
4661 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4663 PR rtl-optimization/83761
4664 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4665 than bytes to mode_for_size.
4667 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4670 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4671 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4674 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4677 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4678 when in layout mode.
4679 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4680 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4683 2018-01-10 Michael Collison <michael.collison@arm.com>
4685 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4686 * config/aarch64/aarch64-option-extension.def: Add
4687 AARCH64_OPT_EXTENSION of 'fp16fml'.
4688 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4689 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4690 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4691 * config/aarch64/constraints.md (Ui7): New constraint.
4692 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4693 (VFMLA_SEL_W): Ditto.
4696 (VFMLA16_LOW): New int iterator.
4697 (VFMLA16_HIGH): Ditto.
4698 (UNSPEC_FMLAL): New unspec.
4699 (UNSPEC_FMLSL): Ditto.
4700 (UNSPEC_FMLAL2): Ditto.
4701 (UNSPEC_FMLSL2): Ditto.
4702 (f16mac): New code attribute.
4703 * config/aarch64/aarch64-simd-builtins.def
4704 (aarch64_fmlal_lowv2sf): Ditto.
4705 (aarch64_fmlsl_lowv2sf): Ditto.
4706 (aarch64_fmlalq_lowv4sf): Ditto.
4707 (aarch64_fmlslq_lowv4sf): Ditto.
4708 (aarch64_fmlal_highv2sf): Ditto.
4709 (aarch64_fmlsl_highv2sf): Ditto.
4710 (aarch64_fmlalq_highv4sf): Ditto.
4711 (aarch64_fmlslq_highv4sf): Ditto.
4712 (aarch64_fmlal_lane_lowv2sf): Ditto.
4713 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4714 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4715 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4716 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4717 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4718 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4719 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4720 (aarch64_fmlal_lane_highv2sf): Ditto.
4721 (aarch64_fmlsl_lane_highv2sf): Ditto.
4722 (aarch64_fmlal_laneq_highv2sf): Ditto.
4723 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4724 (aarch64_fmlalq_lane_highv4sf): Ditto.
4725 (aarch64_fmlsl_lane_highv4sf): Ditto.
4726 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4727 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4728 * config/aarch64/aarch64-simd.md:
4729 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4730 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4731 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4732 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4733 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4734 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4735 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4736 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4737 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4738 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4739 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4740 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4741 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4742 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4743 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4744 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4745 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4746 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4747 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4748 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4749 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4750 (vfmlsl_low_u32): Ditto.
4751 (vfmlalq_low_u32): Ditto.
4752 (vfmlslq_low_u32): Ditto.
4753 (vfmlal_high_u32): Ditto.
4754 (vfmlsl_high_u32): Ditto.
4755 (vfmlalq_high_u32): Ditto.
4756 (vfmlslq_high_u32): Ditto.
4757 (vfmlal_lane_low_u32): Ditto.
4758 (vfmlsl_lane_low_u32): Ditto.
4759 (vfmlal_laneq_low_u32): Ditto.
4760 (vfmlsl_laneq_low_u32): Ditto.
4761 (vfmlalq_lane_low_u32): Ditto.
4762 (vfmlslq_lane_low_u32): Ditto.
4763 (vfmlalq_laneq_low_u32): Ditto.
4764 (vfmlslq_laneq_low_u32): Ditto.
4765 (vfmlal_lane_high_u32): Ditto.
4766 (vfmlsl_lane_high_u32): Ditto.
4767 (vfmlal_laneq_high_u32): Ditto.
4768 (vfmlsl_laneq_high_u32): Ditto.
4769 (vfmlalq_lane_high_u32): Ditto.
4770 (vfmlslq_lane_high_u32): Ditto.
4771 (vfmlalq_laneq_high_u32): Ditto.
4772 (vfmlslq_laneq_high_u32): Ditto.
4773 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4774 (AARCH64_FL_FOR_ARCH8_4): New.
4775 (AARCH64_ISA_F16FML): New ISA flag.
4776 (TARGET_F16FML): New feature flag for fp16fml.
4777 (doc/invoke.texi): Document new fp16fml option.
4779 2018-01-10 Michael Collison <michael.collison@arm.com>
4781 * config/aarch64/aarch64-builtins.c:
4782 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4783 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4784 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4785 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4786 (AARCH64_ISA_SHA3): New ISA flag.
4787 (TARGET_SHA3): New feature flag for sha3.
4788 * config/aarch64/iterators.md (sha512_op): New int attribute.
4789 (CRYPTO_SHA512): New int iterator.
4790 (UNSPEC_SHA512H): New unspec.
4791 (UNSPEC_SHA512H2): Ditto.
4792 (UNSPEC_SHA512SU0): Ditto.
4793 (UNSPEC_SHA512SU1): Ditto.
4794 * config/aarch64/aarch64-simd-builtins.def
4795 (aarch64_crypto_sha512hqv2di): New builtin.
4796 (aarch64_crypto_sha512h2qv2di): Ditto.
4797 (aarch64_crypto_sha512su0qv2di): Ditto.
4798 (aarch64_crypto_sha512su1qv2di): Ditto.
4799 (aarch64_eor3qv8hi): Ditto.
4800 (aarch64_rax1qv2di): Ditto.
4801 (aarch64_xarqv2di): Ditto.
4802 (aarch64_bcaxqv8hi): Ditto.
4803 * config/aarch64/aarch64-simd.md:
4804 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4805 (aarch64_crypto_sha512su0qv2di): Ditto.
4806 (aarch64_crypto_sha512su1qv2di): Ditto.
4807 (aarch64_eor3qv8hi): Ditto.
4808 (aarch64_rax1qv2di): Ditto.
4809 (aarch64_xarqv2di): Ditto.
4810 (aarch64_bcaxqv8hi): Ditto.
4811 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4812 (vsha512h2q_u64): Ditto.
4813 (vsha512su0q_u64): Ditto.
4814 (vsha512su1q_u64): Ditto.
4815 (veor3q_u16): Ditto.
4816 (vrax1q_u64): Ditto.
4818 (vbcaxq_u16): Ditto.
4819 * config/arm/types.md (crypto_sha512): New type attribute.
4820 (crypto_sha3): Ditto.
4821 (doc/invoke.texi): Document new sha3 option.
4823 2018-01-10 Michael Collison <michael.collison@arm.com>
4825 * config/aarch64/aarch64-builtins.c:
4826 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4827 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4828 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4829 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4830 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4831 (AARCH64_ISA_SM4): New ISA flag.
4832 (TARGET_SM4): New feature flag for sm4.
4833 * config/aarch64/aarch64-simd-builtins.def
4834 (aarch64_sm3ss1qv4si): Ditto.
4835 (aarch64_sm3tt1aq4si): Ditto.
4836 (aarch64_sm3tt1bq4si): Ditto.
4837 (aarch64_sm3tt2aq4si): Ditto.
4838 (aarch64_sm3tt2bq4si): Ditto.
4839 (aarch64_sm3partw1qv4si): Ditto.
4840 (aarch64_sm3partw2qv4si): Ditto.
4841 (aarch64_sm4eqv4si): Ditto.
4842 (aarch64_sm4ekeyqv4si): Ditto.
4843 * config/aarch64/aarch64-simd.md:
4844 (aarch64_sm3ss1qv4si): Ditto.
4845 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4846 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4847 (aarch64_sm4eqv4si): Ditto.
4848 (aarch64_sm4ekeyqv4si): Ditto.
4849 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4850 (sm3part_op): Ditto.
4851 (CRYPTO_SM3TT): Ditto.
4852 (CRYPTO_SM3PART): Ditto.
4853 (UNSPEC_SM3SS1): New unspec.
4854 (UNSPEC_SM3TT1A): Ditto.
4855 (UNSPEC_SM3TT1B): Ditto.
4856 (UNSPEC_SM3TT2A): Ditto.
4857 (UNSPEC_SM3TT2B): Ditto.
4858 (UNSPEC_SM3PARTW1): Ditto.
4859 (UNSPEC_SM3PARTW2): Ditto.
4860 (UNSPEC_SM4E): Ditto.
4861 (UNSPEC_SM4EKEY): Ditto.
4862 * config/aarch64/constraints.md (Ui2): New constraint.
4863 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4864 * config/arm/types.md (crypto_sm3): New type attribute.
4865 (crypto_sm4): Ditto.
4866 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4867 (vsm3tt1aq_u32): Ditto.
4868 (vsm3tt1bq_u32): Ditto.
4869 (vsm3tt2aq_u32): Ditto.
4870 (vsm3tt2bq_u32): Ditto.
4871 (vsm3partw1q_u32): Ditto.
4872 (vsm3partw2q_u32): Ditto.
4873 (vsm4eq_u32): Ditto.
4874 (vsm4ekeyq_u32): Ditto.
4875 (doc/invoke.texi): Document new sm4 option.
4877 2018-01-10 Michael Collison <michael.collison@arm.com>
4879 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4880 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4881 (AARCH64_FL_FOR_ARCH8_4): New.
4882 (AARCH64_FL_V8_4): New flag.
4883 (doc/invoke.texi): Document new armv8.4-a option.
4885 2018-01-10 Michael Collison <michael.collison@arm.com>
4887 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4888 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4889 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4890 * config/aarch64/aarch64-option-extension.def: Add
4891 AARCH64_OPT_EXTENSION of 'sha2'.
4892 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4893 (crypto): Disable sha2 and aes if crypto disabled.
4894 (crypto): Enable aes and sha2 if enabled.
4895 (simd): Disable sha2 and aes if simd disabled.
4896 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4898 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4899 (TARGET_SHA2): New feature flag for sha2.
