1 2018-02-16 Jeff Law <law@redhat.com>
3 * config/rx/rx.c (add_pop_cfi_notes): New function.;
6 2018-02-16 Jakub Jelinek <jakub@redhat.com>
9 * ipa-inline.c (inline_small_functions): Fix a typo.
11 2018-02-16 Nathan Sidwell <nathan@acm.org>
13 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
15 2018-02-16 Carl Love <cel@us.ibm.com>
17 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
18 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
19 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
20 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
21 expansion to P8V_BUILTIN_VEC_FLOAT2.
23 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
25 PR rtl-optimization/70023
26 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
27 src_regno into account.
29 2018-02-16 Carl Love <cel@us.ibm.com>
31 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
32 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
33 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
34 * config/rs6000/rs6000.c: Remove case statements for
35 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
36 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
37 and P9V_BUILTIN_VEC_VINSERT4B.
38 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
39 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
40 * config/rs6000/vsx.md:
41 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
44 2018-02-16 Carl Love <cel@us.ibm.com>
46 * config/rs6000/altivec.h: Add builtin names vec_extract4b
48 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
50 * config/rs6000/rs6000-c.c: Add the definitions for
51 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
52 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
53 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
54 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
55 definition for insert4b and define insn *insert3b_internal.
56 * doc/extend.texi: Add documentation for vec_extract4b.
58 2018-02-16 Nathan Sidwell <nathan@acm.org>
60 * doc/extend.texi (Backwards Compatibility): Mention friend
61 injection. Note for-scope is deprecated.
62 * doc/invoke.texi (-ffriend-injection): Deprecate.
64 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
66 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
67 that moved to I2, also allow destinations that are a paradoxical
68 subreg (instead of a normal reg).
70 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
73 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
76 2018-02-16 Richard Biener <rguenther@suse.de>
78 PR tree-optimization/84037
79 PR tree-optimization/84016
81 * config/i386/i386.c (ix86_builtin_vectorization_cost):
82 Adjust vec_construct for the fact we need additional higher latency
83 128bit inserts for AVX256 and AVX512 vector builds.
84 (ix86_add_stmt_cost): Scale vector construction cost for
87 2018-02-16 Richard Biener <rguenther@suse.de>
89 PR tree-optimization/84417
90 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
91 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
92 (non_rewritable_lvalue_p): Likewise, use poly-ints.
94 2018-02-16 Martin Liska <mliska@suse.cz>
97 * internal-fn.def (ASAN_CHECK): Set proper flags.
98 (ASAN_MARK): Likewise.
100 2018-02-16 Julia Koval <julia.koval@intel.com>
102 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
105 2018-02-16 Jakub Jelinek <jakub@redhat.com>
108 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
109 Use ++iter rather than iter++ for std::list iterators.
110 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
111 defer deleting them until all nodes in the forest are processed. Do
112 free even leaf nodes. Change to_process into auto_vec.
115 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
116 * vec.h (vec_default_construct): Use memset instead of placement new
117 if BROKEN_VALUE_INITIALIZATION is defined.
118 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
119 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
122 PR rtl-optimization/83723
123 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
124 * lra.c (lra_substitute_pseudo): Likewise. If true, use
125 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
127 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
129 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
131 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
133 PR rtl-optimization/81443
134 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
135 from inner REGs to paradoxical SUBREGs.
137 2018-02-16 Richard Biener <rguenther@suse.de>
139 PR tree-optimization/84399
140 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
141 For operands we can analyze at their definition make sure we can
142 analyze them at each use as well.
144 2018-02-16 Richard Biener <rguenther@suse.de>
146 PR tree-optimization/84190
147 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
148 volatile accesses if the decl isn't volatile.
150 2018-02-15 Jason Merrill <jason@redhat.com>
152 PR c++/84314 - ICE with templates and fastcall attribute.
153 * attribs.c (build_type_attribute_qual_variant): Don't clobber
154 TYPE_CANONICAL on an existing type.
156 2018-02-15 Jakub Jelinek <jakub@redhat.com>
158 PR tree-optimization/84383
159 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
160 dstoff nor call operand_equal_p if dstbase is NULL.
162 PR tree-optimization/84334
163 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
164 also a CONSTANT_CLASS_P, punt.
166 2018-02-14 Jim Wilson <jimw@sifive.com>
168 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
169 first SMALL_OPERAND check. New local min_second_step. Move assert
170 to where locals are set. Add TARGET_RVC support.
171 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
173 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
175 * doc/invoke.texi: Correct -Wformat-overflow code sample.
177 2018-02-14 Martin Sebor <msebor@redhat.com>
179 PR tree-optimization/83698
180 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
181 arrays constrain the offset range to their bounds.
182 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
183 (builtin_access::overlap): Avoid setting the size of overlap if it's
185 (maybe_diag_overlap): Also consider arrays when deciding what values
186 of offsets to include in diagnostics.
188 2018-02-14 Martin Sebor <msebor@redhat.com>
191 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
192 that correspond to the kind of a declaration.
194 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
197 * config/pa/pa.md: Load address of PIC label using the linkage table
198 if the label is nonlocal.
200 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
202 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
203 warning message if user requests -maltivec=be.
204 * doc/invoke.texi: Document deprecation of -maltivec=be.
206 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
209 * config/rs6000/rs6000-c.c: Update definitions for
210 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
211 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
213 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
216 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
217 add _get_ssp intrinsics. Remove argument from
218 __builtin_ia32_rdssp[d|q].
219 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
220 * config/i386/i386-builtin.def: Remove argument from
221 __builtin_ia32_rdssp[d|q].
222 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
223 ix86_expand_special_args_builtin for _rdssp[d|q].
224 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
225 Clear register before usage.
226 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
227 Add documentation for new _get_ssp and _inc_ssp intrinsics.
229 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
231 PR tree-optimization/84357
232 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
233 operand 1 of an ARRAY_REF too.
235 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
238 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
239 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
241 (set_of_reg): New struct.
242 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
243 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
244 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
246 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
247 Split into bitclr, bitset, bitinvert patterns if appropriate.
248 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
249 use rx_fuse_in_memory_bitop.
250 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
251 to named insn, correct maximum insn length.
253 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
256 * machmode.def: Define a complex mode for PARTIAL_INT.
257 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
259 * doc/rtl.texi: Document CSPImode.
260 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
262 (msp430_hard_regno_nregs_with_padding): Likewise.
264 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
267 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
269 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
271 PR rtl-optimization/84169
272 * combine.c (try_combine): New variable split_i2i3. Set it to true if
273 we generated a parallel as new i3 and we split that to new i2 and i3
274 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
275 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
276 those to i2, not i1. Partially rewrite this scan code.
278 2018-02-13 Jakub Jelinek <jakub@redhat.com>
281 * stor-layout.c (place_field): For variable length fields, adjust
282 offset_align afterwards not just based on the field's alignment,
283 but also on the size.
286 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
287 of exps and logs in the use_exp2 case.
289 2018-02-13 Jeff Law <law@redhat.com>
291 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
294 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
297 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
301 * common.opt (gas-loc-support, gas-locview-support): New.
302 (ginline-points, ginternal-reset-location-views): New.
303 * doc/invoke.texi: Document them. Use @itemx where intended.
304 (gvariable-location-views): Adjust.
305 * target.def (reset_location_view): New.
306 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
307 (TARGET_RESET_LOCATION_VIEW): New.
308 * doc/tm.texi: Rebuilt.
309 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
310 (dwarf2out_default_as_locview_support): New.
311 (output_asm_line_debug_info): Use option variables.
312 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
313 (output_loc_list): Likewise.
314 (add_high_low_attributes): Check option variables.
315 Don't output entry view attribute in strict mode.
316 (gen_inlined_subroutine_die): Check option variables.
317 (dwarf2out_inline_entry): Likewise.
318 (init_sections_and_labels): Likewise.
319 (dwarf2out_early_finish): Likewise.
320 (maybe_reset_location_view): New, from...
321 (dwarf2out_var_location): ... here. Call it.
322 * debug.h (dwarf2out_default_as_loc_support): Declare.
323 (dwarf2out_default_as_locview_support): Declare.
324 * hooks.c (hook_int_rtx_insn_0): New.
325 * hooks.h (hook_int_rtx_insn_0): Declare.
326 * toplev.c (process_options): Take -gas-loc-support and
327 -gas-locview-support from dwarf2out. Enable
328 -gvariable-location-views by default only with locview
329 assembler support. Enable -ginternal-reset-location-views by
330 default only if the target defines the corresponding hook.
331 Enable -ginline-points by default if location views are
332 enabled; force it disabled if statement frontiers are
334 * tree-inline.c (expand_call_inline): Check option variables.
335 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
337 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
339 PR tree-optimization/84321
340 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
341 handling. Also check whether the anti-range contains any values
342 that satisfy the mask; switch to a VR_RANGE if not.
344 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
347 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
349 2018-02-13 Martin Jambor <mjambor@suse.cz>
352 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
353 of call statements, also set location of a load to a temporary.
355 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
357 * config/rl78/rl78.c (add_vector_labels): New function.
358 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
359 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
360 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
361 which checks that no arguments are passed.
362 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
363 * doc/extend.texi: Documentation for the new attribute.
365 2018-02-13 Andreas Schwab <schwab@suse.de>
367 * config/riscv/linux.h (CPP_SPEC): Define.
369 2018-02-13 Jakub Jelinek <jakub@redhat.com>
372 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
373 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
374 OPTION_MASK_ISA_AES as first argument to def_builtin_const
375 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
376 instead of OPTION_MASK_ISA_PCLMUL as first argument to
377 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
378 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
379 temporarily for AES and PCLMUL builtins.
381 PR tree-optimization/84339
382 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
383 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
387 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
388 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
389 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
391 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
393 * omp-simd-clone.h: New file.
394 * omp-simd-clone.c: Include omp-simd-clone.h.
395 (expand_simd_clones): No longer static.
396 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
397 cgraph.h and omp-simd-clone.h.
398 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
399 (vect_recog_widen_shift_pattern): Formatting fix.
400 (vect_pattern_recog_1): Don't check optab for calls.
403 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
404 operands[2] into a REG before using gen_lowpart on it.
406 2018-02-12 Jeff Law <law@redhat.com>
409 * config/sh/sh.c (find_barrier): Consider a sibling call
412 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
413 successfully back substituting a reg.
415 2018-02-12 Richard Biener <rguenther@suse.de>
417 PR tree-optimization/84037
418 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
419 parameter, move visited init to caller.
420 (vect_slp_analyze_operations): Separate cost from validity
421 check, initialize visited once for all instances.
422 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
424 * tree-vect-stmts.c (vect_model_simple_cost): Make early
426 (vect_model_promotion_demotion_cost): Likewise.
427 (vectorizable_bswap): Guard cost modeling with !slp_node
428 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
430 (vectorizable_call): Likewise.
431 (vectorizable_conversion): Likewise.
432 (vectorizable_assignment): Likewise.
433 (vectorizable_shift): Likewise.
434 (vectorizable_operation): Likewise.
435 (vectorizable_store): Likewise.
436 (vectorizable_load): Likewise.
437 (vectorizable_condition): Likewise.
438 (vectorizable_comparison): Likewise.
440 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
443 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
444 (ASAN_MARK): Fix fnspec to account for return value, change pointer
445 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
447 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
450 * params.def (inline-min-speedup): Increase from 8 to 15.
451 (max-inline-insns-auto): Decrease from 40 to 30.
452 * ipa-split.c (consider_split): Add some buffer for function to
453 be considered inlining candidate.
454 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
457 2018-02-12 Richard Biener <rguenther@suse.de>
459 PR tree-optimization/84037
460 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
461 matched stmts if we cannot swap the non-matched ones.
463 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
465 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
466 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
467 _mm_maskz_scalef_round_ss): New intrinsics.
468 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
469 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
470 __builtin_ia32_scalefss_round): Remove.
471 (__builtin_ia32_scalefsd_mask_round,
472 __builtin_ia32_scalefss_mask_round): New intrinsics.
473 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
474 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
475 ((match_operand:VF_128 2 "<round_nimm_predicate>"
476 "<round_constraint>")): Changed to ...
477 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
478 "<round_scalar_constraint>")): ... this.
479 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
480 %0, %1, %2<round_op3>}"): Changed to ...
481 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
482 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
483 %2<round_scalar_mask_op3>}"): ... this.
484 * config/i386/subst.md (round_scalar_nimm_predicate): New.
486 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
488 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
489 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
490 (_mm_maskz_sqrt_round_ss): New intrinsics.
491 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
492 (__builtin_ia32_sqrtsd_mask_round)
493 (__builtin_ia32_sqrtss_mask_round): New builtins.
494 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
495 (__builtin_ia32_sqrtss_round): Remove.
496 (__builtin_ia32_sqrtsd_mask_round)
497 (__builtin_ia32_sqrtss_mask_round): New builtins.
498 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
499 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
500 ((match_operand:VF_128 1 "vector_operand"
501 "xBm,<round_constraint>")): Changed to ...
502 ((match_operand:VF_128 1 "vector_operand"
503 "xBm,<round_scalar_constraint>")): ... this.
504 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
505 %0, %2, %<iptr>1<round_op3>}): Changed to ...
506 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
507 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
508 %<iptr>1<round_scalar_mask_op3>}): ... this.
509 ((set_attr "prefix" "<round_prefix>")): Changed to ...
510 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
512 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
515 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
516 Cast vec_cmpeq result to correct type.
517 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
518 Cast vec_cmpgt result to correct type.
520 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
522 * final.c (final_scan_insn_1): Renamed from...
523 (final_scan_insn): ... this. New wrapper, to recover
524 seen from the outermost call in recursive ones.
525 * config/sparc/sparc.c (output_return): Drop seen from call.
526 (output_sibcall): Likewise.
527 * config/visium/visium.c (output_branch): Likewise.
529 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
531 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
534 2018-02-10 Alan Modra <amodra@gmail.com>
537 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
538 Specify LR as an input.
540 2018-02-10 Jakub Jelinek <jakub@redhat.com>
543 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
544 remove_member_access_dummy_vars): New functions.
545 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
546 lower_omp_1, execute_lower_omp): Use them.
548 PR rtl-optimization/84308
549 * shrink-wrap.c (spread_components): Release todo vector.
551 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
553 PR rtl-optimization/57193
554 * ira-color.c (struct allocno_color_data): Add member
555 conflict_allocno_hard_prefs.
556 (update_conflict_allocno_hard_prefs): New.
557 (bucket_allocno_compare_func): Add a preference based on
558 conflict_allocno_hard_prefs.
559 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
560 (color_allocnos): Remove a dead code. Initiate
561 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
563 2018-02-09 Jakub Jelinek <jakub@redhat.com>
566 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
567 constraint from =wa to wa. Avoid a subreg on the output operand,
568 instead use a pseudo and subreg it in a move.
569 (p9_xxbrd_<mode>): Changed to ...
570 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
571 (p9_xxbrd_v2df): New expander.
572 (p9_xxbrw_<mode>): Changed to ...
573 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
574 (p9_xxbrw_v4sf): New expander.
576 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
578 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
580 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
583 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
584 multiply in 32-bit mode.
585 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
586 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
589 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
591 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
592 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
593 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
594 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
596 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
599 * dwarf2out.c (is_trivial_indirect_ref): New function.
600 (dwarf2out_late_global_decl): Do not generate a location
601 attribute for variables that have a non-trivial DECL_VALUE_EXPR
602 and that are not defined in the current unit.
604 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
606 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
607 instead of a libcall for UNORDERED.
609 2018-02-09 Tamar Christina <tamar.christina@arm.com>
612 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
613 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
615 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
618 * config/s390/s390.c (s390_set_current_function): Invoke
619 s390_indirect_branch_settings also if fndecl didn't change.
621 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
623 * config/rs6000/rs6000.md (blockage): Set length to zero.
625 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
627 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
629 2018-02-09 Jakub Jelinek <jakub@redhat.com>
632 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
633 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
637 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
638 PARALLEL incoming that failed vt_get_decl_and_offset check.
641 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
642 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
644 (get_variable_section): For decls in named .bss* sections pass true as
645 second argument to bss_initializer_p.
647 2018-02-09 Marek Polacek <polacek@redhat.com>
648 Jakub Jelinek <jakub@redhat.com>
651 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
652 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
653 Sync some changes from cxx_fold_indirect_ref.
655 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
657 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
659 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
660 (BLOCK_INLINE_ENTRY_LABEL): New.
661 (dwarf2out_var_location): Disregard inline entry markers.
662 (inline_entry_data): New struct.
663 (inline_entry_data_hasher): New hashtable type.
664 (inline_entry_data_hasher::hash): New.
665 (inline_entry_data_hasher::equal): New.
666 (inline_entry_data_table): New variable.
667 (add_high_low_attributes): Add DW_AT_entry_pc and
668 DW_AT_GNU_entry_view attributes if a pending entry is found
669 in inline_entry_data_table. Add old entry_pc attribute only
670 if debug nonbinding markers are disabled.
671 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
673 (block_within_block_p, dwarf2out_inline_entry): New.
674 (dwarf2out_finish): Check that no entries remained in
675 inline_entry_data_table.
676 * final.c (reemit_insn_block_notes): Handle inline entry notes.
677 (final_scan_insn, notice_source_line): Likewise.
678 (rest_of_clean_state): Skip inline entry markers.
679 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
681 * gimple.c (gimple_build_debug_inline_entry): New.
682 * gimple.h (enum gimple_debug_subcode): Add
683 GIMPLE_DEBUG_INLINE_ENTRY.
684 (gimple_build_debug_inline_entry): Declare.
685 (gimple_debug_inline_entry_p): New.
686 (gimple_debug_nonbind_marker_p): Adjust.
687 * insn-notes.def (INLINE_ENTRY): New.
688 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
689 inline entry marker notes.
690 (print_insn): Likewise.
691 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
692 (INSN_DEBUG_MARKER_KIND): Likewise.
693 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
694 * tree-inline.c (expand_call_inline): Build and insert
695 debug_inline_entry stmt.
696 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
697 inline entry blocks early, if nonbind markers are enabled.
698 (dump_scope_block): Dump fragment info.
699 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
700 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
701 (gimple_build_debug_inline_entry): New.
702 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
703 Enable/disable inline entry points too.
704 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
705 (DEBUG_INSN): Describe inline entry markers.
707 * common.opt (gvariable-location-views): New.
708 (gvariable-location-views=incompat5): New.
709 * config.in: Rebuilt.
710 * configure: Rebuilt.
711 * configure.ac: Test assembler for view support.
712 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
713 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
714 * dwarf2out.c (var_loc_view): New typedef.
715 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
716 (dwarf2out_locviews_in_attribute): New.
717 (dwarf2out_locviews_in_loclist): New.
718 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
719 (enum dw_line_info_opcode): Add LI_adv_address.
720 (struct dw_line_info_table): Add view.
721 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
722 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
723 (zero_view_p): New variable.
724 (ZERO_VIEW_P): New macro.
725 (output_asm_line_debug_info): New.
726 (struct var_loc_node): Add view.
727 (add_AT_view_list, AT_loc_list): New.
728 (add_var_loc_to_decl): Add view param. Test it against last.
729 (new_loc_list): Add view params. Record them.
730 (AT_loc_list_ptr): Handle loc and view lists.
731 (view_list_to_loc_list_val_node): New.
732 (print_dw_val): Handle dw_val_class_view_list.
733 (size_of_die): Likewise.
734 (value_format): Likewise.
735 (loc_list_has_views): New.
736 (gen_llsym): Set vl_symbol too.
737 (maybe_gen_llsym, skip_loc_list_entry): New.
738 (dwarf2out_maybe_output_loclist_view_pair): New.
739 (output_loc_list): Output view list or entries too.
740 (output_view_list_offset): New.
741 (output_die): Handle dw_val_class_view_list.
742 (output_dwarf_version): New.
743 (output_compilation_unit_header): Use it.
744 (output_skeleton_debug_sections): Likewise.
745 (output_rnglists, output_line_info): Likewise.
746 (output_pubnames, output_aranges): Update version comments.
747 (output_one_line_info_table): Output view numbers in asm comments.
748 (dw_loc_list): Determine current endview, pass it to new_loc_list.
749 Call maybe_gen_llsym.
750 (loc_list_from_tree_1): Adjust.
751 (add_AT_location_description): Create view list attribute if
752 needed, check it's absent otherwise.
753 (convert_cfa_to_fb_loc_list): Adjust.
754 (maybe_emit_file): Call output_asm_line_debug_info for test.
755 (dwarf2out_var_location): Reset views as needed. Precompute
756 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
758 (new_line_info_table): Reset next view.
759 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
760 (dwarf2out_source_line): Likewise. Output view resets and labels to
761 the assembler, or select appropriate line info opcodes.
762 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
763 (optimize_string_length): Catch it. Adjust.
764 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
765 dw_val_class_view_list, and remove it if no longer needed.
766 (hash_loc_list): Hash view numbers.
767 (loc_list_hasher::equal): Compare them.
768 (optimize_location_lists): Check whether a view list symbol is
769 needed, and whether the locview attribute is present, and
770 whether they match. Remove the locview attribute if no longer
772 (index_location_lists): Call skip_loc_list_entry for test.
773 (dwarf2out_finish): Call output_asm_line_debug_info for test.
774 Use output_dwarf_version.
775 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
776 (struct dw_val_node): Add val_view_list.
777 * final.c (SEEN_NEXT_VIEW): New.
778 (set_next_view_needed): New.
779 (clear_next_view_needed): New.
780 (maybe_output_next_view): New.
781 (final_start_function): Rename to...
782 (final_start_function_1): ... this. Take pointer to FIRST,
783 add SEEN parameter. Emit param bindings in the initial view.
784 (final_start_function): Reintroduce SEEN-less interface.
785 (final): Rename to...
786 (final_1): ... this. Take SEEN parameter. Output final pending
787 next view at the end.
788 (final): Reintroduce seen-less interface.
789 (final_scan_insn): Output pending next view before switching
790 sections or ending a block. Mark the next view as needed when
791 outputting variable locations. Notify debug backend of section
792 changes, and of location view changes.
793 (rest_of_handle_final): Adjust.
794 * toplev.c (process_options): Autodetect value for debug variable
795 location views option. Warn on incompat5 without -gdwarf-5.
796 * doc/invoke.texi (gvariable-location-views): New.
797 (gvariable-location-views=incompat5): New.
798 (gno-variable-location-views): New.
800 2018-02-08 David Malcolm <dmalcolm@redhat.com>
802 PR tree-optimization/84136
803 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
804 that the result of find_edge is non-NULL.
806 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
809 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
810 storing integer register in SImode. Fix cost of 256 and 512
811 byte aligned SSE register store.
813 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
815 * config/i386/i386.c (ix86_multiplication_cost): Fix
816 multiplication cost for TARGET_AVX512DQ.
818 2018-02-08 Marek Polacek <polacek@redhat.com>
820 PR tree-optimization/84238
821 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
824 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
826 PR tree-optimization/84265
827 * tree-vect-stmts.c (vectorizable_store): Don't treat
828 VMAT_CONTIGUOUS accesses as grouped.
829 (vectorizable_load): Likewise.
831 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
833 PR tree-optimization/81635
834 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
835 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
836 (test_round_for_mask): New functions.
837 (wide_int_cc_tests): Call test_round_for_mask.
838 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
839 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
840 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
841 range returned by get_range_info.
843 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
846 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
847 * symtab.c: Include builtins.h
848 (symtab_node::output_to_lto_symbol_table_p): Move here
849 from lto-streamer-out.c:output_symbol_p.
850 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
851 (output_symbol_p): Move all logic to symtab.c
852 (produce_symtab): Update.
854 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
856 * config/s390/s390-opts.h (enum indirect_branch): Define.
857 * config/s390/s390-protos.h (s390_return_addr_from_memory)
858 (s390_indirect_branch_via_thunk)
859 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
860 (enum s390_indirect_branch_type): Define.
861 * config/s390/s390.c (struct s390_frame_layout, struct
862 machine_function): Remove.
863 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
864 (indirect_branch_table_label_no, indirect_branch_table_name):
866 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
867 (enum s390_indirect_branch_option): Define.
868 (s390_return_addr_from_memory): New function.
869 (s390_handle_string_attribute): New function.
870 (s390_attribute_table): Add new attribute handler.
871 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
872 (s390_indirect_branch_via_thunk): New function.
873 (s390_indirect_branch_via_inline_thunk): New function.
874 (s390_function_ok_for_sibcall): When jumping via thunk disallow
875 sibling call optimization for non z10 compiles.
876 (s390_emit_call): Force indirect branch target to be a single
877 register. Add r1 clobber for non-z10 compiles.
878 (s390_emit_epilogue): Emit return jump via return_use expander.
879 (s390_reorg): Handle JUMP_INSNs as execute targets.
880 (s390_option_override_internal): Perform validity checks for the
881 new command line options.
882 (s390_indirect_branch_attrvalue): New function.
883 (s390_indirect_branch_settings): New function.
884 (s390_set_current_function): Invoke s390_indirect_branch_settings.
885 (s390_output_indirect_thunk_function): New function.
886 (s390_code_end): Implement target hook.
887 (s390_case_values_threshold): Implement target hook.
888 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
890 * config/s390/s390.h (struct s390_frame_layout)
891 (struct machine_function): Move here from s390.c.
892 (TARGET_INDIRECT_BRANCH_NOBP_RET)
893 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
894 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
895 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
896 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
897 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
898 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
899 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
900 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
901 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
902 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
903 (mnemonic attribute): Add values which aren't recognized
905 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
906 pattern for branch conversion. Fix mnemonic attribute.
907 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
908 indirect branch via thunk if requested.
909 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
910 ("*indirect_jump"): Disable for branch conversion using out of
912 ("indirect_jump_via_thunk<mode>_z10")
913 ("indirect_jump_via_thunk<mode>")
914 ("indirect_jump_via_inlinethunk<mode>_z10")
915 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
916 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
917 ("casesi_jump_via_inlinethunk<mode>_z10")
918 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
919 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
920 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
921 ("*indirect2_jump"): Disable for branch conversion.
922 ("casesi_jump"): Turn into expander and expand patterns for branch
924 ("return_use"): New expander.
925 ("*return"): Emit return via thunk and rename it to ...
926 ("*return<mode>"): ... this one.
927 * config/s390/s390.opt: Add new options and and enum for the
930 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
932 * lra-constraints.c (match_reload): Unconditionally use
933 gen_lowpart_SUBREG, rather than selecting between that
934 and equivalent gen_rtx_SUBREG code.
936 2018-02-08 Richard Biener <rguenther@suse.de>
938 PR tree-optimization/84233
939 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
940 changed flag instead of boguously re-using phi_inserted.
942 2018-02-08 Martin Jambor <mjambor@suse.cz>
944 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
945 static local variables.
947 2018-02-08 Richard Biener <rguenther@suse.de>
949 PR tree-optimization/84278
950 * tree-vect-stmts.c (vectorizable_store): When looking for
951 smaller vector types to perform grouped strided loads/stores
952 make sure the mode is supported by the target.
953 (vectorizable_load): Likewise.
955 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
957 * config/aarch64/aarch64.c (aarch64_components_for_bb):
958 Increase LDP/STP opportunities by adding adjacent callee-saves.
960 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
962 PR rtl-optimization/84068
963 PR rtl-optimization/83459
964 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
966 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
968 PR tree-optimization/84224
969 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
970 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
973 2018-02-07 Iain Sandoe <iain@codesourcery.com>
976 * config/rs6000/altivec.md (*restore_world): Remove LR use.
977 * config/rs6000/predicates.md (restore_world_operation): Adjust op
978 count, remove one USE.
980 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
982 * doc/install.texi (Configuration): Document the
983 --with-long-double-format={ibm,ieee} PowerPC configuration
987 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
988 Convert from define_expand to be define_insn_and_split. Rework
989 float/double/_Float128 conversions to QI/HI/SImode to work with
990 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
991 conversions to QI/HImode types did a store and then a load to
992 truncate the value. For conversions to VSX registers, don't split
993 the insn, instead emit the code directly. Use the code iterator
994 any_fix to combine signed and unsigned conversions.
995 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
996 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
997 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
998 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
999 (fix_<mode>di2_hw): Likewise.
1000 (fixuns_<mode>di2_hw): Likewise.
1001 (fix_<mode>si2_hw): Likewise.
1002 (fixuns_<mode>si2_hw): Likewise.
1003 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
1004 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1005 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
1006 fix<uns>_trunc<SFDF:mode>si2_p8.
1007 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
1009 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1010 (fix<uns>_<mode>_mem): Likewise.
1011 (fctiw<u>z_<mode>_mem): Likewise.
1012 (fix<uns>_<mode>_mem): Likewise.
1013 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1014 the register allocator from doing a direct move to the GPRs to do
1015 a store, and instead use the ISA 3.0 store byte/half-word from
1016 vector register instruction. For IEEE 128-bit floating point,
1017 also optimize stores of 32-bit ints.
1018 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1020 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1022 * genextract.c (push_pathstr_operand): New function to support
1024 (walk_rtx): Call push_pathstr_operand.
1025 (print_path): Support [a-zA-Z].
1027 2018-02-07 Richard Biener <rguenther@suse.de>
1029 PR tree-optimization/84037
1030 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1031 (cse_and_gimplify_to_preheader): Declare.
1032 (vect_get_place_in_interleaving_chain): Likewise.
1033 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1035 (_loop_vec_info::~_loop_vec_info): Delete it.
1036 (cse_and_gimplify_to_preheader): New function.
1037 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1038 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1039 (vectorizable_load): Likewise. For grouped stores always base
1040 the IV on the first element.
1041 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1042 condition before gimplifying.
1044 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1046 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1047 *DIV_EXPR and *MOD_EXPR.
1049 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1052 * config/i386/i386.c (ix86_option_override_internal): Mask out
1053 the CF_SET bit when checking -fcf-protection.
1055 2018-02-07 Tom de Vries <tom@codesourcery.com>
1058 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1061 2018-02-07 Richard Biener <rguenther@suse.de>
1063 PR tree-optimization/84204
1064 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1067 PR tree-optimization/84205
1068 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1069 special-case isl_ast_op_zdiv_r.
1071 PR tree-optimization/84223
1072 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1073 Only add conditions from within the region.
1074 (gather_bbs::after_dom_children): Adjust.
1076 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1079 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1080 * config/avr/avr.md: Only post-reload split REG-REG moves if
1081 either register is REGERAL_REG_P.
1083 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1085 PR tree-optimization/84235
1086 * tree-ssa-scopedtables.c
1087 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1088 if the subtraction is performed in floating point type where NaNs are
1089 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1090 build 1. Formatting fix.
1092 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1095 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1096 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1097 and skip it regardless of bb boundaries. Use CALL_P macro,
1098 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1099 unnecessarily, formatting fix.
1101 2018-02-06 Michael Collison <michael.collison@arm.com>
1103 * config/arm/thumb2.md:
1104 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1105 (*thumb_mov_notscc): Ditto.
1107 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1110 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1111 unsigned_fix, not "s".
1113 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1115 * configure.ac (gcc_fn_eh_frame_ro): New function.
1116 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1117 correct .eh_frame permissions.
1118 * configure: Regenerate.
1120 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1122 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1125 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1127 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1128 Display warning message for -mno-speculate-indirect-jumps.