4900 (TARGET_AES): New feature flag for aes.
4901 * config/aarch64/aarch64-simd.md:
4902 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4903 conditional on TARGET_AES.
4904 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4905 (aarch64_crypto_sha1hsi): Make pattern conditional
4907 (aarch64_crypto_sha1hv4si): Ditto.
4908 (aarch64_be_crypto_sha1hv4si): Ditto.
4909 (aarch64_crypto_sha1su1v4si): Ditto.
4910 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4911 (aarch64_crypto_sha1su0v4si): Ditto.
4912 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4913 (aarch64_crypto_sha256su0v4si): Ditto.
4914 (aarch64_crypto_sha256su1v4si): Ditto.
4915 (doc/invoke.texi): Document new aes and sha2 options.
4917 2018-01-10 Martin Sebor <msebor@redhat.com>
4919 PR tree-optimization/83781
4920 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4923 2018-01-11 Martin Sebor <msebor@gmail.com>
4924 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4926 PR tree-optimization/83501
4927 PR tree-optimization/81703
4929 * tree-ssa-strlen.c (get_string_cst): Rename...
4930 (get_string_len): ...to this. Handle global constants.
4931 (handle_char_store): Adjust.
4933 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4934 Jim Wilson <jimw@sifive.com>
4936 * config/riscv/riscv-protos.h (riscv_output_return): New.
4937 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4938 (riscv_attribute_table, riscv_output_return),
4939 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4940 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4941 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4943 (riscv_expand_prologue): Add early return for naked function.
4944 (riscv_expand_epilogue): Likewise.
4945 (riscv_function_ok_for_sibcall): Return false for naked function.
4946 (riscv_set_current_function): New.
4947 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4948 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4949 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4950 * doc/extend.texi (RISC-V Function Attributes): New.
4952 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4954 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4955 check for 128-bit long double before checking TCmode.
4956 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4957 128-bit long doubles before checking TFmode or TCmode.
4958 (FLOAT128_IBM_P): Likewise.
4960 2018-01-10 Martin Sebor <msebor@redhat.com>
4962 PR tree-optimization/83671
4963 * builtins.c (c_strlen): Unconditionally return zero for the empty
4965 Use -Warray-bounds for warnings.
4966 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4967 for non-constant array indices with COMPONENT_REF, arrays of
4968 arrays, and pointers to arrays.
4969 (gimple_fold_builtin_strlen): Determine and set length range for
4970 non-constant character arrays.
4972 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4975 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4978 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4980 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4982 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4985 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4986 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4987 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4988 indexed_or_indirect_operand predicate.
4989 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4990 (*vsx_le_perm_load_v8hi): Likewise.
4991 (*vsx_le_perm_load_v16qi): Likewise.
4992 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4993 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4994 (*vsx_le_perm_store_v8hi): Likewise.
4995 (*vsx_le_perm_store_v16qi): Likewise.
4996 (eight unnamed splitters): Likewise.
4998 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5000 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
5001 * config/rs6000/emmintrin.h: Likewise.
5002 * config/rs6000/mmintrin.h: Likewise.
5003 * config/rs6000/xmmintrin.h: Likewise.
5005 2018-01-10 David Malcolm <dmalcolm@redhat.com>
5008 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
5010 * tree.c (tree_nop_conversion): Return true for location wrapper
5012 (maybe_wrap_with_location): New function.
5013 (selftest::check_strip_nops): New function.
5014 (selftest::test_location_wrappers): New function.
5015 (selftest::tree_c_tests): Call it.
5016 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
5017 (maybe_wrap_with_location): New decl.
5018 (EXPR_LOCATION_WRAPPER_P): New macro.
5019 (location_wrapper_p): New inline function.
5020 (tree_strip_any_location_wrapper): New inline function.
5022 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
5025 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5026 stack_realign_offset for the largest alignment of stack slot
5028 (ix86_find_max_used_stack_alignment): New function.
5029 (ix86_finalize_stack_frame_flags): Use it. Set
5030 max_used_stack_alignment if we don't realign stack.
5031 * config/i386/i386.h (machine_function): Add
5032 max_used_stack_alignment.
5034 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5036 * config/arm/arm.opt (-mbranch-cost): New option.
5037 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5040 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5043 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5044 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5046 2018-01-10 Richard Biener <rguenther@suse.de>
5049 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5050 early out so it also covers the case where we have a non-NULL
5053 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5055 PR tree-optimization/83753
5056 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5057 for non-strided grouped accesses if the number of elements is 1.
5059 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5062 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5063 * i386.h (TARGET_USE_GATHER): Define.
5064 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5066 2018-01-10 Martin Liska <mliska@suse.cz>
5069 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5070 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5072 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5073 CLEANUP_NO_PARTITIONING is not set.
5075 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5077 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5078 for vectors, as a partial revert of r254296.
5079 * rtl.h (const_vec_p): Delete.
5080 (const_vec_duplicate_p): Don't test for vector CONSTs.
5081 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5082 * expmed.c (make_tree): Likewise.
5085 * common.md (E, F): Use CONSTANT_P instead of checking for
5087 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5088 checking for CONST_VECTOR.
5090 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5093 * predict.c (force_edge_cold): Handle in more sane way edges
5096 2018-01-09 Carl Love <cel@us.ibm.com>
5098 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5100 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5101 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5102 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5103 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5104 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5105 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5106 * config/rs6000/rs6000-protos.h: Add extern defition for
5107 rs6000_generate_float2_double_code.
5108 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5110 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5111 (float2_v2df): Add define_expand.
5113 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5116 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5117 op_mode in the force_to_mode call.
5119 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5121 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5122 instead of checking each element individually.
5123 (aarch64_evpc_uzp): Likewise.
5124 (aarch64_evpc_zip): Likewise.
5125 (aarch64_evpc_ext): Likewise.
5126 (aarch64_evpc_rev): Likewise.
5127 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5128 instead of checking each element individually. Return true without
5130 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5131 whether all selected elements come from the same input, instead of
5132 checking each element individually. Remove calls to gen_rtx_REG,
5133 start_sequence and end_sequence and instead assert that no rtl is
5136 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5138 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5139 order of HIGH and CONST checks.
5141 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5143 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5144 if the destination isn't an SSA_NAME.
5146 2018-01-09 Richard Biener <rguenther@suse.de>
5148 PR tree-optimization/83668
5149 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5151 (canonicalize_loop_form): ... here, renamed from ...
5152 (canonicalize_loop_closed_ssa_form): ... this and amended to
5153 swap successor edges for loop exit blocks to make us use
5154 the RPO order we need for initial schedule generation.
5156 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5158 PR tree-optimization/64811
5159 * match.pd: When optimizing comparisons with Inf, avoid
5160 introducing or losing exceptions from comparisons with NaN.
5162 2018-01-09 Martin Liska <mliska@suse.cz>
5165 * asan.c (shadow_mem_size): Add gcc_assert.
5167 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5169 Don't save registers in main().
5172 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5173 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5174 * config/avr/avr.c (avr_set_current_function): Don't error if
5175 naked, OS_task or OS_main are specified at the same time.
5176 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5178 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5180 * common/config/avr/avr-common.c (avr_option_optimization_table):
5181 Switch on -mmain-is-OS_task for optimizing compilations.
5183 2018-01-09 Richard Biener <rguenther@suse.de>
5185 PR tree-optimization/83572
5186 * graphite.c: Include cfganal.h.
5187 (graphite_transform_loops): Connect infinite loops to exit
5188 and remove fake edges at the end.
5190 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5192 * ipa-inline.c (edge_badness): Revert accidental checkin.
5194 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5197 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5198 symbols; not inline clones.
5200 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5203 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5204 hard registers. Formatting fixes.
5206 PR preprocessor/83722
5207 * gcc.c (try_generate_repro): Pass
5208 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5209 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5212 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5213 Kito Cheng <kito.cheng@gmail.com>
5215 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5216 (riscv_leaf_function_p): Delete.
5217 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5219 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5221 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5223 (do_ifelse): New function.
5224 (do_isel): New function.
5225 (do_sub3): New function.
5226 (do_add3): New function.
5227 (do_load_mask_compare): New function.
5228 (do_overlap_load_compare): New function.
5229 (expand_compare_loop): New function.
5230 (expand_block_compare): Call expand_compare_loop() when appropriate.
5231 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5233 (-mblock-compare-inline-loop-limit): New option.
5235 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5238 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5239 Reverse order of second and third operands in first alternative.
5240 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5241 of first and second elements in UNSPEC_VPERMR vector.
5242 (altivec_expand_vec_perm_le): Likewise.
5244 2018-01-08 Jeff Law <law@redhat.com>
5246 PR rtl-optimizatin/81308
5247 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5248 (process_switch): If group_case_labels makes a change, then set
5250 (pass_convert_switch::execute): If a switch is converted, then
5251 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5253 PR rtl-optimization/81308
5254 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5257 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5259 PR target/83663 - Revert r255946
5260 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5261 generation for cases where splatting a value is not useful.
5262 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5263 across a vec_duplicate and a paradoxical subreg forming a vector
5264 mode to a vec_concat.
5266 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5268 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5269 -march=armv8.3-a variants.
5270 * config/arm/t-multilib: Likewise.
5271 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5273 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5275 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5277 (cceq_ior_compare_complement): Give it a name so I can use it, and
5278 change boolean_or_operator predicate to boolean_operator so it can
5279 be used to generate a crand.
5280 (eqne): New code iterator.
5281 (bd/bd_neg): New code_attrs.
5282 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5283 a single define_insn.