1130 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1132 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1134 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1136 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1138 PR tree-optimization/84225
1139 * tree-eh.c (find_trapping_overflow): Only call
1140 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1142 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1145 * config/i386/i386.c: Reimplement the check of possible options
1146 -mibt/-mshstk conbination. Change error messages.
1147 * doc/invoke.texi: Fix a typo: remove extra '='.
1149 2018-02-06 Marek Polacek <polacek@redhat.com>
1151 PR tree-optimization/84228
1152 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1154 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1157 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1158 emitted arch directives.
1159 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1160 __ARM_FEATURE_COPROC before changing architectures.
1162 2018-02-06 Richard Biener <rguenther@suse.de>
1164 * config/i386/i386.c (print_reg): Fix typo.
1165 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1167 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1169 * configure: Regenerate.
1171 2018-02-05 Martin Sebor <msebor@redhat.com>
1173 PR tree-optimization/83369
1174 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1177 2018-02-05 Martin Liska <mliska@suse.cz>
1179 * doc/invoke.texi: Cherry-pick upstream r323995.
1181 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1183 * ira.c (ira_init_register_move_cost): Adjust comment.
1185 2018-02-05 Martin Liska <mliska@suse.cz>
1187 PR gcov-profile/84137
1188 * doc/gcov.texi: Fix typo in documentation.
1190 2018-02-05 Martin Liska <mliska@suse.cz>
1192 PR gcov-profile/83879
1193 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1194 using dlopen functionality.
1196 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1198 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1199 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1200 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1201 _mm_maskz_range_round_ss): New intrinsics.
1202 (__builtin_ia32_rangesd128_round)
1203 (__builtin_ia32_rangess128_round): Remove.
1204 (__builtin_ia32_rangesd128_mask_round,
1205 __builtin_ia32_rangess128_mask_round): New builtins.
1206 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1207 __builtin_ia32_rangess128_round): Remove.
1208 (__builtin_ia32_rangesd128_mask_round,
1209 __builtin_ia32_rangess128_mask_round): New builtins.
1210 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1211 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1212 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1213 "<round_saeonly_constraint>")): Changed to ...
1214 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1215 "<round_saeonly_scalar_constraint>")): ... this.
1216 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1217 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1218 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1219 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1220 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1222 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1224 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1226 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1227 Remove all values except native, 8540 and 8548.
1229 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1231 * config/i386/i386.c (ix86_output_function_return): Pass
1232 INVALID_REGNUM, instead of -1, as invalid register number to
1233 indirect_thunk_name and output_indirect_thunk.
1235 2018-02-02 Julia Koval <julia.koval@intel.com>
1237 * config.gcc: Add -march=icelake.
1238 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1239 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1240 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1241 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1242 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1243 (processor_target_table): Add icelake.
1244 (ix86_option_override_internal): Handle new PTAs.
1245 (get_builtin_code_for_version): Handle icelake.
1246 (M_INTEL_COREI7_ICELAKE): New.
1247 (fold_builtin_cpu): Handle icelake.
1248 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1249 * doc/invoke.texi: Add -march=icelake.
1251 2018-02-02 Julia Koval <julia.koval@intel.com>
1253 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1254 to wide_int_bitmask.
1255 * wide-int-bitmask.h: New.
1257 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1260 * config/i386/i386.md: Replace Pmode with word_mode in
1261 builtin_setjmp_setup and builtin_longjmp to support x32.
1263 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1267 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1269 (rs6000_supported_cpu_names): New static variable.
1270 (linux_cpu_translation_table): Likewise.
1271 (elf_platform) <cpu>: Define new static variable and use it.
1272 Translate kernel AT_PLATFORM name to canonical name if needed.
1273 Error if platform name is unknown.
1275 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1278 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1280 2018-02-01 Jeff Law <law@redhat.com>
1283 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1284 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1285 the scratch if RELEASE_VIA_POP is false.
1286 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1287 If we have to save a temporary register, decrement SIZE appropriately.
1288 Pass new arguments to release_scratch_register_on_entry.
1289 (ix86_adjust_stack_and_probe): Likewise.
1290 (ix86_emit_probe_stack_range): Pass new arguments to
1291 release_scratch_register_on_entry.
1293 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1295 PR rtl-optimization/84157
1296 * combine.c (change_zero_ext): Use REG_P predicate in
1297 front of HARD_REGISTER_P predicate.
1299 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1301 * config/avr/avr.c (avr_option_override): Move disabling of
1302 -fdelete-null-pointer-checks to...
1303 * common/config/avr/avr-common.c (avr_option_optimization_table):
1306 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1308 PR tree-optimization/81635
1309 * tree-data-ref.c (split_constant_offset_1): For types that
1310 wrap on overflow, try to use range info to prove that wrapping
1313 2018-02-01 Renlin Li <renlin.li@arm.com>
1316 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1318 (aarch64_register_move_cost): Likewise.
1319 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1321 (REG_CLASS_NAMES): Likewise.
1322 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1323 TAILCALL_ADDR_REGS. Remove IP registers.
1324 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1326 2018-02-01 Richard Biener <rguenther@suse.de>
1328 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1329 for specifying RPO order and allow NULL for that.
1330 * domwalk.c (dom_walker::dom_walker): Likewise.
1331 (dom_walker::walk): Handle NULL RPO order.
1332 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1334 (rewrite_update_dom_walker): Likewise.
1335 (mark_def_dom_walker): Likewise.
1337 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1339 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1340 (aarch64_maybe_expand_sve_subreg_move): Declare.
1341 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1342 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1344 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1345 that are semantically a reverse operation.
1346 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1347 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1348 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1350 (aarch64_can_change_mode_class): For big-endian, forbid changes
1351 between two SVE modes if they have different element sizes.
1353 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1355 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1356 the TImode handling for big-endian targets.
1358 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1360 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1361 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1363 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1364 Remove BSWAP handing for big-endian targets and use the form of
1365 LD1RQ appropariate for the mode.
1367 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1369 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1370 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1373 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1376 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1377 check for operands that need to go through aarch64_sve_reload_be.
1379 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1381 PR tree-optimization/81661
1382 PR tree-optimization/84117
1383 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1384 * tree-eh.c: Include gimplify.h.
1385 (find_trapping_overflow, replace_trapping_overflow,
1386 rewrite_to_non_trapping_overflow): New functions.
1387 * tree-vect-loop.c: Include tree-eh.h.
1388 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1389 * tree-data-ref.c: Include tree-eh.h.
1390 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1392 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1394 PR rtl-optimization/84123
1395 * combine.c (change_zero_ext): Check if hard register satisfies
1396 can_change_dest_mode before calling gen_lowpart_SUBREG.
1398 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1401 * ira.c (ira_init_register_move_cost): Remove assert.
1403 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1405 PR rtl-optimization/84071
1406 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1407 * doc/tm.texi: Regenerate.
1409 2018-01-31 Richard Biener <rguenther@suse.de>
1411 PR tree-optimization/84132
1412 * tree-data-ref.c (analyze_miv_subscript): Properly
1413 check whether evolution_function_is_affine_multivariate_p
1414 before calling gcd_of_steps_may_divide_p.
1416 2018-01-31 Julia Koval <julia.koval@intel.com>
1419 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1420 * config/i386/i386.md (rdpid_rex64) New.
1421 (rdpid): Make 32bit only.
1423 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1426 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1427 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1429 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1432 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1434 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1436 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1438 PR rtl-optimization/84071
1439 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1440 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1442 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1444 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1445 (arc_attribute_table): Add 'aux' attribute.
1446 (arc_in_small_data_p): Consider aux like variables.
1447 (arc_is_aux_reg_p): New function.
1448 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1449 (arc_get_aux_arg): New function.
1450 (prepare_move_operands): Handle aux-register access.
1451 (arc_handle_aux_attribute): New function.
1452 * doc/extend.texi (ARC Variable attributes): Add subsection.
1454 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1456 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1457 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1458 (arc_attribute_table): Add 'uncached' attribute.
1459 (arc_print_operand): Print '.di' flag for uncached memory
1461 (arc_in_small_data_p): Do not consider for small data the uncached
1463 (arc_is_uncached_mem_p): New function.
1464 * config/arc/predicates.md (compact_store_memory_operand): Check
1465 for uncached memory accesses.
1466 (nonvol_nonimm_operand): Likewise.
1467 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1469 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1472 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1473 falign-loops=): Add Optimization flag.
1475 2018-01-30 Jeff Law <law@redhat.com>
1478 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1479 INT_REGISTERS_SAVED. Check it prior to calling
1480 get_scratch_register_on_entry.
1481 (ix86_adjust_stack_and_probe): Similarly.
1482 (ix86_emit_probe_stack_range): Similarly.
1483 (ix86_expand_prologue): Corresponding changes.
1485 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1488 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1489 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1491 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1494 * lra-constraints.c (curr_insn_transform): Process AND in the
1497 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1499 PR rtl-optimization/83986
1500 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1501 dependence against last_pending_memory_flush in addition to
1504 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1506 PR tree-optimization/81611
1507 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1510 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1513 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1516 2018-01-30 Richard Biener <rguenther@suse.de>
1517 Jakub Jelinek <jakub@redhat.com>
1519 PR tree-optimization/84111
1520 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1521 inner loops added during recursion, as they don't have up-to-date
1524 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1527 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1528 (can_inline_edge_by_limits_p): ... here.
1529 (can_early_inline_edge_p, check_callers,
1530 update_caller_keys, update_callee_keys, recursive_inlining,
1531 add_new_edges_to_heap, speculation_useful_p,
1532 inline_small_functions,
1533 inline_small_functions, flatten_function,
1534 inline_to_all_callers_1): Update.
1536 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1538 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1541 2018-01-30 Richard Biener <rguenther@suse.de>
1543 PR tree-optimization/83008
1544 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1545 invariant and constant vector uses in stmts when they need
1548 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1551 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1552 * configure: Regenerate.
1554 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1556 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1558 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1559 Use gen_rtx_REG rather than gen_lowpart.
1561 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1563 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1564 rather than 0 when creating partial subregs.
1566 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1568 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1571 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1574 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1575 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1576 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1577 flags. This restores the settings used before the 2017-07-24.
1578 Turning off pre increment/decrement/modify allows IVOPTS to
1579 optimize DF/SF loops where the index is an int.
1581 2018-01-29 Richard Biener <rguenther@suse.de>
1582 Kelvin Nilsen <kelvin@gcc.gnu.org>
1585 * tree-vect-stmts.c (vectorizable_call): Don't call
1586 targetm.vectorize_builtin_md_vectorized_function if callee is
1589 2018-01-22 Carl Love <cel@us.ibm.com>
1591 * doc/extend.tex: Fix typo in second arg in
1592 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1594 2018-01-29 Richard Biener <rguenther@suse.de>
1596 PR tree-optimization/84086
1597 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1598 (flush_ssaname_freelist): When SSA names were released reset
1599 the SCEV hash table.
1601 2018-01-29 Richard Biener <rguenther@suse.de>
1603 PR tree-optimization/84057
1604 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1605 removed paths when removing edges.
1607 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1609 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1610 -mfunction-return=@var{choice}.
1612 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1615 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1617 (layout::print_source_line): Likewise.
1618 (test_get_line_width_without_trailing_whitespace): Add test cases.
1620 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1623 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1626 2018-01-26 Jim Wilson <jimw@sifive.com>
1628 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1630 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1633 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1635 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1636 and CMP + SUB-immediate -> SUBS.
1638 2018-01-26 Martin Sebor <msebor@redhat.com>
1640 PR tree-optimization/83896
1641 * tree-ssa-strlen.c (get_string_len): Rename...
1642 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1643 Avoid assuming length is constant.
1644 (handle_char_store): Use HOST_WIDE_INT for string length.
1646 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1649 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1650 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1652 2018-01-26 Richard Biener <rguenther@suse.de>
1654 PR rtl-optimization/84003
1655 * dse.c (record_store): Only record redundant stores when
1656 the earlier store aliases at least all accesses the later one does.
1658 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1660 PR rtl-optimization/83985
1661 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1662 REG_CFA_RESTORE insns.
1663 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1664 REG_CFA_RESTORE insns here.
1667 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1668 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1670 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1672 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1673 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1674 (arc_init): Likewise.
1675 (arc_override_options): Likewise.
1676 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1678 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1680 * config/arc/arc.h (TARGET_DBNZ): Define.
1681 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1682 properly set the tune attribute.
1683 (dbnz): Use TARGET_DBNZ guard.
1684 * config/arc/arc.opt (mtune): Add core3 option.
1686 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1688 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1689 recognize new pic like addresses.
1690 (arc_delegitimize_address): Clean up.
1692 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1694 * config/arc/arc-arches.def: Option mrf16 valid for all
1696 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1697 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1698 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1699 * config/arc/arc-tables.opt: Regenerate.
1700 * config/arc/arc.c (arc_conditional_register_usage): Handle
1701 reduced register file case.
1702 (arc_file_start): Set must have build attributes.
1703 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1705 * config/arc/arc.opt (mrf16): Add new option.
1706 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1707 * config/arc/genmultilib.awk: Handle new mrf16 option.
1708 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1709 * config/arc/t-multilib: Regenerate.
1710 * doc/invoke.texi (ARC Options): Document mrf16 option.
1712 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1714 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1715 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1716 (arc_attribute_table): Add 'secure_call' attribute.
1717 (arc_print_operand): Print secure call operand.
1718 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1720 (arc_is_secure_call_p): New function. * config/arc/arc.md
1721 (call_i): Add support for sjli instruction.
1722 (call_value_i): Likewise.
1723 * config/arc/constraints.md (Csc): New constraint.
1725 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1726 John Eric Martin <John.Martin@emmicro-us.com>
1728 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1729 * config/arc/arc.c (_arc_jli_section): New struct.
1730 (arc_jli_section): New type.
1731 (rc_jli_sections): New static variable.
1732 (arc_handle_jli_attribute): New function.
1733 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1734 (arc_file_end): New function.
1735 (TARGET_ASM_FILE_END): Define.
1736 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1737 (arc_add_jli_section): New function.
1738 (jli_call_scan): Likewise.
1739 (arc_reorg): Call jli_call_scan.
1740 (arc_output_addsi): Remove 'S' from printing asm operand.
1741 (arc_is_jli_call_p): New function.
1742 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1744 (movhi_insn): Likewise.
1745 (movsi_insn): Likewise.
1746 (movsi_set_cc_insn): Likewise.
1747 (loadqi_update): Likewise.
1748 (load_zeroextendqisi_update): Likewise.
1749 (load_signextendqisi_update): Likewise.
1750 (loadhi_update): Likewise.
1751 (load_zeroextendhisi_update): Likewise.
1752 (load_signextendhisi_update): Likewise.
1753 (loadsi_update): Likewise.
1754 (loadsf_update): Likewise.
1755 (movsicc_insn): Likewise.
1756 (bset_insn): Likewise.
1757 (bxor_insn): Likewise.
1758 (bclr_insn): Likewise.
1759 (bmsk_insn): Likewise.
1760 (bicsi3_insn): Likewise.
1761 (cmpsi_cc_c_insn): Likewise.
1762 (movsi_ne): Likewise.
1763 (movsi_cond_exec): Likewise.
1764 (clrsbsi2): Likewise.
1775 (call_i): Remove 'S' asm letter, add jli instruction.
1776 (call_value_i): Likewise.
1777 * config/arc/arc.op (mjli-always): New option.
1778 * config/arc/constraints.md (Cji): New constraint.
1779 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1781 (subsf3_fpx): Likewise.
1782 (mulsf3_fpx): Likewise.
1783 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1785 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1787 * doc/invoke.texi (ARC): Document mjli-always option.
1789 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1791 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
1792 avoid addition with 0 and use incw and decw where possible.
1794 2018-01-26 Richard Biener <rguenther@suse.de>
1796 PR tree-optimization/81082
1797 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1798 association if it requires casting to unsigned.
1799 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1800 from fold_plusminus_mult_expr to catch important cases late when
1801 range info is available.