5284 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5285 decrement (bdnzt/bdnzf/bdzt/bdzf).
5286 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5287 with the new names of the branch decrement patterns, and added the
5288 names of the branch decrement conditional patterns.
5290 2018-01-08 Richard Biener <rguenther@suse.de>
5292 PR tree-optimization/83563
5293 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5296 2018-01-08 Richard Biener <rguenther@suse.de>
5299 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5301 2018-01-08 Richard Biener <rguenther@suse.de>
5303 PR tree-optimization/83685
5304 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5305 references to abnormals.
5307 2018-01-08 Richard Biener <rguenther@suse.de>
5310 * dwarf2out.c (output_indirect_strings): Handle empty
5311 skeleton_debug_str_hash.
5312 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5314 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5316 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5317 (emit_store_direct): Likewise.
5318 (arc_trampoline_adjust_address): Likewise.
5319 (arc_asm_trampoline_template): New function.
5320 (arc_initialize_trampoline): Use asm_trampoline_template.
5321 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5322 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5323 * config/arc/arc.md (flush_icache): Delete pattern.
5325 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5327 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5328 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5331 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5334 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5335 by not USED_FOR_TARGET.
5336 (make_pass_resolve_sw_modes): Likewise.
5338 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5340 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5343 2018-01-08 Richard Biener <rguenther@suse.de>
5346 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5348 2018-01-08 Richard Biener <rguenther@suse.de>
5351 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5353 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5356 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5357 basic blocks with a small number of successors.
5358 (convert_control_dep_chain_into_preds): Improve handling of
5360 (dump_predicates): Split apart into...
5361 (dump_pred_chain): ...here...
5362 (dump_pred_info): ...and here.
5363 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5364 (can_chain_union_be_invalidated_p): Improve check for invalidation
5366 (uninit_uses_cannot_happen): Avoid unnecessary if
5367 convert_control_dep_chain_into_preds yielded nothing.
5369 2018-01-06 Martin Sebor <msebor@redhat.com>
5371 PR tree-optimization/83640
5372 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5373 subtracting negative offset from size.
5374 (builtin_access::overlap): Adjust offset bounds of the access to fall
5375 within the size of the object if possible.
5377 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5379 PR rtl-optimization/83699
5380 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5381 extract_bit_field_as_subreg to cases in which the extracted
5382 value is also a vector.
5384 * lra-constraints.c (process_alt_operands): Test for the equivalence
5385 substitutions when detecting a possible reload cycle.
5387 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5390 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5391 by default if flag_selective_schedling{,2}. Formatting fixes.
5393 PR rtl-optimization/83682
5394 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5395 if it has non-VECTOR_MODE element mode.
5396 (vec_duplicate_p): Likewise.
5399 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5400 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5402 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5405 * config/i386/i386-builtin.def
5406 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5407 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5408 Require also OPTION_MASK_ISA_AVX512F in addition to
5409 OPTION_MASK_ISA_GFNI.
5410 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5411 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5412 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5413 to OPTION_MASK_ISA_GFNI.
5414 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5415 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5416 OPTION_MASK_ISA_AVX512BW.
5417 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5418 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5419 addition to OPTION_MASK_ISA_GFNI.
5420 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5421 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5422 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5423 to OPTION_MASK_ISA_GFNI.
5424 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5425 a requirement for all ISAs rather than any of them with a few
5427 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5429 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5430 bitmasks to be enabled with 3 exceptions, instead of requiring any
5431 enabled ISA with lots of exceptions.
5432 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5433 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5434 Change avx512bw in isa attribute to avx512f.
5435 * config/i386/sgxintrin.h: Add license boilerplate.
5436 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5437 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5438 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5439 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5441 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5442 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5443 temporarily sse2 rather than sse if not enabled already.
5446 * config/i386/sse.md (VI248_VLBW): Rename to ...
5447 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5448 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5449 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5450 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5451 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5452 mode iterator instead of VI248_VLBW.
5454 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5456 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5457 (record_modified): Skip clobbers; add debug output.
5458 (param_change_prob): Use sreal frequencies.
5460 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5462 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5463 punt for user-aligned variables.
5465 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5467 * tree-chrec.c (chrec_contains_symbols): Return true for
5470 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5473 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5474 of (x|y) == x for BICS pattern.
5476 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5478 PR tree-optimization/83605
5479 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5480 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5483 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5485 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5486 * config/epiphany/rtems.h: New file.
5488 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5489 Uros Bizjak <ubizjak@gmail.com>
5492 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5493 QIreg_operand instead of register_operand predicate.
5494 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5495 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5496 comments instead of -fmitigate[-_]rop.
5498 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5501 * cgraphunit.c (symbol_table::compile): Switch to text_section
5502 before calling assembly_start debug hook.
5503 * run-rtl-passes.c (run_rtl_passes): Likewise.
5506 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5508 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5509 range_int_cst_p rather than !symbolic_range_p before calling
5510 extract_range_from_multiplicative_op_1.
5512 2018-01-04 Jeff Law <law@redhat.com>
5514 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5515 redundant test in assertion.
5517 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5519 * doc/rtl.texi: Document machine_mode wrapper classes.
5521 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5523 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5526 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5528 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5529 the VEC_PERM_EXPR fold to fail.
5531 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5534 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5535 to switched_sections.
5537 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5540 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5543 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5546 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5547 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5549 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5552 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5553 is BLKmode and bitpos not zero or mode change is needed.
5555 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5558 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5561 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5564 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5565 instead of MULT rtx. Update all corresponding splitters.
5567 (*ssub<modesuffix>): Ditto.
5569 (*cmp_sadd_di): Update split patterns.
5570 (*cmp_sadd_si): Ditto.
5571 (*cmp_sadd_sidi): Ditto.
5572 (*cmp_ssub_di): Ditto.
5573 (*cmp_ssub_si): Ditto.
5574 (*cmp_ssub_sidi): Ditto.
5575 * config/alpha/predicates.md (const23_operand): New predicate.
5576 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5577 Look for ASHIFT, not MULT inner operand.
5578 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5580 2018-01-04 Martin Liska <mliska@suse.cz>
5582 PR gcov-profile/83669
5583 * gcov.c (output_intermediate_file): Add version to intermediate
5585 * doc/gcov.texi: Document new field 'version' in intermediate
5586 file format. Fix location of '-k' option of gcov command.
5588 2018-01-04 Martin Liska <mliska@suse.cz>
5591 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5593 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5595 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5597 2018-01-03 Martin Sebor <msebor@redhat.com>
5599 PR tree-optimization/83655
5600 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5601 checking calls with invalid arguments.
5603 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5605 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5606 (vectorizable_mask_load_store): Delete.
5607 (vectorizable_call): Return false for masked loads and stores.
5608 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5609 instead of gimple_assign_rhs1.
5610 (vectorizable_load): Handle IFN_MASK_LOAD.
5611 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5613 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5615 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5617 (vectorizable_mask_load_store): ...here.
5618 (vectorizable_load): ...and here.
5620 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5622 * tree-vect-stmts.c (vect_build_all_ones_mask)
5623 (vect_build_zero_merge_argument): New functions, split out from...
5624 (vectorizable_load): ...here.
5626 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5628 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5630 (vectorizable_mask_load_store): ...here.
5631 (vectorizable_store): ...and here.
5633 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5635 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5637 (vectorizable_mask_load_store): ...here.
5639 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5641 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5642 (vect_model_store_cost): Take a vec_load_store_type instead of a
5644 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5645 (vect_model_store_cost): Take a vec_load_store_type instead of a
5647 (vectorizable_mask_load_store): Update accordingly.
5648 (vectorizable_store): Likewise.
5649 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5651 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5653 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5654 IFN_MASK_LOAD calls here rather than...
5655 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5657 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5658 Alan Hayward <alan.hayward@arm.com>
5659 David Sherwood <david.sherwood@arm.com>
5661 * expmed.c (extract_bit_field_1): For vector extracts,
5662 fall back to extract_bit_field_as_subreg if vec_extract
5665 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5666 Alan Hayward <alan.hayward@arm.com>
5667 David Sherwood <david.sherwood@arm.com>
5669 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5670 they are variable or constant sized.
5671 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5672 slots for constant-sized data.
5674 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5675 Alan Hayward <alan.hayward@arm.com>
5676 David Sherwood <david.sherwood@arm.com>
5678 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5679 handling COND_EXPRs with boolean comparisons, try to find a better
5680 basis for the mask type than the boolean itself.
5682 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5684 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5685 is calculated and how it can be overridden.
5686 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5687 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5689 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5692 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5693 Alan Hayward <alan.hayward@arm.com>
5694 David Sherwood <david.sherwood@arm.com>
5696 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5697 Remove the mode argument.
5698 (aarch64_simd_valid_immediate): Remove the mode and inverse
5700 * config/aarch64/iterators.md (bitsize): New iterator.
5701 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5702 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5703 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5704 aarch64_simd_valid_immediate.
5705 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5706 (aarch64_reg_or_bic_imm): Likewise.
5707 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5708 with an insn_type enum and msl with a modifier_type enum.
5709 Replace element_width with a scalar_mode. Change the shift
5710 to unsigned int. Add constructors for scalar_float_mode and
5711 scalar_int_mode elements.
5712 (aarch64_vect_float_const_representable_p): Delete.
5713 (aarch64_can_const_movi_rtx_p)
5714 (aarch64_simd_scalar_immediate_valid_for_move)
5715 (aarch64_simd_make_constant): Update call to
5716 aarch64_simd_valid_immediate.
5717 (aarch64_advsimd_valid_immediate_hs): New function.
5718 (aarch64_advsimd_valid_immediate): Likewise.