1803 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1805 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1806 * configure.ac (hidden_linkonce): New test.
1807 * configure: Regenerate.
1808 * config.in: Regenerate.
1810 2018-01-26 Julia Koval <julia.koval@intel.com>
1812 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1813 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1814 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1815 _mm_mask_bitshuffle_epi64_mask): Fix type.
1816 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1817 USI_FTYPE_V4DI_V4DI_USI): Remove.
1818 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1819 __builtin_ia32_vpshufbitqmb256_mask,
1820 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1821 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1822 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1824 2018-01-26 Alan Modra <amodra@gmail.com>
1827 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1828 UNSPEC_VBPERMQ. Sort other unspecs.
1830 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1832 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1834 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1837 * predict.c (drop_profile): Do not push/pop cfun; update also
1839 (handle_missing_profiles): Fix logic looking for zero profiles.
1841 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1844 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1845 on functions with #pragma omp declare simd or functions with simd
1847 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1848 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1849 Remove trailing \n from warning_at calls.
1851 2018-01-25 Tom de Vries <tom@codesourcery.com>
1854 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1855 for neutered workers.
1857 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1860 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1861 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1863 2018-01-24 Jeff Law <law@redhat.com>
1866 * i386.c (get_probe_interval): Move to earlier point.
1867 (ix86_compute_frame_layout): If -fstack-clash-protection and
1868 the frame is larger than the probe interval, then use pushes
1869 to save registers rather than reg->mem moves.
1870 (ix86_expand_prologue): Remove conditional for int_registers_saved
1873 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1876 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1877 min/max for never referenced object.
1879 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1882 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1884 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1885 attributes from DECL_ATTRIBUTES (decl) without affecting
1886 DECL_ATTRIBUTES (current_function_decl).
1887 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1888 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1890 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1892 PR tree-optimization/83979
1893 * fold-const.c (fold_comparison): Use constant_boolean_node
1894 instead of boolean_{true,false}_node.
1896 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1898 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1901 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1903 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1904 Simplify the clause that sets the length attribute.
1905 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1906 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1907 clause that sets the length attribute.
1908 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1910 2018-01-24 Tom de Vries <tom@codesourcery.com>
1913 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1914 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1915 Add strict parameter.
1916 (prevent_branch_around_nothing): Insert dummy insn between branch to
1917 label and label with no ptx insn inbetween.
1918 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1920 2018-01-24 Tom de Vries <tom@codesourcery.com>
1923 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1924 for neutered threads in warp.
1925 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1927 2018-01-24 Richard Biener <rguenther@suse.de>
1929 PR tree-optimization/83176
1930 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1933 2018-01-24 Richard Biener <rguenther@suse.de>
1935 PR tree-optimization/82819
1936 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1937 code generating pluses that are no-ops in the target precision.
1939 2018-01-24 Richard Biener <rguenther@suse.de>
1942 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1944 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1946 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1947 to merge probabilities.
1948 * predict.c (probably_never_executed): Also mark as cold functions
1949 with global 0 profile and guessed local profile.
1950 * profile-count.c (profile_probability::combine_with_count): New
1952 * profile-count.h (profile_probability::operator*,
1953 profile_probability::operator*=, profile_probability::operator/,
1954 profile_probability::operator/=): Reduce precision to adjusted
1955 and set value to guessed on contradictory divisions.
1956 (profile_probability::combine_with_freq): Remove.
1957 (profile_probability::combine_wiht_count): Declare.
1958 (profile_count::force_nonzero):: Set to adjusted.
1959 (profile_count::probability_in):: Set quality to adjusted.
1960 * tree-ssa-tail-merge.c (replace_block_by): Use
1963 2018-01-23 Andrew Waterman <andrew@sifive.com>
1964 Jim Wilson <jimw@sifive.com>
1966 * config/riscv/riscv.c (riscv_stack_boundary): New.
1967 (riscv_option_override): Set riscv_stack_boundary. Handle
1968 riscv_preferred_stack_boundary_arg.
1969 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1970 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1971 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1972 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1973 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1974 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1976 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1979 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1980 of struct ix86_frame.
1981 (ix86_expand_epilogue): Likewise. Add a local variable for
1982 the reg_save_offset field in struct ix86_frame.
1984 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1986 PR tree-optimization/82604
1987 * tree-loop-distribution.c (enum partition_kind): New enum item
1988 PKIND_PARTIAL_MEMSET.
1989 (partition_builtin_p): Support above new enum item.
1990 (generate_code_for_partition): Ditto.
1991 (compute_access_range): Differentiate cases that equality can be
1992 proven at all loops, the innermost loops or no loops.
1993 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1994 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1995 (finalize_partitions, distribute_loop): Don't fuse partition of
1996 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1997 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
2000 2018-01-23 Martin Liska <mliska@suse.cz>
2002 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
2003 order to ignore the predictor.
2004 (PRED_POLYMORPHIC_CALL): Likewise.
2005 (PRED_RECURSIVE_CALL): Likewise.
2007 2018-01-23 Martin Liska <mliska@suse.cz>
2009 * tree-profile.c (tree_profiling): Print function header to
2010 aware reader which function we are working on.
2011 * value-prof.c (gimple_find_values_to_profile): Do not print
2012 not interesting value histograms.
2014 2018-01-23 Martin Liska <mliska@suse.cz>
2016 * profile-count.h (enum profile_quality): Add
2017 profile_uninitialized as the first value. Do not number values
2018 as they are zero based.
2019 (profile_count::verify): Update sanity check.
2020 (profile_probability::verify): Likewise.
2022 2018-01-23 Nathan Sidwell <nathan@acm.org>
2024 * doc/invoke.texi (ffor-scope): Deprecate.
2026 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2028 PR tree-optimization/83510
2029 * domwalk.c (set_all_edges_as_executable): New function.
2030 (dom_walker::dom_walker): Convert bool param
2031 "skip_unreachable_blocks" to enum reachability. Move setup of
2032 edge flags to set_all_edges_as_executable and only do it when
2033 reachability is REACHABLE_BLOCKS.
2034 * domwalk.h (enum dom_walker::reachability): New enum.
2035 (dom_walker::dom_walker): Convert bool param
2036 "skip_unreachable_blocks" to enum reachability.
2037 (set_all_edges_as_executable): New decl.
2038 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2039 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2041 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2042 but converting true to REACHABLE_BLOCKS.
2043 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2045 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2046 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2047 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2049 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2050 if check_all_array_refs will be called.
2052 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2054 * tree.c (selftest::test_location_wrappers): Add more test
2057 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2059 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2060 (selftest::test_bit_in_range): Likewise.
2062 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2065 * doc/sourcebuild.texi (vect_float): Say that the selector
2066 only describes the situation when -funsafe-math-optimizations is on.
2067 (vect_float_strict): Document.
2069 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2071 PR tree-optimization/83965
2072 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2073 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2074 instead of checking only for a reduction.
2075 (vect_recog_widen_sum_pattern): Likewise.
2077 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2079 * predict.c (probably_never_executed): Only use precise profile info.
2080 (compute_function_frequency): Skip after inlining hack since we now
2081 have quality checking.
2083 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2085 * profile-count.h (profile_probability::very_unlikely,
2086 profile_probability::unlikely, profile_probability::even): Set
2087 precision to guessed.
2089 2018-01-23 Richard Biener <rguenther@suse.de>
2091 PR tree-optimization/83963
2092 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2093 Properly terminate dominator walk when crossing the exit edge not
2094 when visiting its source block.
2096 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2099 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2100 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2102 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2104 PR tree-optimization/83957
2105 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2106 semicolon after for body surrounded by braces.
2108 PR tree-optimization/83081
2109 * profile-count.h (profile_probability::split): New method.
2110 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2111 Use profile_probability::split.
2112 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2113 when splitting a single conditional jump into 2.
2115 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2117 PR tree-optimization/69452
2118 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2121 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2123 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2124 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2125 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2127 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2129 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2130 * config/rl78/rl78.md (movdi): New define_expand.
2131 * config/rl78/rl78.c (rl78_split_movdi): New function.
2133 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2136 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2138 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2139 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2140 128-bit to produce an UNSPEC move to get the double word with the
2141 signbit and then a shift directly to do signbit.
2142 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2143 implementation with a new version that just does either a direct
2144 move or a regular move. Move memory interface to separate insns.
2145 Move insns so they are next to the expander.
2146 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2147 with signbit move. Split big and little endian case.
2148 (signbit<mode>2_dm_mem_le): Likewise.
2149 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2150 (signbit<mode>2_dm2): Likewise.
2152 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2154 * config/rl78/rl78.md (anddi3): New define_expand.
2156 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2158 * config/rl78/rl78.md (umindi3): New define_expand.
2160 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2162 * config/rl78/rl78.md (smindi3): New define_expand.
2164 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2166 * config/rl78/rl78.md (smaxdi3): New define_expand.
2168 2018-01-22 Carl Love <cel@us.ibm.com>
2170 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2171 LVX_V1TI): Add macro expansion.
2172 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2173 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2174 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2175 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2176 Change check to determine if the instruction is a byte reversing
2177 entry. Fix typo in comment.
2178 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2179 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2180 Add def_builtin calls for new builtins.
2181 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2182 Add define_insn expansion.
2184 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2186 * config/rl78/rl78.md (umaxdi3): New define_expand.
2188 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2190 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2191 for non-QImode registers.
2193 2018-01-22 Richard Biener <rguenther@suse.de>
2195 PR tree-optimization/83963
2196 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2197 including the loop exit block.
2198 (scop_detection::merge_sese): Likewise.
2199 (scop_detection::add_scop): Do it here instead.
2201 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2203 * doc/sourcebuild.texi (arm_softfloat): Document.
2205 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2208 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2209 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2210 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2212 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2213 David Edelsohn <dje.gcc@gmail.com>
2216 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2217 Change "crset eq" to "crset 2".
2218 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2219 (*call_indirect_aix<mode>_nospec): Likewise.
2220 (*call_value_indirect_aix<mode>_nospec): Likewise.
2221 (*call_indirect_elfv2<mode>_nospec): Likewise.
2222 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2223 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2224 change assembly output from . to $.
2225 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2226 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2227 (*tablejump<mode>_internal1_nospec): Likewise.
2229 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2232 * config/sh/sh_optimize_sett_clrt.cc:
2233 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2235 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2237 PR tree-optimization/83940
2238 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2239 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2240 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2241 use it to pass back the definition type.
2242 (vect_check_store_rhs): Likewise rhs_dt_out.
2243 (vect_build_gather_load_calls): Add a mask_dt argument and use
2244 it instead of a call to vect_is_simple_use.
2245 (vectorizable_store): Update calls to vect_check_load_store_mask
2246 and vect_check_store_rhs. Use the dt returned by the latter instead
2247 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2248 instead of calls to vect_is_simple_use. Pass the scalar rather
2249 than the vector operand to vect_is_simple_use when handling
2250 second and subsequent copies of an rhs value.
2251 (vectorizable_load): Update calls to vect_check_load_store_mask
2252 and vect_build_gather_load_calls. Use the cached mask_dt and
2253 gs_info.offset_dt instead of calls to vect_is_simple_use.
2255 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2258 * tree-emutls.c: Include gimplify.h.
2259 (lower_emutls_2): New function.
2260 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2261 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2262 it before further processing.
2265 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2266 UINTVAL (trueop1) instead of INTVAL (op1).
2268 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2272 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2273 INCOMING_FRAME_SP_OFFSET if not defined.
2274 (scan_trace): Add ENTRY argument. If true and
2275 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2276 emit a note to adjust the CFA offset.
2277 (create_cfi_notes): Adjust scan_trace callers.
2278 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2279 INCOMING_FRAME_SP_OFFSET in the CIE.
2280 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2281 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2283 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2284 * doc/tm.texi: Regenerated.
2286 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2288 PR rtl-optimization/83147
2289 * lra-constraints.c (remove_inheritance_pseudos): Use
2290 lra_substitute_pseudo_within_insn.
2292 2018-01-19 Tom de Vries <tom@codesourcery.com>
2293 Cesar Philippidis <cesar@codesourcery.com>
2296 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2298 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2301 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2302 spaces for function labels.
2304 2018-01-19 Martin Liska <mliska@suse.cz>
2306 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2307 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2308 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2309 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2310 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2311 (PRED_CONST_RETURN): Change from 69 to 65.
2312 (PRED_NULL_RETURN): Change from 91 to 71.
2313 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2314 (PRED_LOOP_GUARD): Change from 66 to 73.
2316 2018-01-19 Martin Liska <mliska@suse.cz>
2318 * predict.c (predict_insn_def): Add new assert.
2319 (struct branch_predictor): Change type to signed integer.
2320 (test_prediction_value_range): Amend test to cover
2322 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2323 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2324 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2325 (PRED_LOOP_IV_COMPARE): Likewise.
2326 * predict.h (PROB_UNINITIALIZED): Define new constant.
2328 2018-01-19 Martin Liska <mliska@suse.cz>
2330 * predict.c (dump_prediction): Add new format for
2331 analyze_brprob.py script which is enabled with -details
2333 * profile-count.h (precise_p): New function.
2335 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2337 PR tree-optimization/83922
2338 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2339 there are no statements that need masking.
2340 (vect_active_double_reduction_p): New function.
2341 (vect_analyze_loop_operations): Use it when handling phis that
2342 are not in the loop header.
2344 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2346 PR tree-optimization/83914
2347 * tree-vect-loop.c (vectorizable_induction): Don't convert
2348 init_expr or apply the peeling adjustment for inductions
2349 that are nested within the vectorized loop.
2351 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2353 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2356 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2360 * function.h (gimplify_parameters): Add gimple_seq * argument.
2361 * function.c: Include gimple.h and options.h.
2362 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2363 for the added local temporaries if needed.
2364 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2365 if there are any parameter cleanups, wrap whole body into a
2366 try/finally with the cleanups.
2368 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2371 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2372 Use GET_MODE_CLASS for scalar floating point.
2374 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2378 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2379 Fix call of call_cgraph_insertion_hooks.
2381 2018-01-18 Martin Sebor <msebor@redhat.com>
2383 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2385 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2388 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2391 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2394 * common.opt: (-ffile-prefix-map): New option.
2395 * opts.c (common_handle_option): Defer it.
2396 * opts-global.c (handle_common_deferred_options): Handle it.
2397 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2398 * file-prefix-map.h: New file.
2399 (remap_debug_filename, add_debug_prefix_map): ...here.
2400 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2401 * final.c (debug_prefix_map, add_debug_prefix_map
2402 remap_debug_filename): Move to...
2403 * file-prefix-map.c: New file.
2404 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2405 generalize, get rid of alloca(), use strrchr() instead of strchr().
2406 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2407 Implement in terms of add_prefix_map().
2408 (remap_macro_filename, remap_debug_filename): Implement in term of
2410 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2411 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2412 * dbxout.c: Include file-prefix-map.h.
2413 * varasm.c: Likewise.
2414 * vmsdbgout.c: Likewise.
2415 * xcoffout.c: Likewise.
2416 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2417 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2418 * doc/invoke.texi (-ffile-prefix-map): Document.
2419 (-fdebug-prefix-map): Update description.
2421 2018-01-18 Martin Liska <mliska@suse.cz>
2423 * config/i386/i386.c (indirect_thunk_name): Document that also
2425 (output_indirect_thunk): Document why both instructions
2426 (pause and lfence) are generated.
2428 2018-01-18 Richard Biener <rguenther@suse.de>
2430 PR tree-optimization/83887
2431 * graphite-scop-detection.c
2432 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2433 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2434 (scop_detection::merge_sese): Re-implement with a flood-fill
2435 algorithm that properly finds a SESE region if it exists.
2437 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2440 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2441 pointer_diff optimizations use view_convert instead of convert.
2443 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2445 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2446 Generate different code for -mno-speculate-indirect-jumps.