5719 (aarch64_simd_valid_immediate): Remove mode and inverse
5720 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5721 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5722 and aarch64_float_const_representable_p on the result.
5723 (aarch64_output_simd_mov_immediate): Remove mode argument.
5724 Update call to aarch64_simd_valid_immediate and use of
5725 simd_immediate_info.
5726 (aarch64_output_scalar_simd_mov_immediate): Update call
5729 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5730 Alan Hayward <alan.hayward@arm.com>
5731 David Sherwood <david.sherwood@arm.com>
5733 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5734 (mode_nunits): Likewise CONST_MODE_NUNITS.
5735 * machmode.def (ADJUST_NUNITS): Document.
5736 * genmodes.c (mode_data::need_nunits_adj): New field.
5737 (blank_mode): Update accordingly.
5738 (adj_nunits): New variable.
5739 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5741 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5742 listed in adj_nunits.
5743 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5744 listed in adj_nunits. Don't emit case statements for such modes.
5745 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5746 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5747 nothing if adj_nunits is nonnull.
5748 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5749 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5750 (emit_mode_fbit): Update use of print_maybe_const_decl.
5751 (emit_move_size): Likewise. Treat the array as non-const
5753 (emit_mode_adjustments): Handle adj_nunits.
5755 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5757 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5758 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5759 (VECTOR_MODES): Use it.
5760 (make_vector_modes): Take the prefix as an argument.
5762 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5763 Alan Hayward <alan.hayward@arm.com>
5764 David Sherwood <david.sherwood@arm.com>
5766 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5767 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5768 for MODE_VECTOR_BOOL.
5769 * machmode.def (VECTOR_BOOL_MODE): Document.
5770 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5771 (make_vector_bool_mode): New function.
5772 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5774 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5775 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5777 * stor-layout.c (int_mode_for_mode): Likewise.
5778 * tree.c (build_vector_type_for_mode): Likewise.
5779 * varasm.c (output_constant_pool_2): Likewise.
5780 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5781 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5782 for MODE_VECTOR_BOOL.
5783 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5784 of mode class checks.
5785 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5786 instead of a list of mode class checks.
5787 (expand_vector_scalar_condition): Likewise.
5788 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5790 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5791 Alan Hayward <alan.hayward@arm.com>
5792 David Sherwood <david.sherwood@arm.com>
5794 * machmode.h (mode_size): Change from unsigned short to
5796 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5797 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5798 or if measurement_type is not polynomial.
5799 (fixed_size_mode::includes_p): Check for constant-sized modes.
5800 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5801 return a poly_uint16 rather than an unsigned short.
5802 (emit_mode_size): Change the type of mode_size from unsigned short
5803 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5804 (emit_mode_adjustments): Cope with polynomial vector sizes.
5805 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5807 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5809 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5810 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5811 * caller-save.c (setup_save_areas): Likewise.
5812 (replace_reg_with_saved_mem): Likewise.
5813 * calls.c (emit_library_call_value_1): Likewise.
5814 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5815 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5816 (gen_lowpart_for_combine): Likewise.
5817 * convert.c (convert_to_integer_1): Likewise.
5818 * cse.c (equiv_constant, cse_insn): Likewise.
5819 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5820 (cselib_subst_to_values): Likewise.
5821 * dce.c (word_dce_process_block): Likewise.
5822 * df-problems.c (df_word_lr_mark_ref): Likewise.
5823 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5824 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5825 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5826 (rtl_for_decl_location): Likewise.
5827 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5828 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5829 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5830 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5831 (expand_expr_real_1): Likewise.
5832 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5833 (pad_below): Likewise.
5834 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5835 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5836 * ira.c (get_subreg_tracking_sizes): Likewise.
5837 * ira-build.c (ira_create_allocno_objects): Likewise.
5838 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5839 (ira_sort_regnos_for_alter_reg): Likewise.
5840 * ira-costs.c (record_operand_costs): Likewise.
5841 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5842 (resolve_simple_move): Likewise.
5843 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5844 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5845 (lra_constraints): Likewise.
5846 (CONST_POOL_OK_P): Reject variable-sized modes.
5847 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5848 (add_pseudo_to_slot, lra_spill): Likewise.
5849 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5850 * optabs-query.c (get_best_extraction_insn): Likewise.
5851 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5852 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5853 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5854 * recog.c (offsettable_address_addr_space_p): Likewise.
5855 * regcprop.c (maybe_mode_change): Likewise.
5856 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5857 * regrename.c (build_def_use): Likewise.
5858 * regstat.c (dump_reg_info): Likewise.
5859 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5860 (find_reloads, find_reloads_subreg_address): Likewise.
5861 * reload1.c (eliminate_regs_1): Likewise.
5862 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5863 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5864 (simplify_binary_operation_1, simplify_subreg): Likewise.
5865 * targhooks.c (default_function_arg_padding): Likewise.
5866 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5867 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5868 (verify_gimple_assign_ternary): Likewise.
5869 * tree-inline.c (estimate_move_cost): Likewise.
5870 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5871 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5872 (get_address_cost_ainc): Likewise.
5873 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5874 (vect_supportable_dr_alignment): Likewise.
5875 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5876 (vectorizable_reduction): Likewise.
5877 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5878 (vectorizable_operation, vectorizable_load): Likewise.
5879 * tree.c (build_same_sized_truth_vector_type): Likewise.
5880 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5881 * var-tracking.c (emit_note_insn_var_location): Likewise.
5882 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5883 (ADDR_VEC_ALIGN): Likewise.
5885 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5886 Alan Hayward <alan.hayward@arm.com>
5887 David Sherwood <david.sherwood@arm.com>
5889 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5891 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5892 or if measurement_type is polynomial.
5893 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5894 * combine.c (make_extraction): Likewise.
5895 * dse.c (find_shift_sequence): Likewise.
5896 * dwarf2out.c (mem_loc_descriptor): Likewise.
5897 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5898 (extract_bit_field, extract_low_bits): Likewise.
5899 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5900 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5901 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5902 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5903 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5904 * reload.c (find_reloads): Likewise.
5905 * reload1.c (alter_reg): Likewise.
5906 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5907 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5908 * tree-if-conv.c (predicate_mem_writes): Likewise.
5909 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5910 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5911 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5912 * valtrack.c (dead_debug_insert_temp): Likewise.
5913 * varasm.c (mergeable_constant_section): Likewise.
5914 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5916 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5917 Alan Hayward <alan.hayward@arm.com>
5918 David Sherwood <david.sherwood@arm.com>
5920 * expr.c (expand_assignment): Cope with polynomial mode sizes
5921 when assigning to a CONCAT.
5923 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5924 Alan Hayward <alan.hayward@arm.com>
5925 David Sherwood <david.sherwood@arm.com>
5927 * machmode.h (mode_precision): Change from unsigned short to
5929 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5931 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5932 or if measurement_type is not polynomial.
5933 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5934 in which the mode is already known to be a scalar_int_mode.
5935 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5936 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5938 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5939 for GET_MODE_PRECISION.
5940 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5941 for GET_MODE_PRECISION.
5942 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5944 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5945 (expand_field_assignment, make_extraction): Likewise.
5946 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5947 (get_last_value): Likewise.
5948 * convert.c (convert_to_integer_1): Likewise.
5949 * cse.c (cse_insn): Likewise.
5950 * expr.c (expand_expr_real_1): Likewise.
5951 * lra-constraints.c (simplify_operand_subreg): Likewise.
5952 * optabs-query.c (can_atomic_load_p): Likewise.
5953 * optabs.c (expand_atomic_load): Likewise.
5954 (expand_atomic_store): Likewise.
5955 * ree.c (combine_reaching_defs): Likewise.
5956 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5957 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5958 * tree.h (type_has_mode_precision_p): Likewise.
5959 * ubsan.c (instrument_si_overflow): Likewise.
5961 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5962 Alan Hayward <alan.hayward@arm.com>
5963 David Sherwood <david.sherwood@arm.com>
5965 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5966 polynomial numbers of units.
5967 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5968 (valid_vector_subparts_p): New function.
5969 (build_vector_type): Remove temporary shim and take the number
5970 of units as a poly_uint64 rather than an int.
5971 (build_opaque_vector_type): Take the number of units as a
5972 poly_uint64 rather than an int.
5973 * tree.c (build_vector_from_ctor): Handle polynomial
5974 TYPE_VECTOR_SUBPARTS.
5975 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5976 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5977 (build_vector_from_val): If the number of units is variable,
5978 use build_vec_duplicate_cst for constant operands and
5979 VEC_DUPLICATE_EXPR otherwise.
5980 (make_vector_type): Remove temporary is_constant ().
5981 (build_vector_type, build_opaque_vector_type): Take the number of
5982 units as a poly_uint64 rather than an int.
5983 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5985 * cfgexpand.c (expand_debug_expr): Likewise.
5986 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5987 (store_constructor, expand_expr_real_1): Likewise.
5988 (const_scalar_mask_from_tree): Likewise.
5989 * fold-const-call.c (fold_const_reduction): Likewise.
5990 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5991 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5992 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5993 (fold_relational_const): Likewise.
5994 (native_interpret_vector): Likewise. Change the size from an
5995 int to an unsigned int.
5996 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5997 TYPE_VECTOR_SUBPARTS.
5998 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5999 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
6000 duplicating a non-constant operand into a variable-length vector.
6001 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
6002 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
6003 * ipa-icf.c (sem_variable::equals): Likewise.
6004 * match.pd: Likewise.
6005 * omp-simd-clone.c (simd_clone_subparts): Likewise.
6006 * print-tree.c (print_node): Likewise.
6007 * stor-layout.c (layout_type): Likewise.
6008 * targhooks.c (default_builtin_vectorization_cost): Likewise.
6009 * tree-cfg.c (verify_gimple_comparison): Likewise.
6010 (verify_gimple_assign_binary): Likewise.
6011 (verify_gimple_assign_ternary): Likewise.
6012 (verify_gimple_assign_single): Likewise.
6013 * tree-pretty-print.c (dump_generic_node): Likewise.
6014 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6015 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
6016 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
6017 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
6018 (vect_shift_permute_load_chain): Likewise.
6019 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
6020 (expand_vector_condition, optimize_vector_constructor): Likewise.
6021 (lower_vec_perm, get_compute_type): Likewise.
6022 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6023 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
6024 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
6025 (vect_recog_mask_conversion_pattern): Likewise.
6026 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6027 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6028 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6029 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6030 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6031 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6032 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6033 (supportable_widening_operation): Likewise.
6034 (supportable_narrowing_operation): Likewise.
6035 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6037 * varasm.c (output_constant): Likewise.
6039 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6040 Alan Hayward <alan.hayward@arm.com>
6041 David Sherwood <david.sherwood@arm.com>
6043 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6044 so that both the length == 3 and length != 3 cases set up their
6045 own permute vectors. Add comments explaining why we know the
6046 number of elements is constant.
6047 (vect_permute_load_chain): Likewise.
6049 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6050 Alan Hayward <alan.hayward@arm.com>
6051 David Sherwood <david.sherwood@arm.com>
6053 * machmode.h (mode_nunits): Change from unsigned char to
6055 (ONLY_FIXED_SIZE_MODES): New macro.
6056 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6057 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6058 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6060 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6061 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6062 or if measurement_type is not polynomial.
6063 * genmodes.c (ZERO_COEFFS): New macro.
6064 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6066 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6067 Use ZERO_COEFFS when emitting initializers.
6068 * data-streamer.h (bp_pack_poly_value): New function.
6069 (bp_unpack_poly_value): Likewise.
6070 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6071 for GET_MODE_NUNITS.
6072 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6073 for GET_MODE_NUNITS.
6074 * tree.c (make_vector_type): Remove temporary shim and make
6075 the real function take the number of units as a poly_uint64
6077 (build_vector_type_for_mode): Handle polynomial nunits.
6078 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6079 * emit-rtl.c (const_vec_series_p_1): Likewise.
6080 (gen_rtx_CONST_VECTOR): Likewise.
6081 * fold-const.c (test_vec_duplicate_folding): Likewise.
6082 * genrecog.c (validate_pattern): Likewise.
6083 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6084 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6085 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6086 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6087 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6088 * rtlanal.c (subreg_get_info): Likewise.
6089 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6090 (vect_grouped_load_supported): Likewise.
6091 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6092 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6093 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6094 (simplify_const_unary_operation, simplify_binary_operation_1)
6095 (simplify_const_binary_operation, simplify_ternary_operation)
6096 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6097 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6098 instead of CONST_VECTOR_NUNITS.
6099 * varasm.c (output_constant_pool_2): Likewise.
6100 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6101 explicit-encoded elements in the XVEC for variable-length vectors.
6103 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6105 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6107 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6108 Alan Hayward <alan.hayward@arm.com>
6109 David Sherwood <david.sherwood@arm.com>
6111 * coretypes.h (fixed_size_mode): Declare.
6112 (fixed_size_mode_pod): New typedef.
6113 * builtins.h (target_builtins::x_apply_args_mode)
6114 (target_builtins::x_apply_result_mode): Change type to
6115 fixed_size_mode_pod.
6116 * builtins.c (apply_args_size, apply_result_size, result_vector)
6117 (expand_builtin_apply_args_1, expand_builtin_apply)
6118 (expand_builtin_return): Update accordingly.
6120 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6122 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6123 * cselib.c (cselib_hash_rtx): Likewise.
6124 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6125 CONST_VECTOR encoding.
6127 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6128 Jeff Law <law@redhat.com>
6131 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6132 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6133 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6134 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6137 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6138 explicitly probe *sp in a noreturn function if there were any callee
6139 register saves or frame pointer is needed.
6141 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6144 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6145 BLKmode for ternary, binary or unary expressions.
6148 * var-tracking.c (delete_vta_debug_insn): New inline function.
6149 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6150 insns from get_insns () to NULL instead of each bb separately.
6151 Use delete_vta_debug_insn. No longer static.
6152 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6153 delete_vta_debug_insns callers.
6154 * rtl.h (delete_vta_debug_insns): Declare.
6155 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6156 instead of variable_tracking_main.
6158 2018-01-03 Martin Sebor <msebor@redhat.com>
6160 PR tree-optimization/83603
6161 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6162 arguments past the endof the argument list in functions declared
6163 without a prototype.
6164 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6165 Avoid checking when arguments are null.
6167 2018-01-03 Martin Sebor <msebor@redhat.com>
6170 * doc/extend.texi (attribute const): Fix a typo.
6171 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6172 issuing -Wsuggest-attribute for void functions.
6174 2018-01-03 Martin Sebor <msebor@redhat.com>
6176 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6177 offset_int::from instead of wide_int::to_shwi.
6178 (maybe_diag_overlap): Remove assertion.
6179 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6180 * gimple-ssa-sprintf.c (format_directive): Same.
6181 (parse_directive): Same.
6182 (sprintf_dom_walker::compute_format_length): Same.
6183 (try_substitute_return_value): Same.
6185 2018-01-03 Jeff Law <law@redhat.com>
6188 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6189 non-constant residual for zero at runtime and avoid probing in
6190 that case. Reorganize code for trailing problem to mirror handling
6193 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6195 PR tree-optimization/83501
6196 * tree-ssa-strlen.c (get_string_cst): New.
6197 (handle_char_store): Call get_string_cst.
6199 2018-01-03 Martin Liska <mliska@suse.cz>
6201 PR tree-optimization/83593
6202 * tree-ssa-strlen.c: Include tree-cfg.h.
6203 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6204 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6205 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6207 (strlen_dom_walker::before_dom_children): Call
6208 gimple_purge_dead_eh_edges. Dump tranformation with details
6210 (strlen_dom_walker::before_dom_children): Update call by adding
6211 new argument cleanup_eh.
6212 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6214 2018-01-03 Martin Liska <mliska@suse.cz>
6217 * cif-code.def (VARIADIC_THUNK): New enum value.
6218 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6221 2018-01-03 Jan Beulich <jbeulich@suse.com>
6223 * sse.md (mov<mode>_internal): Tighten condition for when to use
6224 vmovdqu<ssescalarsize> for TI and OI modes.
6226 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6228 Update copyright years.
6230 2018-01-03 Martin Liska <mliska@suse.cz>
6233 * ipa-visibility.c (function_and_variable_visibility): Skip
6234 functions with noipa attribure.
6236 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6238 * gcc.c (process_command): Update copyright notice dates.
6239 * gcov-dump.c (print_version): Ditto.
6240 * gcov.c (print_version): Ditto.
6241 * gcov-tool.c (print_version): Ditto.
6242 * gengtype.c (create_file): Ditto.
6243 * doc/cpp.texi: Bump @copying's copyright year.
6244 * doc/cppinternals.texi: Ditto.
6245 * doc/gcc.texi: Ditto.
6246 * doc/gccint.texi: Ditto.
6247 * doc/gcov.texi: Ditto.
6248 * doc/install.texi: Ditto.
6249 * doc/invoke.texi: Ditto.
6251 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6253 * vector-builder.h (vector_builder::m_full_nelts): Change from
6254 unsigned int to poly_uint64.
6255 (vector_builder::full_nelts): Update prototype accordingly.
6256 (vector_builder::new_vector): Likewise.
6257 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6258 (vector_builder::operator ==): Likewise.
6259 (vector_builder::finalize): Likewise.
6260 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6261 Take the number of elements as a poly_uint64 rather than an
6263 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6264 from unsigned int to poly_uint64.
6265 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6266 (vec_perm_indices::new_vector): Likewise.
6267 (vec_perm_indices::length): Likewise.
6268 (vec_perm_indices::nelts_per_input): Likewise.
6269 (vec_perm_indices::input_nelts): Likewise.
6270 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6271 number of elements per input as a poly_uint64 rather than an
6272 unsigned int. Use the original encoding for variable-length
6273 vectors, rather than clamping each individual element.
6274 For the second and subsequent elements in each pattern,
6275 clamp the step and base before clamping their sum.
6276 (vec_perm_indices::series_p): Handle polynomial element counts.
6277 (vec_perm_indices::all_in_range_p): Likewise.
6278 (vec_perm_indices_to_tree): Likewise.
6279 (vec_perm_indices_to_rtx): Likewise.
6280 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6281 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6282 (tree_vector_builder::new_binary_operation): Handle polynomial
6283 element counts. Return false if we need to know the number
6284 of elements at compile time.
6285 * fold-const.c (fold_vec_perm): Punt if the number of elements
6286 isn't known at compile time.
6288 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6290 * vec-perm-indices.h (vec_perm_builder): Change element type
6291 from HOST_WIDE_INT to poly_int64.
6292 (vec_perm_indices::element_type): Update accordingly.
6293 (vec_perm_indices::clamp): Handle polynomial element_types.
6294 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6295 (vec_perm_indices::all_in_range_p): Likewise.
6296 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6298 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6299 polynomial vec_perm_indices element types.
6300 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6301 * fold-const.c (fold_vec_perm): Likewise.
6302 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6303 * tree-vect-generic.c (lower_vec_perm): Likewise.