2447 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2448 (*call_indirect_aix<mode>): Disable for
2449 -mno-speculate-indirect-jumps.
2450 (*call_indirect_aix<mode>_nospec): New define_insn.
2451 (*call_value_indirect_aix<mode>): Disable for
2452 -mno-speculate-indirect-jumps.
2453 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2454 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2455 -mno-speculate-indirect-jumps.
2456 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2458 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2460 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2461 long double type, set the flags for noting the default long double
2462 type, even if we don't pass or return a long double type.
2464 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2467 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2470 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2472 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2473 support for merge[hl].
2474 (fold_mergehl_helper): New helper function.
2475 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2476 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2477 (altivec_vmrglw_direct): Add xxmrglw insn.
2479 2018-01-17 Andrew Waterman <andrew@sifive.com>
2481 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2482 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2484 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2487 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2488 call the lto_location_cache before reading the
2489 DECL_SOURCE_LOCATION of the types.
2491 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2492 Richard Sandiford <richard.sandiford@linaro.org>
2494 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2495 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2496 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2497 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2498 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2500 * config/aarch64/constraints.md (aarch64_movti_operand):
2502 * config/aarch64/predicates.md (Uti): Add new constraint.
2504 2018-01-17 Carl Love <cel@us.ibm.com>
2506 * config/rs6000/vsx.md (define_expand xl_len_r,
2507 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2508 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2510 (define_expand, define_insn): Move the shift left from the
2511 define_insn to the define_expand for lxvl and stxvl instructions.
2512 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2513 and XL_LEN_R definitions to PURE.
2515 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2517 * config/i386/i386.c (indirect_thunk_name): Declare regno
2518 as unsigned int. Compare regno with INVALID_REGNUM.
2519 (output_indirect_thunk): Ditto.
2520 (output_indirect_thunk_function): Ditto.
2521 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2522 in the call to output_indirect_thunk_function.
2524 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2527 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2528 rather than the size of inner_type to determine the stack slot size
2529 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2531 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2534 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2537 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2539 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2540 endian Linux systems to optionally enable multilibs for selecting
2541 the long double type if the user configured an explicit type.
2542 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2543 have no long double multilibs if not defined.
2544 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2545 warn if the user used -mabi={ieee,ibm}longdouble and we built
2546 multilibs for long double.
2547 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2548 appropriate multilib option.
2549 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2551 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2552 for building long double multilibs.
2553 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2555 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2557 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2560 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2562 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2565 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2568 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2571 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2573 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2574 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2576 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2578 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2579 different rtl trees depending on TARGET_64BIT.
2580 (rs6000_gen_lvx): Likewise.
2582 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2584 * config/visium/visium.md (nop): Tweak comment.
2585 (hazard_nop): Likewise.
2587 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2589 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2590 -mspeculate-indirect-jumps.
2591 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2592 for -mno-speculate-indirect-jumps.
2593 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2594 (*call_value_indirect_elfv2<mode>): Disable for
2595 -mno-speculate-indirect-jumps.
2596 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2597 (indirect_jump): Emit different RTL for
2598 -mno-speculate-indirect-jumps.
2599 (*indirect_jump<mode>): Disable for
2600 -mno-speculate-indirect-jumps.
2601 (*indirect_jump<mode>_nospec): New define_insn.
2602 (tablejump): Emit different RTL for
2603 -mno-speculate-indirect-jumps.
2604 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2605 (tablejumpsi_nospec): New define_expand.
2606 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2607 (tablejumpdi_nospec): New define_expand.
2608 (*tablejump<mode>_internal1): Disable for
2609 -mno-speculate-indirect-jumps.
2610 (*tablejump<mode>_internal1_nospec): New define_insn.
2611 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2614 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2616 * caller-save.c (insert_save): Drop unnecessary parameter. All
2619 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2620 Richard Biener <rguenth@suse.de>
2623 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2624 return early, inline manually is_gimple_sizepos. Make sure if we
2625 call gimplify_expr we don't end up with a gimple constant.
2626 * tree.c (variably_modified_type_p): Don't return true for
2627 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2628 * gimplify.h (is_gimple_sizepos): Remove.
2630 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2632 PR tree-optimization/83857
2633 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2634 vectorizable_live_operation for pure SLP statements.
2635 (vectorizable_live_operation): Handle PHIs.
2637 2018-01-16 Richard Biener <rguenther@suse.de>
2639 PR tree-optimization/83867
2640 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2641 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2643 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2646 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2647 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2648 If off is not INTEGER_CST, issue a may not be aligned warning
2649 rather than isn't aligned. Use isn%'t rather than isn't.
2650 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2652 <case MULT_EXPR>: Improve the case when bottom and one of the
2653 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2654 operand, in that case check if the other operand is multiple of
2655 bottom divided by the INTEGER_CST operand.
2657 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2660 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2661 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2662 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2663 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2664 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2665 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2666 (pa_function_arg_size): New function.
2668 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2670 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2671 in a separate statement.
2673 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2675 PR tree-optimization/83847
2676 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2677 group gathers and scatters.
2679 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2681 PR rtl-optimization/86620
2682 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2684 PR rtl-optimization/83213
2685 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2686 to last if both are JUMP_INSNs.
2688 PR tree-optimization/83843
2689 * gimple-ssa-store-merging.c
2690 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2691 store_immediate_info for bswap/nop orig_stores.
2693 2018-01-15 Andrew Waterman <andrew@sifive.com>
2695 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2697 <UDIV>: Increase cost if !TARGET_DIV.
2699 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2701 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2702 (define_attr "cr_logical_3op"): New.
2703 (cceq_ior_compare): Adjust.
2704 (cceq_ior_compare_complement): Adjust.
2705 (*cceq_rev_compare): Adjust.
2706 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2707 (is_cracked_insn): Adjust.
2708 (insn_must_be_first_in_group): Adjust.
2709 * config/rs6000/40x.md: Adjust.
2710 * config/rs6000/440.md: Adjust.
2711 * config/rs6000/476.md: Adjust.
2712 * config/rs6000/601.md: Adjust.
2713 * config/rs6000/603.md: Adjust.
2714 * config/rs6000/6xx.md: Adjust.
2715 * config/rs6000/7450.md: Adjust.
2716 * config/rs6000/7xx.md: Adjust.
2717 * config/rs6000/8540.md: Adjust.
2718 * config/rs6000/cell.md: Adjust.
2719 * config/rs6000/e300c2c3.md: Adjust.
2720 * config/rs6000/e500mc.md: Adjust.
2721 * config/rs6000/e500mc64.md: Adjust.
2722 * config/rs6000/e5500.md: Adjust.
2723 * config/rs6000/e6500.md: Adjust.
2724 * config/rs6000/mpc.md: Adjust.
2725 * config/rs6000/power4.md: Adjust.
2726 * config/rs6000/power5.md: Adjust.
2727 * config/rs6000/power6.md: Adjust.
2728 * config/rs6000/power7.md: Adjust.
2729 * config/rs6000/power8.md: Adjust.
2730 * config/rs6000/power9.md: Adjust.
2731 * config/rs6000/rs64.md: Adjust.
2732 * config/rs6000/titan.md: Adjust.
2734 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2736 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2737 ix86_indirect_branch_register logic.
2739 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2741 * config/i386/constraints.md (Bs): Update
2742 ix86_indirect_branch_register check. Don't check
2743 ix86_indirect_branch_register with GOT_memory_operand.
2745 * config/i386/predicates.md (GOT_memory_operand): Don't check
2746 ix86_indirect_branch_register here.
2747 (GOT32_symbol_operand): Likewise.
2749 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2751 * config/i386/predicates.md (constant_call_address_operand):
2752 Rewrite ix86_indirect_branch_register logic.
2753 (sibcall_insn_operand): Likewise.
2755 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2757 * config/i386/constraints.md (Bs): Replace
2758 ix86_indirect_branch_thunk_register with
2759 ix86_indirect_branch_register.
2761 * config/i386/i386.md (indirect_jump): Likewise.
2762 (tablejump): Likewise.
2763 (*sibcall_memory): Likewise.
2764 (*sibcall_value_memory): Likewise.
2765 Peepholes of indirect call and jump via memory: Likewise.
2766 * config/i386/i386.opt: Likewise.
2767 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2768 (GOT_memory_operand): Likewise.
2769 (call_insn_operand): Likewise.
2770 (sibcall_insn_operand): Likewise.
2771 (GOT32_symbol_operand): Likewise.
2773 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2776 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2777 type rather than type addr's type points to.
2778 (expand_omp_atomic_mutex): Likewise.
2779 (expand_omp_atomic): Likewise.
2781 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2784 * config/i386/i386.c (output_indirect_thunk_function): Use
2785 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2786 for __x86_return_thunk.
2788 2018-01-15 Richard Biener <rguenther@suse.de>
2791 * expmed.c (extract_bit_field_1): Fix typo.
2793 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2796 * config/arm/iterators.md (VF): New mode iterator.
2797 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2798 Remove integer-related logic from pattern.
2799 (neon_vabd<mode>_3): Likewise.
2801 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2804 * common.opt (fstrict-overflow): No longer an alias.
2805 (fwrapv-pointer): New option.
2806 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2807 also for pointer types based on flag_wrapv_pointer.
2808 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2809 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2810 opts->x_flag_wrapv got set.
2811 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2812 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2813 POINTER_TYPE_OVERFLOW_UNDEFINED.
2814 * match.pd: Likewise in address comparison pattern.
2815 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2817 2018-01-15 Richard Biener <rguenther@suse.de>
2820 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2821 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2822 Reset type names to their identifier if their TYPE_DECL doesn't
2823 have linkage (and thus is used for ODR and devirt).
2824 (save_debug_info_for_decl): Remove.
2825 (save_debug_info_for_type): Likewise.
2826 (add_tree_to_fld_list): Adjust.
2827 * tree-pretty-print.c (dump_generic_node): Make dumping of
2828 type names more robust.
2830 2018-01-15 Richard Biener <rguenther@suse.de>
2832 * BASE-VER: Bump to 8.0.1.
2834 2018-01-14 Martin Sebor <msebor@redhat.com>
2837 * builtins.c (check_access): Avoid warning when the no-warning bit
2840 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2842 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2843 * ira-color (allocno_hard_regs_compare): Likewise.
2845 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2848 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2849 Use .pushsection/.popsection.
2851 2018-01-14 Martin Sebor <msebor@redhat.com>
2854 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2856 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2858 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2859 entry from extra_headers.
2860 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2861 extra_headers, make the list bitwise identical to the i?86-*-* one.
2863 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2865 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2866 -mcmodel=large with -mindirect-branch=thunk,
2867 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2868 -mfunction-return=thunk-extern.
2869 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2870 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2871 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2873 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2875 * config/i386/i386.c (print_reg): Print the name of the full
2876 integer register without '%'.
2877 (ix86_print_operand): Handle 'V'.
2878 * doc/extend.texi: Document 'V' modifier.
2880 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2882 * config/i386/constraints.md (Bs): Disallow memory operand for
2883 -mindirect-branch-register.
2885 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2886 (GOT_memory_operand): Likewise.
2887 (call_insn_operand): Likewise.
2888 (sibcall_insn_operand): Likewise.
2889 (GOT32_symbol_operand): Likewise.
2890 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2891 for -mindirect-branch-register.
2892 (tablejump): Likewise.
2893 (*sibcall_memory): Likewise.
2894 (*sibcall_value_memory): Likewise.
2895 Disallow peepholes of indirect call and jump via memory for
2896 -mindirect-branch-register.
2897 (*call_pop): Replace m with Bw.
2898 (*call_value_pop): Likewise.
2899 (*sibcall_pop_memory): Replace m with Bs.
2900 * config/i386/i386.opt (mindirect-branch-register): New option.
2901 * doc/invoke.texi: Document -mindirect-branch-register option.
2903 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2905 * config/i386/i386-protos.h (ix86_output_function_return): New.
2906 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2907 set function_return_type.
2908 (indirect_thunk_name): Add ret_p to indicate thunk for function
2910 (output_indirect_thunk_function): Pass false to
2911 indirect_thunk_name.
2912 (ix86_output_indirect_branch_via_reg): Likewise.
2913 (ix86_output_indirect_branch_via_push): Likewise.
2914 (output_indirect_thunk_function): Create alias for function
2915 return thunk if regno < 0.
2916 (ix86_output_function_return): New function.
2917 (ix86_handle_fndecl_attribute): Handle function_return.
2918 (ix86_attribute_table): Add function_return.
2919 * config/i386/i386.h (machine_function): Add
2920 function_return_type.
2921 * config/i386/i386.md (simple_return_internal): Use
2922 ix86_output_function_return.
2923 (simple_return_internal_long): Likewise.
2924 * config/i386/i386.opt (mfunction-return=): New option.
2925 (indirect_branch): Mention -mfunction-return=.
2926 * doc/extend.texi: Document function_return function attribute.
2927 * doc/invoke.texi: Document -mfunction-return= option.
2929 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2931 * config/i386/i386-opts.h (indirect_branch): New.
2932 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2933 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2934 with local indirect jump when converting indirect call and jump.
2935 (ix86_set_indirect_branch_type): New.
2936 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2937 (indirectlabelno): New.
2938 (indirect_thunk_needed): Likewise.
2939 (indirect_thunk_bnd_needed): Likewise.
2940 (indirect_thunks_used): Likewise.
2941 (indirect_thunks_bnd_used): Likewise.
2942 (INDIRECT_LABEL): Likewise.
2943 (indirect_thunk_name): Likewise.
2944 (output_indirect_thunk): Likewise.
2945 (output_indirect_thunk_function): Likewise.
2946 (ix86_output_indirect_branch_via_reg): Likewise.
2947 (ix86_output_indirect_branch_via_push): Likewise.
2948 (ix86_output_indirect_branch): Likewise.
2949 (ix86_output_indirect_jmp): Likewise.
2950 (ix86_code_end): Call output_indirect_thunk_function if needed.
2951 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2953 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2954 (ix86_attribute_table): Add indirect_branch.
2955 * config/i386/i386.h (machine_function): Add indirect_branch_type
2956 and has_local_indirect_jump.
2957 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2959 (tablejump): Likewise.
2960 (*indirect_jump): Use ix86_output_indirect_jmp.
2961 (*tablejump_1): Likewise.
2962 (simple_return_indirect_internal): Likewise.
2963 * config/i386/i386.opt (mindirect-branch=): New option.
2964 (indirect_branch): New.
2967 (thunk-inline): Likewise.
2968 (thunk-extern): Likewise.
2969 * doc/extend.texi: Document indirect_branch function attribute.
2970 * doc/invoke.texi: Document -mindirect-branch= option.
2972 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2975 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2977 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2979 * ipa-inline.c (want_inline_small_function_p): Return false if
2980 inlining has already failed with CIF_FINAL_ERROR.
2981 (update_caller_keys): Call want_inline_small_function_p before
2983 (update_callee_keys): Likewise.
2985 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2987 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2989 (rs6000_quadword_masked_address_p): Likewise.
2990 (quad_aligned_load_p): Likewise.
2991 (quad_aligned_store_p): Likewise.
2992 (const_load_sequence_p): Add comment to describe the outer-most loop.
2993 (mimic_memory_attributes_and_flags): New function.
2994 (rs6000_gen_stvx): Likewise.
2995 (replace_swapped_aligned_store): Likewise.
2996 (rs6000_gen_lvx): Likewise.
2997 (replace_swapped_aligned_load): Likewise.
2998 (replace_swapped_load_constant): Capitalize argument name in
2999 comment describing this function.
3000 (rs6000_analyze_swaps): Add a third pass to search for vector loads
3001 and stores that access quad-word aligned addresses and replace
3002 with stvx or lvx instructions when appropriate.
3003 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
3004 New function prototype.
3005 (rs6000_quadword_masked_address_p): Likewise.