6304 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6305 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6306 element type to HOST_WIDE_INT.
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 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6313 rather than an int. Use plus_constant.
6314 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6315 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6317 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6318 Alan Hayward <alan.hayward@arm.com>
6319 David Sherwood <david.sherwood@arm.com>
6321 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6322 a HOST_WIDE_INT to a poly_int64.
6324 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6325 Alan Hayward <alan.hayward@arm.com>
6326 David Sherwood <david.sherwood@arm.com>
6328 * calls.c (load_register_parameters): Cope with polynomial
6329 mode sizes. Require a constant size for BLKmode parameters
6330 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6331 forces a parameter to be padded at the lsb end in order to
6332 fill a complete number of words, require the parameter size
6333 to be ordered wrt UNITS_PER_WORD.
6335 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6336 Alan Hayward <alan.hayward@arm.com>
6337 David Sherwood <david.sherwood@arm.com>
6339 * reload1.c (spill_stack_slot_width): Change element type
6340 from unsigned int to poly_uint64_pod.
6341 (alter_reg): Treat mode sizes as polynomial.
6343 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6344 Alan Hayward <alan.hayward@arm.com>
6345 David Sherwood <david.sherwood@arm.com>
6347 * reload.c (complex_word_subreg_p): New function.
6348 (reload_inner_reg_of_subreg, push_reload): Use it.
6350 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6351 Alan Hayward <alan.hayward@arm.com>
6352 David Sherwood <david.sherwood@arm.com>
6354 * lra-constraints.c (process_alt_operands): Reject matched
6355 operands whose sizes aren't ordered.
6356 (match_reload): Refer to this check here.
6358 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6359 Alan Hayward <alan.hayward@arm.com>
6360 David Sherwood <david.sherwood@arm.com>
6362 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6363 that the mode size is in the set {1, 2, 4, 8, 16}.
6365 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6366 Alan Hayward <alan.hayward@arm.com>
6367 David Sherwood <david.sherwood@arm.com>
6369 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6370 Use plus_constant instead of gen_rtx_PLUS.
6372 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6373 Alan Hayward <alan.hayward@arm.com>
6374 David Sherwood <david.sherwood@arm.com>
6376 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6377 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6378 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6379 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6380 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6381 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6382 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6383 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6384 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6385 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6387 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6388 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6389 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6390 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6391 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6392 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6393 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6394 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6395 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6396 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6398 * expr.c (emit_move_resolve_push): Treat the input and result
6399 of PUSH_ROUNDING as a poly_int64.
6400 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6401 (emit_push_insn): Likewise.
6402 * lra-eliminations.c (mark_not_eliminable): Likewise.
6403 * recog.c (push_operand): Likewise.
6404 * reload1.c (elimination_effects): Likewise.
6405 * rtlanal.c (nonzero_bits1): Likewise.
6406 * calls.c (store_one_arg): Likewise. Require the padding to be
6407 known at compile time.
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 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6414 Use plus_constant instead of gen_rtx_PLUS.
6416 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6417 Alan Hayward <alan.hayward@arm.com>
6418 David Sherwood <david.sherwood@arm.com>
6420 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6423 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6424 Alan Hayward <alan.hayward@arm.com>
6425 David Sherwood <david.sherwood@arm.com>
6427 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6428 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6429 via stack temporaries. Treat the mode size as polynomial too.
6431 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6432 Alan Hayward <alan.hayward@arm.com>
6433 David Sherwood <david.sherwood@arm.com>
6435 * expr.c (expand_expr_real_2): When handling conversions involving
6436 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6437 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6438 as a poly_uint64 too.
6440 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6441 Alan Hayward <alan.hayward@arm.com>
6442 David Sherwood <david.sherwood@arm.com>
6444 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6446 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6447 Alan Hayward <alan.hayward@arm.com>
6448 David Sherwood <david.sherwood@arm.com>
6450 * combine.c (can_change_dest_mode): Handle polynomial
6451 REGMODE_NATURAL_SIZE.
6452 * expmed.c (store_bit_field_1): Likewise.
6453 * expr.c (store_constructor): Likewise.
6454 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6455 and polynomial REGMODE_NATURAL_SIZE.
6456 (gen_lowpart_common): Likewise.
6457 * reginfo.c (record_subregs_of_mode): Likewise.
6458 * rtlanal.c (read_modify_subreg_p): Likewise.
6460 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6461 Alan Hayward <alan.hayward@arm.com>
6462 David Sherwood <david.sherwood@arm.com>
6464 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6465 numbers of elements.
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 * match.pd: Cope with polynomial numbers of vector elements.
6473 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6474 Alan Hayward <alan.hayward@arm.com>
6475 David Sherwood <david.sherwood@arm.com>
6477 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6478 in a POINTER_PLUS_EXPR.
6480 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6481 Alan Hayward <alan.hayward@arm.com>
6482 David Sherwood <david.sherwood@arm.com>
6484 * omp-simd-clone.c (simd_clone_subparts): New function.
6485 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6486 (ipa_simd_modify_function_body): Likewise.
6488 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6489 Alan Hayward <alan.hayward@arm.com>
6490 David Sherwood <david.sherwood@arm.com>
6492 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6493 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6494 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6495 (expand_vector_condition, vector_element): Likewise.
6496 (subparts_gt): New function.
6497 (get_compute_type): Use subparts_gt.
6498 (count_type_subparts): Delete.
6499 (expand_vector_operations_1): Use subparts_gt instead of
6500 count_type_subparts.
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-data-refs.c (vect_no_alias_p): Replace with...
6507 (vect_compile_time_alias): ...this new function. Do the calculation
6508 on poly_ints rather than trees.
6509 (vect_prune_runtime_alias_test_list): Update call accordingly.
6511 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6512 Alan Hayward <alan.hayward@arm.com>
6513 David Sherwood <david.sherwood@arm.com>
6515 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6517 (vect_schedule_slp_instance): Likewise.
6519 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6520 Alan Hayward <alan.hayward@arm.com>
6521 David Sherwood <david.sherwood@arm.com>
6523 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6524 constant and extern definitions for variable-length vectors.
6525 (vect_get_constant_vectors): Note that the number of units
6526 is known to be constant.
6528 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6529 Alan Hayward <alan.hayward@arm.com>
6530 David Sherwood <david.sherwood@arm.com>
6532 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6533 of units as polynomial. Choose between WIDE and NARROW based
6536 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6537 Alan Hayward <alan.hayward@arm.com>
6538 David Sherwood <david.sherwood@arm.com>
6540 * tree-vect-stmts.c (simd_clone_subparts): New function.
6541 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6543 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6544 Alan Hayward <alan.hayward@arm.com>
6545 David Sherwood <david.sherwood@arm.com>
6547 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6548 vectors as polynomial. Use build_index_vector for
6551 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6552 Alan Hayward <alan.hayward@arm.com>
6553 David Sherwood <david.sherwood@arm.com>
6555 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6556 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6557 for variable-length vectors.
6558 (vectorizable_mask_load_store): Treat the number of units as
6559 polynomial, asserting that it is constant if the condition has
6560 already been enforced.
6561 (vectorizable_store, vectorizable_load): Likewise.
6563 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6564 Alan Hayward <alan.hayward@arm.com>
6565 David Sherwood <david.sherwood@arm.com>
6567 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6568 of units as polynomial. Punt if we can't tell at compile time
6569 which vector contains the final result.
6571 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6572 Alan Hayward <alan.hayward@arm.com>
6573 David Sherwood <david.sherwood@arm.com>
6575 * tree-vect-loop.c (vectorizable_induction): Treat the number
6576 of units as polynomial. Punt on SLP inductions. Use an integer
6577 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6578 cast of such a series for variable-length floating-point
6581 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6582 Alan Hayward <alan.hayward@arm.com>
6583 David Sherwood <david.sherwood@arm.com>
6585 * tree.h (build_index_vector): Declare.
6586 * tree.c (build_index_vector): New function.
6587 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6588 of units as polynomial, forcibly converting it to a constant if
6589 vectorizable_reduction has already enforced the condition.
6590 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6591 to create a {1,2,3,...} vector.
6592 (vectorizable_reduction): Treat the number of units as polynomial.
6593 Choose vectype_in based on the largest scalar element size rather
6594 than the smallest number of units. Enforce the restrictions
6597 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6598 Alan Hayward <alan.hayward@arm.com>
6599 David Sherwood <david.sherwood@arm.com>
6601 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6602 number of units as polynomial.
6604 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6605 Alan Hayward <alan.hayward@arm.com>
6606 David Sherwood <david.sherwood@arm.com>
6608 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6609 * target.def (autovectorize_vector_sizes): Return the vector sizes
6610 by pointer, using vector_sizes rather than a bitmask.
6611 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6612 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6613 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6615 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6616 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6617 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6618 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6619 * omp-general.c (omp_max_vf): Likewise.
6620 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6621 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6622 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6623 * tree-vect-slp.c (vect_slp_bb): Likewise.
6624 * doc/tm.texi: Regenerate.
6625 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6627 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6628 the vector size as a poly_uint64 rather than an unsigned int.
6629 (current_vector_size): Change from an unsigned int to a poly_uint64.
6630 (get_vectype_for_scalar_type): Update accordingly.
6631 * tree.h (build_truth_vector_type): Take the size and number of
6632 units as a poly_uint64 rather than an unsigned int.
6633 (build_vector_type): Add a temporary overload that takes
6634 the number of units as a poly_uint64 rather than an unsigned int.
6635 * tree.c (make_vector_type): Likewise.
6636 (build_truth_vector_type): Take the number of units as a poly_uint64
6637 rather than an unsigned int.
6639 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6640 Alan Hayward <alan.hayward@arm.com>
6641 David Sherwood <david.sherwood@arm.com>
6643 * target.def (get_mask_mode): Take the number of units and length
6644 as poly_uint64s rather than unsigned ints.
6645 * targhooks.h (default_get_mask_mode): Update accordingly.
6646 * targhooks.c (default_get_mask_mode): Likewise.
6647 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6648 * doc/tm.texi: Regenerate.
6650 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6651 Alan Hayward <alan.hayward@arm.com>
6652 David Sherwood <david.sherwood@arm.com>
6654 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6655 * omp-general.c (omp_max_vf): Likewise.
6656 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6657 (expand_omp_simd): Handle polynomial safelen.
6658 * omp-low.c (omplow_simd_context): Add a default constructor.
6659 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6660 (lower_rec_simd_input_clauses): Update accordingly.
6661 (lower_rec_input_clauses): Likewise.
6663 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6664 Alan Hayward <alan.hayward@arm.com>
6665 David Sherwood <david.sherwood@arm.com>
6667 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6668 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6669 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6670 (vect_analyze_slp_cost): Likewise.
6671 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6672 (vect_model_load_cost): Likewise.
6674 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6675 Alan Hayward <alan.hayward@arm.com>
6676 David Sherwood <david.sherwood@arm.com>
6678 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6679 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6680 from an unsigned int * to a poly_uint64_pod *.
6681 (calculate_unrolling_factor): New function.
6682 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6684 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6685 Alan Hayward <alan.hayward@arm.com>
6686 David Sherwood <david.sherwood@arm.com>
6688 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6689 from an unsigned int to a poly_uint64.
6690 (_loop_vec_info::slp_unrolling_factor): Likewise.
6691 (_loop_vec_info::vectorization_factor): Change from an int
6693 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6694 (vect_get_num_vectors): New function.
6695 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6696 (vect_get_num_copies): Use vect_get_num_vectors.
6697 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6698 to an unsigned int *.
6699 (vect_analyze_data_refs): Change min_vf from an int * to a
6701 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6702 than an unsigned HOST_WIDE_INT.
6703 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6704 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6705 to an unsigned int *.
6706 (vect_analyze_data_ref_dependences): Likewise.
6707 (vect_compute_data_ref_alignment): Handle polynomial vf.
6708 (vect_enhance_data_refs_alignment): Likewise.
6709 (vect_prune_runtime_alias_test_list): Likewise.
6710 (vect_shift_permute_load_chain): Likewise.
6711 (vect_supportable_dr_alignment): Likewise.
6712 (dependence_distance_ge_vf): Take the vectorization factor as a
6713 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6714 (vect_analyze_data_refs): Change min_vf from an int * to a
6716 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6717 vfm1 as a poly_uint64 rather than an int. Make the same change
6718 for the returned bound_scalar.
6719 (vect_gen_vector_loop_niters): Handle polynomial vf.
6720 (vect_do_peeling): Likewise. Update call to
6721 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6722 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6724 * tree-vect-loop.c (vect_determine_vectorization_factor)
6725 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6726 (vect_get_known_peeling_cost): Likewise.
6727 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6728 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6729 (vect_transform_loop): Likewise. Use the lowest possible VF when
6730 updating the upper bounds of the loop.
6731 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6733 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6734 polynomial unroll factors.
6735 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6736 (vect_make_slp_decision): Likewise.
6737 (vect_supported_load_permutation_p): Likewise, and polynomial
6739 (vect_analyze_slp_cost): Handle polynomial vf.
6740 (vect_slp_analyze_node_operations): Likewise.
6741 (vect_slp_analyze_bb_1): Likewise.
6742 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6743 than an unsigned HOST_WIDE_INT.
6744 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6745 (vectorizable_load): Handle polynomial vf.
6746 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6748 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6750 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6751 Alan Hayward <alan.hayward@arm.com>
6752 David Sherwood <david.sherwood@arm.com>
6754 * match.pd: Handle bit operations involving three constants
6755 and try to fold one pair.
6757 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6759 * tree-vect-loop-manip.c: Include gimple-fold.h.
6760 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6761 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6762 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6763 Add a path that uses a step of VF instead of 1, but disable it
6765 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6766 and niters_no_overflow parameters. Update calls to
6767 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6768 Create a new SSA name if the latter choses to use a ste other
6769 than zero, and return it via niters_vector_mult_vf_var.
6770 * tree-vect-loop.c (vect_transform_loop): Update calls to
6771 vect_do_peeling, vect_gen_vector_loop_niters and
6772 slpeel_make_loop_iterate_ntimes.
6773 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6774 (vect_gen_vector_loop_niters): Update declarations after above changes.
6776 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6778 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6779 128-bit round to integer instructions.
6780 (ceil<mode>2): Likewise.
6781 (btrunc<mode>2): Likewise.
6782 (round<mode>2): Likewise.
6784 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6786 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6787 unaligned VSX load/store on P8/P9.
6788 (expand_block_clear): Allow the use of unaligned VSX
6789 load/store on P8/P9.
6791 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6793 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6795 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6796 swap associated with both a load and a store.
6798 2018-01-02 Andrew Waterman <andrew@sifive.com>
6800 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6801 * config/riscv/riscv.md (clear_cache): Use it.
6803 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6805 * web.c: Remove out-of-date comment.
6807 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6809 * expr.c (fixup_args_size_notes): Check that any existing
6810 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6811 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6812 (emit_single_push_insn): ...here.
6814 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6816 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6817 (const_vector_encoded_nelts): New function.
6818 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6819 (const_vector_int_elt, const_vector_elt): Declare.
6820 * emit-rtl.c (const_vector_int_elt_1): New function.
6821 (const_vector_elt): Likewise.
6822 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6823 of CONST_VECTOR_ELT.
6825 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6827 * expr.c: Include rtx-vector-builder.h.
6828 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6829 directly on the tree encoding.
6830 (const_vector_from_tree): Likewise.
6831 * optabs.c: Include rtx-vector-builder.h.
6832 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6833 sequence of "u" values.
6834 * vec-perm-indices.c: Include rtx-vector-builder.h.
6835 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6836 directly on the vec_perm_indices encoding.
6838 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6840 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6841 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6842 * rtx-vector-builder.h: New file.
6843 * rtx-vector-builder.c: Likewise.
6844 * rtl.h (rtx_def::u2): Add a const_vector field.
6845 (CONST_VECTOR_NPATTERNS): New macro.
6846 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6847 (CONST_VECTOR_DUPLICATE_P): Likewise.
6848 (CONST_VECTOR_STEPPED_P): Likewise.
6849 (CONST_VECTOR_ENCODED_ELT): Likewise.
6850 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6851 (unwrap_const_vec_duplicate): Likewise.
6852 (const_vec_series_p): Check for a non-duplicated vector encoding.
6853 Say that the function only returns true for integer vectors.
6854 * emit-rtl.c: Include rtx-vector-builder.h.
6855 (gen_const_vec_duplicate_1): Delete.
6856 (gen_const_vector): Call gen_const_vec_duplicate instead of
6857 gen_const_vec_duplicate_1.
6858 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6859 (gen_const_vec_duplicate): Use rtx_vector_builder.
6860 (gen_const_vec_series): Likewise.
6861 (gen_rtx_CONST_VECTOR): Likewise.
6862 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6863 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6864 Build a new vector rather than modifying a CONST_VECTOR in-place.
6865 (handle_special_swappables): Update call accordingly.
6866 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6867 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6868 Build a new vector rather than modifying a CONST_VECTOR in-place.
6869 (handle_special_swappables): Update call accordingly.
6871 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6873 * simplify-rtx.c (simplify_const_binary_operation): Use
6874 CONST_VECTOR_ELT instead of XVECEXP.
6876 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6878 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6879 the selector elements to be different from the data elements
6880 if the selector is a VECTOR_CST.
6881 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6882 ssizetype for the selector.
6884 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6886 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6887 before testing each element individually.
6888 * tree-vect-generic.c (lower_vec_perm): Likewise.
6890 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6892 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6893 * selftest-run-tests.c (selftest::run_tests): Call it.
6894 * vector-builder.h (vector_builder::operator ==): New function.
6895 (vector_builder::operator !=): Likewise.
6896 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6897 (vec_perm_indices::all_from_input_p): New function.
6898 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6899 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6900 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6901 instead of reading the VECTOR_CST directly. Detect whether both
6902 vector inputs are the same before constructing the vec_perm_indices,
6903 and update the number of inputs argument accordingly. Use the
6904 utility functions added above. Only construct sel2 if we need to.
6906 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6908 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6909 the broadcast of the low byte.
6910 (expand_mult_highpart): Use an explicit encoding for the permutes.
6911 * optabs-query.c (can_mult_highpart_p): Likewise.
6912 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6913 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6914 (vectorizable_bswap): Likewise.
6915 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6916 explicit encoding for the power-of-2 permutes.
6917 (vect_permute_store_chain): Likewise.
6918 (vect_grouped_load_supported): Likewise.
6919 (vect_permute_load_chain): Likewise.
6921 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6923 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6924 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6925 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6926 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6927 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6928 (vect_gen_perm_mask_any): Likewise.
6930 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6932 * int-vector-builder.h: New file.
6933 * vec-perm-indices.h: Include int-vector-builder.h.
6934 (vec_perm_indices): Redefine as an int_vector_builder.
6935 (auto_vec_perm_indices): Delete.
6936 (vec_perm_builder): Redefine as a stand-alone class.