3006 (rs6000_gen_lvx): Likewise.
3007 (rs6000_gen_stvx): Likewise.
3008 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
3009 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3010 when memory address is aligned.
3011 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3012 this split to select lvx instruction when memory address is aligned.
3013 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3014 instruction when memory address is aligned.
3015 (*vsx_le_perm_load_v16qi): Likewise.
3016 (four unnamed splitters): Modify to select the stvx instruction
3017 when memory is aligned.
3019 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3021 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3022 which appears in the queue multiple times.
3024 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3025 Alan Hayward <alan.hayward@arm.com>
3026 David Sherwood <david.sherwood@arm.com>
3028 * tree-vectorizer.h (vec_lower_bound): New structure.
3029 (_loop_vec_info): Add check_nonzero and lower_bounds.
3030 (LOOP_VINFO_CHECK_NONZERO): New macro.
3031 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3032 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3033 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3034 fields. Make seg_len the distance travelled, not including the
3036 (dr_direction_indicator): Declare.
3037 (dr_zero_step_indicator): Likewise.
3038 (dr_known_forward_stride_p): Likewise.
3039 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3041 (runtime_alias_check_p): Allow runtime alias checks with
3043 (operator ==): Compare access_size and align.
3044 (prune_runtime_alias_test_list): Rework for new distinction between
3045 the access_size and seg_len.
3046 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3048 (get_segment_min_max): New function.
3049 (create_intersect_range_checks): Use it.
3050 (dr_step_indicator): New function.
3051 (dr_direction_indicator): Likewise.
3052 (dr_zero_step_indicator): Likewise.
3053 (dr_known_forward_stride_p): Likewise.
3054 * tree-loop-distribution.c (data_ref_segment_size): Return
3055 DR_STEP * (niters - 1).
3056 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3058 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3059 (vect_preserves_scalar_order_p): New function, split out from...
3060 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3061 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3062 (vect_vfa_access_size): New function.
3063 (vect_vfa_align): Likewise.
3064 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3065 (dump_lower_bound): New function.
3066 (vect_check_lower_bound): Likewise.
3067 (vect_small_gap_p): Likewise.
3068 (vectorizable_with_step_bound_p): Likewise.
3069 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3070 depencies if the vectorization factor is 1. Convert the checks
3071 for nonzero steps into checks on the bounds of DR_STEP. Try using
3072 a bunds check for variable steps if the minimum required step is
3073 relatively small. Update calls to the dr_with_seg_len
3074 constructor and to vect_compile_time_alias.
3075 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3077 (vect_loop_versioning): Call it.
3078 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3080 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3082 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3083 Alan Hayward <alan.hayward@arm.com>
3084 David Sherwood <david.sherwood@arm.com>
3086 * doc/sourcebuild.texi (vect_scatter_store): Document.
3087 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3089 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3091 * genopinit.c (main): Add supports_vec_scatter_store and
3092 supports_vec_scatter_store_cached to target_optabs.
3093 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3094 IFN_MASK_SCATTER_STORE.
3095 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3097 * internal-fn.h (internal_store_fn_p): Declare.
3098 (internal_fn_stored_value_index): Likewise.
3099 * internal-fn.c (scatter_store_direct): New macro.
3100 (expand_scatter_store_optab_fn): New function.
3101 (direct_scatter_store_optab_supported_p): New macro.
3102 (internal_store_fn_p): New function.
3103 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3104 IFN_MASK_SCATTER_STORE.
3105 (internal_fn_mask_index): Likewise.
3106 (internal_fn_stored_value_index): New function.
3107 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3109 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3110 * optabs-query.c (supports_vec_scatter_store_p): New function.
3111 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3112 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3113 true for scatter stores.
3114 (vect_gather_scatter_fn_p): Handle scatter stores too.
3115 (vect_check_gather_scatter): Consider using scatter stores if
3116 supports_vec_scatter_store_p.
3117 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3119 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3120 internal_fn_stored_value_index.
3121 (check_load_store_masking): Handle scatter stores too.
3122 (vect_get_store_rhs): Make public.
3123 (vectorizable_call): Use internal_store_fn_p.
3124 (vectorizable_store): Handle scatter store internal functions.
3125 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3126 when deciding whether the end of the group has been reached.
3127 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3128 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3129 (mask_scatter_store<mode>): New insns.
3131 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3132 Alan Hayward <alan.hayward@arm.com>
3133 David Sherwood <david.sherwood@arm.com>
3135 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3136 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3137 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3139 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3140 Use vect_truncate_gather_scatter_offset if we can't treat the
3141 operation as a normal gather load or scatter store.
3142 (get_group_load_store_type): Take the gather_scatter_info
3143 as argument. Try using a gather load or scatter store for
3144 single-element groups.
3145 (get_load_store_type): Update calls to get_group_load_store_type
3146 and vect_use_strided_gather_scatters_p.
3148 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3149 Alan Hayward <alan.hayward@arm.com>
3150 David Sherwood <david.sherwood@arm.com>
3152 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3153 optional tree argument.
3154 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3156 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3157 but continue to use the current value as a fallback.
3158 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3159 to compare the updates.
3160 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3161 (get_load_store_type): Use it when handling a strided access.
3162 (vect_get_strided_load_store_ops): New function.
3163 (vect_get_data_ptr_increment): Likewise.
3164 (vectorizable_load): Handle strided gather loads. Always pass
3165 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3167 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3168 Alan Hayward <alan.hayward@arm.com>
3169 David Sherwood <david.sherwood@arm.com>
3171 * doc/md.texi (gather_load@var{m}): Document.
3172 (mask_gather_load@var{m}): Likewise.
3173 * genopinit.c (main): Add supports_vec_gather_load and
3174 supports_vec_gather_load_cached to target_optabs.
3175 * optabs-tree.c (init_tree_optimization_optabs): Use
3176 ggc_cleared_alloc to allocate target_optabs.
3177 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3178 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3180 * internal-fn.h (internal_load_fn_p): Declare.
3181 (internal_gather_scatter_fn_p): Likewise.
3182 (internal_fn_mask_index): Likewise.
3183 (internal_gather_scatter_fn_supported_p): Likewise.
3184 * internal-fn.c (gather_load_direct): New macro.
3185 (expand_gather_load_optab_fn): New function.
3186 (direct_gather_load_optab_supported_p): New macro.
3187 (direct_internal_fn_optab): New function.
3188 (internal_load_fn_p): Likewise.
3189 (internal_gather_scatter_fn_p): Likewise.
3190 (internal_fn_mask_index): Likewise.
3191 (internal_gather_scatter_fn_supported_p): Likewise.
3192 * optabs-query.c (supports_at_least_one_mode_p): New function.
3193 (supports_vec_gather_load_p): Likewise.
3194 * optabs-query.h (supports_vec_gather_load_p): Declare.
3195 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3196 and memory_type field.
3197 (NUM_PATTERNS): Bump to 15.
3198 * tree-vect-data-refs.c: Include internal-fn.h.
3199 (vect_gather_scatter_fn_p): New function.
3200 (vect_describe_gather_scatter_call): Likewise.
3201 (vect_check_gather_scatter): Try using internal functions for
3202 gather loads. Recognize existing calls to a gather load function.
3203 (vect_analyze_data_refs): Consider using gather loads if
3204 supports_vec_gather_load_p.
3205 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3206 (vect_get_gather_scatter_offset_type): Likewise.
3207 (vect_convert_mask_for_vectype): Likewise.
3208 (vect_add_conversion_to_patterm): Likewise.
3209 (vect_try_gather_scatter_pattern): Likewise.
3210 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3211 (vect_vect_recog_func_ptrs): Add it.
3212 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3213 internal_fn_mask_index and internal_gather_scatter_fn_p.
3214 (check_load_store_masking): Take the gather_scatter_info as an
3215 argument and handle gather loads.
3216 (vect_get_gather_scatter_ops): New function.
3217 (vectorizable_call): Check internal_load_fn_p.
3218 (vectorizable_load): Likewise. Handle gather load internal
3220 (vectorizable_store): Update call to check_load_store_masking.
3221 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3222 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3223 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3224 (aarch64_gather_scale_operand_d): New predicates.
3225 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3226 (mask_gather_load<mode>): New insns.
3228 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3229 Alan Hayward <alan.hayward@arm.com>
3230 David Sherwood <david.sherwood@arm.com>
3232 * optabs.def (fold_left_plus_optab): New optab.
3233 * doc/md.texi (fold_left_plus_@var{m}): Document.
3234 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3235 * internal-fn.c (fold_left_direct): Define.
3236 (expand_fold_left_optab_fn): Likewise.
3237 (direct_fold_left_optab_supported_p): Likewise.
3238 * fold-const-call.c (fold_const_fold_left): New function.
3239 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3240 * tree-parloops.c (valid_reduction_p): New function.
3241 (gather_scalar_reductions): Use it.
3242 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3243 (vect_finish_replace_stmt): Declare.
3244 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3245 (needs_fold_left_reduction_p): New function, split out from...
3246 (vect_is_simple_reduction): ...here. Accept reductions that
3247 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3248 (vect_force_simple_reduction): Also store the reduction type in
3249 the assignment's STMT_VINFO_REDUC_TYPE.
3250 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3251 (merge_with_identity): New function.
3252 (vect_expand_fold_left): Likewise.
3253 (vectorize_fold_left_reduction): Likewise.
3254 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3255 scalar phi in place for it. Check for target support and reject
3256 cases that would reassociate the operation. Defer the transform
3257 phase to vectorize_fold_left_reduction.
3258 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3259 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3260 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3262 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3264 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3265 call to ifc_temp_var.
3267 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3268 Alan Hayward <alan.hayward@arm.com>
3269 David Sherwood <david.sherwood@arm.com>
3271 * target.def (legitimize_address_displacement): Take the original
3272 offset as a poly_int.
3273 * targhooks.h (default_legitimize_address_displacement): Update
3275 * targhooks.c (default_legitimize_address_displacement): Likewise.
3276 * doc/tm.texi: Regenerate.
3277 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3278 as an argument, moving assert of ad->disp == ad->disp_term to...
3279 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3280 Try calling targetm.legitimize_address_displacement before expanding
3281 the address rather than afterwards, and adjust for the new interface.
3282 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3283 Match the new hook interface. Handle SVE addresses.
3284 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3287 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3289 * Makefile.in (OBJS): Add early-remat.o.
3290 * target.def (select_early_remat_modes): New hook.
3291 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3292 * doc/tm.texi: Regenerate.
3293 * targhooks.h (default_select_early_remat_modes): Declare.
3294 * targhooks.c (default_select_early_remat_modes): New function.
3295 * timevar.def (TV_EARLY_REMAT): New timevar.
3296 * passes.def (pass_early_remat): New pass.
3297 * tree-pass.h (make_pass_early_remat): Declare.
3298 * early-remat.c: New file.
3299 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3301 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3303 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3304 Alan Hayward <alan.hayward@arm.com>
3305 David Sherwood <david.sherwood@arm.com>
3307 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3308 vfm1 with a bound_epilog parameter.
3309 (vect_do_peeling): Update calls accordingly, and move the prologue
3310 call earlier in the function. Treat the base bound_epilog as 0 for
3311 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3312 this base when peeling for gaps.
3313 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3314 with fully-masked loops.
3315 (vect_estimate_min_profitable_iters): Handle the single peeled
3316 iteration in that case.
3318 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3319 Alan Hayward <alan.hayward@arm.com>
3320 David Sherwood <david.sherwood@arm.com>
3322 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3323 single-element interleaving even if the size is not a power of 2.
3324 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3325 accesses for single-element interleaving if the group size is
3328 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3329 Alan Hayward <alan.hayward@arm.com>
3330 David Sherwood <david.sherwood@arm.com>
3332 * doc/md.texi (fold_extract_last_@var{m}): Document.
3333 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3334 * optabs.def (fold_extract_last_optab): New optab.
3335 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3336 * internal-fn.c (fold_extract_direct): New macro.
3337 (expand_fold_extract_optab_fn): Likewise.
3338 (direct_fold_extract_optab_supported_p): Likewise.
3339 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3340 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3341 EXTRACT_LAST_REDUCTION.
3342 (get_initial_def_for_reduction): Do not create an initial vector
3343 for EXTRACT_LAST_REDUCTION reductions.
3344 (vectorizable_reduction): Leave the scalar phi in place for
3345 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3346 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3347 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3348 transform phase to vectorizable_condition.
3349 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3351 (vect_finish_stmt_generation): ...here.
3352 (vect_finish_replace_stmt): New function.
3353 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3354 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3356 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3358 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3359 Alan Hayward <alan.hayward@arm.com>
3360 David Sherwood <david.sherwood@arm.com>
3362 * doc/md.texi (extract_last_@var{m}): Document.
3363 * optabs.def (extract_last_optab): New optab.
3364 * internal-fn.def (EXTRACT_LAST): New internal function.
3365 * internal-fn.c (cond_unary_direct): New macro.
3366 (expand_cond_unary_optab_fn): Likewise.
3367 (direct_cond_unary_optab_supported_p): Likewise.
3368 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3369 loops using EXTRACT_LAST.
3370 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3371 (extract_last_<mode>): ...this optab.
3372 (vec_extract<mode><Vel>): Update accordingly.
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 * target.def (empty_mask_is_expensive): New hook.
3379 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3380 * doc/tm.texi: Regenerate.
3381 * targhooks.h (default_empty_mask_is_expensive): Declare.
3382 * targhooks.c (default_empty_mask_is_expensive): New function.
3383 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3384 if the target says that empty masks are expensive.
3385 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3387 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3389 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3390 Alan Hayward <alan.hayward@arm.com>
3391 David Sherwood <david.sherwood@arm.com>
3393 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3394 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3395 (vect_use_loop_mask_for_alignment_p): New function.
3396 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3397 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3398 niters_skip argument. Make sure that the first niters_skip elements
3399 of the first iteration are inactive.
3400 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3401 Update call to vect_set_loop_masks_directly.
3402 (get_misalign_in_elems): New function, split out from...
3403 (vect_gen_prolog_loop_niters): ...here.
3404 (vect_update_init_of_dr): Take a code argument that specifies whether
3405 the adjustment should be added or subtracted.
3406 (vect_update_init_of_drs): Likewise.
3407 (vect_prepare_for_masked_peels): New function.
3408 (vect_do_peeling): Skip prologue peeling if we're using a mask
3409 instead. Update call to vect_update_inits_of_drs.
3410 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3412 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3413 alignment. Do not include the number of peeled iterations in
3414 the minimum threshold in that case.
3415 (vectorizable_induction): Adjust the start value down by
3416 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3417 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3418 Take the number of skipped iterations into account when calculating
3420 * tree-vect-stmts.c (vect_gen_while_not): New function.
3422 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3423 Alan Hayward <alan.hayward@arm.com>
3424 David Sherwood <david.sherwood@arm.com>
3426 * doc/sourcebuild.texi (vect_fully_masked): Document.
3427 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3429 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3431 (vect_analyze_loop_2): ...here. Don't check the vectorization
3432 factor against the number of loop iterations if the loop is
3435 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3436 Alan Hayward <alan.hayward@arm.com>
3437 David Sherwood <david.sherwood@arm.com>
3439 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3440 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3441 (dump_groups): Update accordingly.
3442 (iv_use::mem_type): New member variable.
3443 (address_p): New function.
3444 (record_use): Add a mem_type argument and initialize the new
3446 (record_group_use): Add a mem_type argument. Use address_p.
3447 Remove obsolete null checks of base_object. Update call to record_use.
3448 (find_interesting_uses_op): Update call to record_group_use.
3449 (find_interesting_uses_cond): Likewise.
3450 (find_interesting_uses_address): Likewise.
3451 (get_mem_type_for_internal_fn): New function.
3452 (find_address_like_use): Likewise.
3453 (find_interesting_uses_stmt): Try find_address_like_use before
3454 calling find_interesting_uses_op.
3455 (addr_offset_valid_p): Use the iv mem_type field as the type
3456 of the addressed memory.
3457 (add_autoinc_candidates): Likewise.
3458 (get_address_cost): Likewise.
3459 (split_small_address_groups_p): Use address_p.
3460 (split_address_groups): Likewise.
3461 (add_iv_candidate_for_use): Likewise.
3462 (autoinc_possible_for_pair): Likewise.
3463 (rewrite_groups): Likewise.
3464 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3465 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3466 (get_alias_ptr_type_for_ptr_address): New function.
3467 (rewrite_use_address): Rewrite address uses in calls that were
3468 identified by find_address_like_use.
3470 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3471 Alan Hayward <alan.hayward@arm.com>
3472 David Sherwood <david.sherwood@arm.com>
3474 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3476 * gimple-expr.h (is_gimple_addressable: Likewise.
3477 * gimple-expr.c (is_gimple_address): Likewise.
3478 * internal-fn.c (expand_call_mem_ref): New function.
3479 (expand_mask_load_optab_fn): Use it.
3480 (expand_mask_store_optab_fn): Likewise.
3482 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3483 Alan Hayward <alan.hayward@arm.com>
3484 David Sherwood <david.sherwood@arm.com>
3486 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3487 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3488 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3489 (cond_umax@var{mode}): Document.
3490 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3491 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3492 (cond_umin_optab, cond_umax_optab): New optabs.
3493 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3494 (COND_IOR, COND_XOR): New internal functions.
3495 * internal-fn.h (get_conditional_internal_fn): Declare.
3496 * internal-fn.c (cond_binary_direct): New macro.
3497 (expand_cond_binary_optab_fn): Likewise.
3498 (direct_cond_binary_optab_supported_p): Likewise.
3499 (get_conditional_internal_fn): New function.
3500 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3501 Cope with reduction statements that are vectorized as calls rather
3503 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3504 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3505 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3506 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3507 (UNSPEC_COND_EOR): New unspecs.
3508 (optab): Add mappings for them.
3509 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3510 (sve_int_op, sve_fp_op): New int attributes.
3512 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3513 Alan Hayward <alan.hayward@arm.com>
3514 David Sherwood <david.sherwood@arm.com>
3516 * optabs.def (while_ult_optab): New optab.
3517 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3518 * internal-fn.def (WHILE_ULT): New internal function.
3519 * internal-fn.h (direct_internal_fn_supported_p): New override
3520 that takes two types as argument.
3521 * internal-fn.c (while_direct): New macro.
3522 (expand_while_optab_fn): New function.
3523 (convert_optab_supported_p): Likewise.
3524 (direct_while_optab_supported_p): New macro.
3525 * wide-int.h (wi::udiv_ceil): New function.
3526 * tree-vectorizer.h (rgroup_masks): New structure.
3527 (vec_loop_masks): New typedef.
3528 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3530 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3531 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3532 (vect_max_vf): New function.
3533 (slpeel_make_loop_iterate_ntimes): Delete.
3534 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3535 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3536 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3537 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3538 internal-fn.h, stor-layout.h and optabs-query.h.
3539 (vect_set_loop_mask): New function.
3540 (add_preheader_seq): Likewise.
3541 (add_header_seq): Likewise.
3542 (interleave_supported_p): Likewise.
3543 (vect_maybe_permute_loop_masks): Likewise.
3544 (vect_set_loop_masks_directly): Likewise.
3545 (vect_set_loop_condition_masked): Likewise.
3546 (vect_set_loop_condition_unmasked): New function, split out from
3547 slpeel_make_loop_iterate_ntimes.
3548 (slpeel_make_loop_iterate_ntimes): Rename to..
3549 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3550 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3551 (vect_do_peeling): Update call accordingly.
3552 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3554 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3555 mask_compare_type, can_fully_mask_p and fully_masked_p.
3556 (release_vec_loop_masks): New function.
3557 (_loop_vec_info): Use it to free the loop masks.
3558 (can_produce_all_loop_masks_p): New function.
3559 (vect_get_max_nscalars_per_iter): Likewise.
3560 (vect_verify_full_masking): Likewise.
3561 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3562 retries, and free the mask rgroups before retrying. Check loop-wide
3563 reasons for disallowing fully-masked loops. Make the final decision
3564 about whether use a fully-masked loop or not.
3565 (vect_estimate_min_profitable_iters): Do not assume that peeling
3566 for the number of iterations will be needed for fully-masked loops.
3567 (vectorizable_reduction): Disable fully-masked loops.
3568 (vectorizable_live_operation): Likewise.
3569 (vect_halve_mask_nunits): New function.
3570 (vect_double_mask_nunits): Likewise.
3571 (vect_record_loop_mask): Likewise.
3572 (vect_get_loop_mask): Likewise.
3573 (vect_transform_loop): Handle the case in which the final loop
3574 iteration might handle a partial vector. Call vect_set_loop_condition
3575 instead of slpeel_make_loop_iterate_ntimes.
3576 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3577 (check_load_store_masking): New function.
3578 (prepare_load_store_mask): Likewise.
3579 (vectorizable_store): Handle fully-masked loops.
3580 (vectorizable_load): Likewise.
3581 (supportable_widening_operation): Use vect_halve_mask_nunits for
3583 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3584 (vect_gen_while): New function.
3585 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3586 (aarch64_uqdec<mode>): New insn.
3588 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3589 Alan Hayward <alan.hayward@arm.com>
3590 David Sherwood <david.sherwood@arm.com>
3592 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3593 (reduc_xor_scal_optab): New optabs.
3594 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3595 (reduc_xor_scal_@var{m}): Document.
3596 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3597 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3599 * fold-const-call.c (fold_const_call): Handle them.
3600 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3601 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3602 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3603 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3604 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3605 (UNSPEC_XORV): New unspecs.
3606 (optab): Add entries for them.
3607 (BITWISEV): New int iterator.
3608 (bit_reduc_op): New int attributes.
3610 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3611 Alan Hayward <alan.hayward@arm.com>
3612 David Sherwood <david.sherwood@arm.com>
3614 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3615 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3616 * optabs.def (vec_shl_insert_optab): New optab.
3617 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3618 (duplicate_and_interleave): Likewise.
3619 * tree-vect-loop.c: Include internal-fn.h.
3620 (neutral_op_for_slp_reduction): New function, split out from
3621 get_initial_defs_for_reduction.
3622 (get_initial_def_for_reduction): Handle option 2 for variable-length
3623 vectors by loading the neutral value into a vector and then shifting
3624 the initial value into element 0.
3625 (get_initial_defs_for_reduction): Replace the code argument with
3626 the neutral value calculated by neutral_op_for_slp_reduction.
3627 Use gimple_build_vector for constant-length vectors.
3628 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3629 but the first group_size elements have a neutral value.
3630 Use duplicate_and_interleave otherwise.
3631 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3632 Update call to get_initial_defs_for_reduction. Handle SLP
3633 reductions for variable-length vectors by creating one vector
3634 result for each scalar result, with the elements associated
3635 with other scalar results stubbed out with the neutral value.
3636 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3637 Require IFN_VEC_SHL_INSERT for double reductions on
3638 variable-length vectors, or SLP reductions that have
3639 a neutral value. Require can_duplicate_and_interleave_p
3640 support for variable-length unchained SLP reductions if there
3641 is no neutral value, such as for MIN/MAX reductions. Also require
3642 the number of vector elements to be a multiple of the number of
3643 SLP statements when doing variable-length unchained SLP reductions.
3644 Update call to vect_create_epilog_for_reduction.
3645 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3646 and remove initial values.
3647 (duplicate_and_interleave): Make public.
3648 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3649 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3651 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3652 Alan Hayward <alan.hayward@arm.com>
3653 David Sherwood <david.sherwood@arm.com>
3655 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3656 (can_duplicate_and_interleave_p): New function.
3657 (vect_get_and_check_slp_defs): Take the vector of statements
3658 rather than just the current one. Remove excess parentheses.
3659 Restriction rejectinon of vect_constant_def and vect_external_def
3660 for variable-length vectors to boolean types, or types for which
3661 can_duplicate_and_interleave_p is false.
3662 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3663 (duplicate_and_interleave): New function.
3664 (vect_get_constant_vectors): Use gimple_build_vector for
3665 constant-length vectors and suitable variable-length constant
3666 vectors. Use duplicate_and_interleave for other variable-length
3667 vectors. Don't defer the update when inserting new statements.
3669 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3670 Alan Hayward <alan.hayward@arm.com>
3671 David Sherwood <david.sherwood@arm.com>
3673 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3674 min_profitable_iters doesn't go negative.
3676 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3677 Alan Hayward <alan.hayward@arm.com>
3678 David Sherwood <david.sherwood@arm.com>
3680 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3681 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3682 * optabs.def (vec_mask_load_lanes_optab): New optab.
3683 (vec_mask_store_lanes_optab): Likewise.
3684 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3685 (MASK_STORE_LANES): Likewise.
3686 * internal-fn.c (mask_load_lanes_direct): New macro.
3687 (mask_store_lanes_direct): Likewise.
3688 (expand_mask_load_optab_fn): Handle masked operations.
3689 (expand_mask_load_lanes_optab_fn): New macro.
3690 (expand_mask_store_optab_fn): Handle masked operations.
3691 (expand_mask_store_lanes_optab_fn): New macro.
3692 (direct_mask_load_lanes_optab_supported_p): Likewise.
3693 (direct_mask_store_lanes_optab_supported_p): Likewise.
3694 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3696 (vect_load_lanes_supported): Likewise.
3697 * tree-vect-data-refs.c (strip_conversion): New function.
3698 (can_group_stmts_p): Likewise.
3699 (vect_analyze_data_ref_accesses): Use it instead of checking
3700 for a pair of assignments.
3701 (vect_store_lanes_supported): Take a masked_p parameter.
3702 (vect_load_lanes_supported): Likewise.
3703 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3704 vect_store_lanes_supported and vect_load_lanes_supported.
3705 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3706 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3707 parameter. Don't allow gaps for masked accesses.
3708 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3709 and vect_load_lanes_supported.
3710 (get_load_store_type): Take a masked_p parameter and update
3711 call to get_group_load_store_type.
3712 (vectorizable_store): Update call to get_load_store_type.
3713 Handle IFN_MASK_STORE_LANES.
3714 (vectorizable_load): Update call to get_load_store_type.
3715 Handle IFN_MASK_LOAD_LANES.
3717 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3718 Alan Hayward <alan.hayward@arm.com>
3719 David Sherwood <david.sherwood@arm.com>
3721 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3723 * config/aarch64/aarch64-protos.h
3724 (aarch64_sve_struct_memory_operand_p): Declare.
3725 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3726 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3727 (VPRED, vpred): Handle SVE structure modes.
3728 * config/aarch64/constraints.md (Utx): New constraint.
3729 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3730 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3731 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3732 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3733 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3734 structure modes. Split into pieces after RA.
3735 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3736 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3738 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3739 SVE structure modes.
3740 (aarch64_classify_address): Likewise.
3741 (sizetochar): Move earlier in file.
3742 (aarch64_print_operand): Handle SVE register lists.
3743 (aarch64_array_mode): New function.
3744 (aarch64_sve_struct_memory_operand_p): Likewise.
3745 (TARGET_ARRAY_MODE): Redefine.
3747 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3748 Alan Hayward <alan.hayward@arm.com>
3749 David Sherwood <david.sherwood@arm.com>
3751 * target.def (array_mode): New target hook.
3752 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3753 * doc/tm.texi: Regenerate.
3754 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3755 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3756 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3758 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3761 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3762 Alan Hayward <alan.hayward@arm.com>
3763 David Sherwood <david.sherwood@arm.com>
3765 * fold-const.c (fold_binary_loc): Check the argument types
3766 rather than the result type when testing for a vector operation.
3768 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3770 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3771 * doc/tm.texi: Regenerate.
3773 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3774 Alan Hayward <alan.hayward@arm.com>
3775 David Sherwood <david.sherwood@arm.com>
3777 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3778 (sve): Document new AArch64 extension.
3779 * doc/md.texi (w): Extend the description of the AArch64
3780 constraint to include SVE vectors.
3781 (Upl, Upa): Document new AArch64 predicate constraints.
3782 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3784 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3785 (msve-vector-bits=): New option.
3786 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3787 SVE when these are disabled.
3788 (sve): New extension.
3789 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3790 modes. Adjust their number of units based on aarch64_sve_vg.
3791 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3792 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3793 aarch64_addr_query_type.
3794 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3795 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3796 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3797 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3798 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3799 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3800 (aarch64_simd_imm_zero_p): Delete.
3801 (aarch64_check_zero_based_sve_index_immediate): Declare.
3802 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3803 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3804 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3805 (aarch64_sve_float_mul_immediate_p): Likewise.
3806 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3808 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3809 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3810 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3811 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3812 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3813 (aarch64_regmode_natural_size): Likewise.
3814 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3815 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3817 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3818 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3819 for VG and the SVE predicate registers.
3820 (V_ALIASES): Add a "z"-prefixed alias.
3821 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3822 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3823 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3824 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3825 (REG_CLASS_NAMES): Add entries for them.
3826 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3827 and the predicate registers.
3828 (aarch64_sve_vg): Declare.
3829 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3830 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3831 (REGMODE_NATURAL_SIZE): Define.
3832 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3834 * config/aarch64/aarch64.c: Include cfgrtl.h.
3835 (simd_immediate_info): Add a constructor for series vectors,
3836 and an associated step field.
3837 (aarch64_sve_vg): New variable.
3838 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3839 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3840 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3841 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3842 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3843 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3844 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3845 (aarch64_get_mask_mode): New functions.
3846 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3847 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3848 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3849 predicate modes and predicate registers. Explicitly restrict
3850 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3851 to store a vector mode if it is recognized by
3852 aarch64_classify_vector_mode.
3853 (aarch64_regmode_natural_size): New function.
3854 (aarch64_hard_regno_caller_save_mode): Return the original mode
3856 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3857 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3858 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3859 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3861 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3862 does not overlap dest if the function is frame-related. Handle
3864 (aarch64_split_add_offset): New function.
3865 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3866 them aarch64_add_offset.
3867 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3868 and update call to aarch64_sub_sp.
3869 (aarch64_add_cfa_expression): New function.
3870 (aarch64_expand_prologue): Pass extra temporary registers to the
3871 functions above. Handle the case in which we need to emit new
3872 DW_CFA_expressions for registers that were originally saved
3873 relative to the stack pointer, but now have to be expressed
3874 relative to the frame pointer.
3875 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3877 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3878 IP0 and IP1 values for SVE frames.
3879 (aarch64_expand_vec_series): New function.
3880 (aarch64_expand_sve_widened_duplicate): Likewise.
3881 (aarch64_expand_sve_const_vector): Likewise.
3882 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3883 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3884 into the register, rather than emitting a SET directly.
3885 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3886 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3887 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3888 (offset_9bit_signed_scaled_p): New functions.
3889 (aarch64_replicate_bitmask_imm): New function.
3890 (aarch64_bitmask_imm): Use it.
3891 (aarch64_cannot_force_const_mem): Reject expressions involving
3892 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3893 (aarch64_classify_index): Handle SVE indices, by requiring
3894 a plain register index with a scale that matches the element size.
3895 (aarch64_classify_address): Handle SVE addresses. Assert that
3896 the mode of the address is VOIDmode or an integer mode.
3897 Update call to aarch64_classify_symbol.
3898 (aarch64_classify_symbolic_expression): Update call to
3899 aarch64_classify_symbol.
3900 (aarch64_const_vec_all_in_range_p): New function.
3901 (aarch64_print_vector_float_operand): Likewise.
3902 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3903 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3904 and the FP immediates 1.0 and 0.5.