6937 (vec_perm_indices::vec_perm_indices): New function.
6938 (vec_perm_indices::clamp): Likewise.
6939 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6940 (vec_perm_indices::new_vector): New function.
6941 (vec_perm_indices::new_expanded_vector): Update for new
6942 vec_perm_indices class.
6943 (vec_perm_indices::rotate_inputs): New function.
6944 (vec_perm_indices::all_in_range_p): Operate directly on the
6945 encoded form, without computing elided elements.
6946 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6947 encoding. Update for new vec_perm_indices class.
6948 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6949 the given vec_perm_builder.
6950 (expand_vec_perm_var): Update vec_perm_builder constructor.
6951 (expand_mult_highpart): Use vec_perm_builder instead of
6952 auto_vec_perm_indices.
6953 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6954 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6955 or double series encoding as appropriate.
6956 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6957 vec_perm_indices instead of auto_vec_perm_indices.
6958 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6959 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6960 (vect_permute_store_chain): Likewise.
6961 (vect_grouped_load_supported): Likewise.
6962 (vect_permute_load_chain): Likewise.
6963 (vect_shift_permute_load_chain): Likewise.
6964 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6965 (vect_transform_slp_perm_load): Likewise.
6966 (vect_schedule_slp_instance): Likewise.
6967 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6968 (vectorizable_mask_load_store): Likewise.
6969 (vectorizable_bswap): Likewise.
6970 (vectorizable_store): Likewise.
6971 (vectorizable_load): Likewise.
6972 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6973 vec_perm_indices instead of auto_vec_perm_indices. Use
6974 tree_to_vec_perm_builder to read the vector from a tree.
6975 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6976 vec_perm_builder instead of a vec_perm_indices.
6977 (have_whole_vector_shift): Use vec_perm_builder and
6978 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6979 truncation to calc_vec_perm_mask_for_shift.
6980 (vect_create_epilog_for_reduction): Likewise.
6981 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6982 from auto_vec_perm_indices to vec_perm_indices.
6983 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6984 instead of changing individual elements.
6985 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6986 the vector in d.perm.
6987 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6988 from auto_vec_perm_indices to vec_perm_indices.
6989 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6990 instead of changing individual elements.
6991 (arm_vectorize_vec_perm_const): Use new_vector to install
6992 the vector in d.perm.
6993 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6994 Update vec_perm_builder constructor.
6995 (rs6000_expand_interleave): Likewise.
6996 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6997 (rs6000_expand_interleave): Likewise.
6999 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7001 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
7002 to qimode could truncate the indices.
7003 * optabs.c (expand_vec_perm_var): Likewise.
7005 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7007 * Makefile.in (OBJS): Add vec-perm-indices.o.
7008 * vec-perm-indices.h: New file.
7009 * vec-perm-indices.c: Likewise.
7010 * target.h (vec_perm_indices): Replace with a forward class
7012 (auto_vec_perm_indices): Move to vec-perm-indices.h.
7013 * optabs.h: Include vec-perm-indices.h.
7014 (expand_vec_perm): Delete.
7015 (selector_fits_mode_p, expand_vec_perm_var): Declare.
7016 (expand_vec_perm_const): Declare.
7017 * target.def (vec_perm_const_ok): Replace with...
7018 (vec_perm_const): ...this new hook.
7019 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
7020 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
7021 * doc/tm.texi: Regenerate.
7022 * optabs.def (vec_perm_const): Delete.
7023 * doc/md.texi (vec_perm_const): Likewise.
7024 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
7025 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7026 expand_vec_perm for constant permutation vectors. Assert that
7027 the mode of variable permutation vectors is the integer equivalent
7028 of the mode that is being permuted.
7029 * optabs-query.h (selector_fits_mode_p): Declare.
7030 * optabs-query.c: Include vec-perm-indices.h.
7031 (selector_fits_mode_p): New function.
7032 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7033 is defined, instead of checking whether the vec_perm_const_optab
7034 exists. Use targetm.vectorize.vec_perm_const instead of
7035 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7036 fit in the vector mode before using a variable permute.
7037 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7038 vec_perm_indices instead of an rtx.
7039 (expand_vec_perm): Replace with...
7040 (expand_vec_perm_const): ...this new function. Take the selector
7041 as a vec_perm_indices rather than an rtx. Also take the mode of
7042 the selector. Update call to shift_amt_for_vec_perm_mask.
7043 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7044 Use vec_perm_indices::new_expanded_vector to expand the original
7045 selector into bytes. Check whether the indices fit in the vector
7046 mode before using a variable permute.
7047 (expand_vec_perm_var): Make global.
7048 (expand_mult_highpart): Use expand_vec_perm_const.
7049 * fold-const.c: Includes vec-perm-indices.h.
7050 * tree-ssa-forwprop.c: Likewise.
7051 * tree-vect-data-refs.c: Likewise.
7052 * tree-vect-generic.c: Likewise.
7053 * tree-vect-loop.c: Likewise.
7054 * tree-vect-slp.c: Likewise.
7055 * tree-vect-stmts.c: Likewise.
7056 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7058 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7059 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7060 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7061 (aarch64_vectorize_vec_perm_const): ...this new function.
7062 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7063 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7064 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7065 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7066 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7067 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7068 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7070 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7071 check for NEON modes.
7072 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7073 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7074 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7075 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7077 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7078 the old VEC_PERM_CONST conditions.
7079 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7080 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7081 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7082 (ia64_vectorize_vec_perm_const_ok): Merge into...
7083 (ia64_vectorize_vec_perm_const): ...this new function.
7084 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7085 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7086 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7087 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7088 * config/mips/mips.c (mips_expand_vec_perm_const)
7089 (mips_vectorize_vec_perm_const_ok): Merge into...
7090 (mips_vectorize_vec_perm_const): ...this new function.
7091 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7092 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7093 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7094 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7095 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7096 (rs6000_expand_vec_perm_const): Delete.
7097 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7099 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7100 (altivec_expand_vec_perm_const_le): Take each operand individually.
7101 Operate on constant selectors rather than rtxes.
7102 (altivec_expand_vec_perm_const): Likewise. Update call to
7103 altivec_expand_vec_perm_const_le.
7104 (rs6000_expand_vec_perm_const): Delete.
7105 (rs6000_vectorize_vec_perm_const_ok): Delete.
7106 (rs6000_vectorize_vec_perm_const): New function.
7107 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7108 an element count and rtx array.
7109 (rs6000_expand_extract_even): Update call accordingly.
7110 (rs6000_expand_interleave): Likewise.
7111 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7112 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7113 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7114 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7115 (rs6000_expand_vec_perm_const): Delete.
7116 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7117 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7118 (altivec_expand_vec_perm_const_le): Take each operand individually.
7119 Operate on constant selectors rather than rtxes.
7120 (altivec_expand_vec_perm_const): Likewise. Update call to
7121 altivec_expand_vec_perm_const_le.
7122 (rs6000_expand_vec_perm_const): Delete.
7123 (rs6000_vectorize_vec_perm_const_ok): Delete.
7124 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7125 reference to the SPE evmerge intructions.
7126 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7127 an element count and rtx array.
7128 (rs6000_expand_extract_even): Update call accordingly.
7129 (rs6000_expand_interleave): Likewise.
7130 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7131 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7133 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7135 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7137 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7138 vector mode and that that mode matches the mode of the data
7140 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7141 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7142 directly using expand_vec_perm_1 when forcing selectors into
7144 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7146 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7148 * optabs-query.h (can_vec_perm_p): Delete.
7149 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7150 * optabs-query.c (can_vec_perm_p): Split into...
7151 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7152 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7153 particular selector is valid.
7154 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7155 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7156 (vect_grouped_load_supported): Likewise.
7157 (vect_shift_permute_load_chain): Likewise.
7158 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7159 (vect_transform_slp_perm_load): Likewise.
7160 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7161 (vectorizable_bswap): Likewise.
7162 (vect_gen_perm_mask_checked): Likewise.
7163 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7164 implementations of variable permutation vectors into account
7165 when deciding which selector to use.
7166 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7167 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7168 with a false third argument.
7169 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7170 to test whether the constant selector is valid and can_vec_perm_var_p
7171 to test whether a variable selector is valid.
7173 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7175 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7176 * optabs-query.c (can_vec_perm_p): Likewise.
7177 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7178 instead of vec_perm_indices.
7179 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7180 (vect_gen_perm_mask_checked): Likewise,
7181 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7182 (vect_gen_perm_mask_checked): Likewise,
7184 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7186 * optabs-query.h (qimode_for_vec_perm): Declare.
7187 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7188 (qimode_for_vec_perm): ...this new function.
7189 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7191 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7193 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7194 does not have a conditional at the top.
7196 2018-01-02 Richard Biener <rguenther@suse.de>
7198 * ipa-inline.c (big_speedup_p): Fix expression.
7200 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7203 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7206 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7210 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7211 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7212 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7213 cond_taken_branch_cost 3->4.
7215 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7217 PR tree-optimization/83581
7218 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7219 TODO_cleanup_cfg if any changes have been made.
7222 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7223 convert_modes if target mode has the right side, but different mode
7227 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7228 last argument when extracting from CONCAT. If either from_real or
7229 from_imag is NULL, use expansion through memory. If result is not
7230 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7231 the parts directly to inner mode, if even that fails, use expansion
7235 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7236 check for bswap in mode rather than HImode and use that in expand_unop
7239 Copyright (C) 2018 Free Software Foundation, Inc.
7241 Copying and distribution of this file, with or without modification,
7242 are permitted in any medium without royalty provided the copyright
7243 notice and this notice are preserved.