3905 (aarch64_print_address_internal): Handle SVE addresses.
3906 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3907 (aarch64_regno_regclass): Handle predicate registers.
3908 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3910 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3911 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3912 (aarch64_convert_sve_vector_bits): New function.
3913 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3914 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3916 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3917 Handle SVE vector and predicate modes. Accept VL-based constants
3918 that need only one temporary register, and VL offsets that require
3919 no temporary registers.
3920 (aarch64_conditional_register_usage): Mark the predicate registers
3921 as fixed if SVE isn't available.
3922 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3923 Return true for SVE vector and predicate modes.
3924 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3925 rather than an unsigned int. Handle SVE modes.
3926 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3928 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3930 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3931 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3932 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3933 (aarch64_sve_float_mul_immediate_p): New functions.
3934 (aarch64_sve_valid_immediate): New function.
3935 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3936 Explicitly reject structure modes. Check for INDEX constants.
3937 Handle PTRUE and PFALSE constants.
3938 (aarch64_check_zero_based_sve_index_immediate): New function.
3939 (aarch64_simd_imm_zero_p): Delete.
3940 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3941 vector modes. Accept constants in the range of CNT[BHWD].
3942 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3943 ask for an Advanced SIMD mode.
3944 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3945 (aarch64_simd_vector_alignment): Handle SVE predicates.
3946 (aarch64_vectorize_preferred_vector_alignment): New function.
3947 (aarch64_simd_vector_alignment_reachable): Use it instead of
3949 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3950 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3952 (MAX_VECT_LEN): Delete.
3953 (expand_vec_perm_d): Add a vec_flags field.
3954 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3955 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3956 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3958 (aarch64_evpc_rev): Rename to...
3959 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3960 (aarch64_evpc_rev_global): New function.
3961 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3962 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3964 (aarch64_evpc_sve_tbl): New function.
3965 (aarch64_expand_vec_perm_const_1): Update after rename of
3966 aarch64_evpc_rev. Handle SVE permutes too, trying
3967 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3968 than aarch64_evpc_tbl.
3969 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3970 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3971 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3972 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3973 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3974 (aarch64_expand_sve_vcond): New functions.
3975 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3976 of aarch64_vector_mode_p.
3977 (aarch64_dwarf_poly_indeterminate_value): New function.
3978 (aarch64_compute_pressure_classes): Likewise.
3979 (aarch64_can_change_mode_class): Likewise.
3980 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3981 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3982 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3983 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3984 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3985 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3986 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3987 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3989 (Dn, Dl, Dr): Accept const as well as const_vector.
3990 (Dz): Likewise. Compare against CONST0_RTX.
3991 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3992 of "vector" where appropriate.
3993 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3994 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3995 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3996 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3997 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3998 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3999 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
4000 (v_int_equiv): Extend to SVE modes.
4001 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
4003 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
4004 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
4005 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
4006 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
4007 (SVE_COND_FP_CMP): New int iterators.
4008 (perm_hilo): Handle the new unpack unspecs.
4009 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4011 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4012 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4013 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4014 (aarch64_equality_operator, aarch64_constant_vector_operand)
4015 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4016 (aarch64_sve_nonimmediate_operand): Likewise.
4017 (aarch64_sve_general_operand): Likewise.
4018 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4019 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4020 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4021 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4022 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4023 (aarch64_sve_float_arith_immediate): Likewise.
4024 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4025 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4026 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4027 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4028 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4029 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4030 (aarch64_sve_float_arith_operand): Likewise.
4031 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4032 (aarch64_sve_float_mul_operand): Likewise.
4033 (aarch64_sve_vec_perm_operand): Likewise.
4034 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4035 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4036 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4037 as well as const_vector.
4038 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4039 in file. Use CONST0_RTX and CONSTM1_RTX.
4040 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4041 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4042 Use aarch64_simd_imm_zero.
4043 * config/aarch64/aarch64-sve.md: New file.
4044 * config/aarch64/aarch64.md: Include it.
4045 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4046 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4047 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4048 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4049 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4050 (sve): New attribute.
4051 (enabled): Disable instructions with the sve attribute unless
4053 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4054 aarch64_expand_mov_immediate.
4055 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4056 CNT[BHSD] immediates.
4057 (movti): Split CONST_POLY_INT moves into two halves.
4058 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4059 Split additions that need a temporary here if the destination
4060 is the stack pointer.
4061 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4062 (*add<mode>3_poly_1): New instruction.
4063 (set_clobber_cc): New expander.
4065 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4067 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4068 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4069 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4070 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4071 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4072 Change innermode from fixed_mode_size to machine_mode.
4073 (simplify_subreg): Update call accordingly. Handle a constant-sized
4074 subreg of a variable-length CONST_VECTOR.
4076 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4077 Alan Hayward <alan.hayward@arm.com>
4078 David Sherwood <david.sherwood@arm.com>
4080 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4081 (add_offset_to_base): New function, split out from...
4082 (create_mem_ref): ...here. When handling a scale other than 1,
4083 check first whether the address is valid without the offset.
4084 Add it into the base if so, leaving the index and scale as-is.
4086 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4089 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4090 fold_for_warn before checking if arg2 is INTEGER_CST.
4092 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4094 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4095 (store_multiple_operation): Delete.
4096 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4097 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4098 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4099 guarded by TARGET_STRING.
4100 (rs6000_output_load_multiple): Delete.
4101 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4102 OPTION_MASK_STRING / TARGET_STRING handling.
4103 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4104 (const rs6000_opt_masks) <"string">: Change mask to 0.
4105 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4106 (MASK_STRING): Delete.
4107 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4109 (load_multiple): Delete.
4116 (store_multiple): Delete.
4123 (movmemsi_8reg): Delete.
4124 (corresponding unnamed define_insn): Delete.
4125 (movmemsi_6reg): Delete.
4126 (corresponding unnamed define_insn): Delete.
4127 (movmemsi_4reg): Delete.
4128 (corresponding unnamed define_insn): Delete.
4129 (movmemsi_2reg): Delete.
4130 (corresponding unnamed define_insn): Delete.
4131 (movmemsi_1reg): Delete.
4132 (corresponding unnamed define_insn): Delete.
4133 * config/rs6000/rs6000.opt (mno-string): New.
4134 (mstring): Replace by deprecation warning stub.
4135 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4137 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4139 * regrename.c (regrename_do_replace): If replacing the same
4140 reg multiple times, try to reuse last created gen_raw_REG.
4143 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4144 main to workaround a bug in GDB.
4146 2018-01-12 Tom de Vries <tom@codesourcery.com>
4149 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4151 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4153 PR rtl-optimization/80481
4154 * ira-color.c (get_cap_member): New function.
4155 (allocnos_conflict_by_live_ranges_p): Use it.
4156 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4157 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4159 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4162 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4163 (*saddl_se_1): Ditto.
4165 (*ssubl_se_1): Ditto.
4167 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4169 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4170 rather than wi::to_widest for DR_INITs.
4171 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4172 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4173 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4175 (vect_analyze_group_access_1): Note that here.
4177 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4179 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4180 polynomial type sizes.
4182 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4184 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4185 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4186 (gimple_add_tmp_var): Likewise.
4188 2018-01-12 Martin Liska <mliska@suse.cz>
4190 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4191 (gimple_alloc_sizes): Likewise.
4192 (dump_gimple_statistics): Use PRIu64 in printf format.
4193 * gimple.h: Change uint64_t to int.
4195 2018-01-12 Martin Liska <mliska@suse.cz>
4197 * tree-core.h: Use uint64_t instead of int.
4198 * tree.c (tree_node_counts): Likewise.
4199 (tree_node_sizes): Likewise.
4200 (dump_tree_statistics): Use PRIu64 in printf format.
4202 2018-01-12 Martin Liska <mliska@suse.cz>
4204 * Makefile.in: As qsort_chk is implemented in vec.c, add
4205 vec.o to linkage of gencfn-macros.
4206 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4207 passing the info to record_node_allocation_statistics.
4208 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4210 * ggc-common.c (struct ggc_usage): Add operator== and use
4211 it in operator< and compare function.
4212 * mem-stats.h (struct mem_usage): Likewise.
4213 * vec.c (struct vec_usage): Remove operator< and compare
4214 function. Can be simply inherited.
4216 2018-01-12 Martin Jambor <mjambor@suse.cz>
4219 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4220 * tree-ssa-math-opts.c: Include domwalk.h.
4221 (convert_mult_to_fma_1): New function.
4222 (fma_transformation_info): New type.
4223 (fma_deferring_state): Likewise.
4224 (cancel_fma_deferring): New function.
4225 (result_of_phi): Likewise.
4226 (last_fma_candidate_feeds_initial_phi): Likewise.
4227 (convert_mult_to_fma): Added deferring logic, split actual
4228 transformation to convert_mult_to_fma_1.
4229 (math_opts_dom_walker): New type.
4230 (math_opts_dom_walker::after_dom_children): New method, body moved
4231 here from pass_optimize_widening_mul::execute, added deferring logic
4233 (pass_optimize_widening_mul::execute): Moved most of code to
4234 math_opts_dom_walker::after_dom_children.
4235 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4236 * config/i386/i386.c (ix86_option_override_internal): Added
4237 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4239 2018-01-12 Richard Biener <rguenther@suse.de>
4242 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4243 inline instance vars.
4245 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4248 * config/rx/rx.c (rx_is_restricted_memory_address):
4251 2018-01-12 Richard Biener <rguenther@suse.de>
4253 PR tree-optimization/80846
4254 * target.def (split_reduction): New target hook.
4255 * targhooks.c (default_split_reduction): New function.
4256 * targhooks.h (default_split_reduction): Declare.
4257 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4258 target requests first reduce vectors by combining low and high
4260 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4261 (get_vectype_for_scalar_type_and_size): Export.
4262 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4263 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4264 * doc/tm.texi: Regenerate.
4265 * config/i386/i386.c (ix86_split_reduction): Implement
4266 TARGET_VECTORIZE_SPLIT_REDUCTION.
4268 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4271 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4272 in PIC mode except for TARGET_VXWORKS_RTP.
4273 * config/sparc/sparc.c: Include cfgrtl.h.
4274 (TARGET_INIT_PIC_REG): Define.
4275 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4276 (sparc_pic_register_p): New predicate.
4277 (sparc_legitimate_address_p): Use it.
4278 (sparc_legitimize_pic_address): Likewise.
4279 (sparc_delegitimize_address): Likewise.
4280 (sparc_mode_dependent_address_p): Likewise.
4281 (gen_load_pcrel_sym): Remove 4th parameter.
4282 (load_got_register): Adjust call to above. Remove obsolete stuff.
4283 (sparc_expand_prologue): Do not call load_got_register here.
4284 (sparc_flat_expand_prologue): Likewise.
4285 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4286 (sparc_use_pseudo_pic_reg): New function.
4287 (sparc_init_pic_reg): Likewise.
4288 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4289 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4291 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4293 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4294 Add item for branch_cost.
4296 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4298 PR rtl-optimization/83565
4299 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4300 not extend the result to a larger mode for rotate operations.
4301 (num_sign_bit_copies1): Likewise.
4303 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4306 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4308 Use values-Xc.o for -pedantic.
4309 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4311 2018-01-12 Martin Liska <mliska@suse.cz>
4314 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4316 (possible_polymorphic_call_targets): Use it.
4317 (ipa_devirt): Likewise.
4319 2018-01-12 Martin Liska <mliska@suse.cz>
4321 * profile-count.h (enum profile_quality): Use 0 as invalid
4322 enum value of profile_quality.
4324 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4326 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4327 -mext-string options.
4329 2018-01-12 Richard Biener <rguenther@suse.de>
4331 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4332 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4333 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4335 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4337 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4339 * configure.ac (--with-long-double-format): Add support for the
4340 configuration option to change the default long double format on
4342 * config.gcc (powerpc*-linux*-*): Likewise.
4343 * configure: Regenerate.
4344 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4345 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4346 used without modification.
4348 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4350 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4351 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4352 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4353 MISC_BUILTIN_SPEC_BARRIER.
4354 (rs6000_init_builtins): Likewise.
4355 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4357 (speculation_barrier): New define_insn.
4358 * doc/extend.texi: Document __builtin_speculation_barrier.
4360 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4363 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4364 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4365 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4367 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4368 integral modes instead of "ss" and "sd".
4369 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4370 vectors with 32-bit and 64-bit elements.
4371 (vecdupssescalarmodesuffix): New mode attribute.
4372 (vec_dup<mode>): Use it.
4374 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4377 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4378 frame if argument is passed on stack.
4380 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4383 * ree.c (combine_reaching_defs): Optimize also
4384 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4385 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4387 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4390 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4392 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4395 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4396 after they are computed.
4398 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4400 PR tree-optimization/83695
4401 * gimple-loop-linterchange.cc
4402 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4403 reset cached scev information after interchange.
4404 (pass_linterchange::execute): Remove call to scev_reset_htab.
4406 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4408 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4409 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4410 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4411 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4412 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4413 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4414 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4415 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4416 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4417 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4418 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4419 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4420 (V_lane_reg): Likewise.
4421 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4423 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4424 (vfmal_lane_low<mode>_intrinsic,
4425 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4426 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4427 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4428 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4429 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4430 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4432 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4434 * config/arm/arm-cpus.in (fp16fml): New feature.
4435 (ALL_SIMD): Add fp16fml.
4436 (armv8.2-a): Add fp16fml as an option.
4437 (armv8.3-a): Likewise.
4438 (armv8.4-a): Add fp16fml as part of fp16.
4439 * config/arm/arm.h (TARGET_FP16FML): Define.
4440 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4442 * config/arm/arm-modes.def (V2HF): Define.
4443 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4444 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4445 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4446 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4447 vfmsl_low, vfmsl_high): New set of builtins.
4448 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4449 (vfml_op): New code attribute.
4450 (VFMLHALVES): New int iterator.
4451 (VFML, VFMLSEL): New mode attributes.
4452 (V_reg): Define mapping for V2HF.
4453 (V_hi, V_lo): New mode attributes.
4454 (VF_constraint): Likewise.
4455 (vfml_half, vfml_half_selector): New int attributes.
4456 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4458 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4459 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4461 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4462 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4463 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4464 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4466 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4467 Document new effective target and option set.
4469 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4471 * config/arm/arm-cpus.in (armv8_4): New feature.
4472 (ARMv8_4a): New fgroup.
4473 (armv8.4-a): New arch.
4474 * config/arm/arm-tables.opt: Regenerate.
4475 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4476 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4477 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4478 Add matching rules for -march=armv8.4-a and extensions.
4479 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4481 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4484 * config/rx/rx.md (BW): New mode attribute.
4485 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4487 2018-01-11 Richard Biener <rguenther@suse.de>
4489 PR tree-optimization/83435
4490 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4491 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4492 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4494 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4495 Alan Hayward <alan.hayward@arm.com>
4496 David Sherwood <david.sherwood@arm.com>
4498 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4500 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4501 (aarch64_print_address_internal): Use it to check for a zero offset.
4503 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4504 Alan Hayward <alan.hayward@arm.com>
4505 David Sherwood <david.sherwood@arm.com>
4507 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4508 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4509 Return a poly_int64 rather than a HOST_WIDE_INT.
4510 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4511 rather than a HOST_WIDE_INT.
4512 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4513 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4514 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4515 final_offset from HOST_WIDE_INT to poly_int64.
4516 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4517 to_constant when getting the number of units in an Advanced SIMD
4519 (aarch64_builtin_vectorized_function): Check for a constant number
4521 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4523 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4524 attribute instead of GET_MODE_NUNITS.
4525 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4526 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4527 GET_MODE_SIZE for fixed-size registers.
4528 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4529 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4530 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4531 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4532 (aarch64_print_operand, aarch64_print_address_internal)
4533 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4534 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4535 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4536 Handle polynomial GET_MODE_SIZE.
4537 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4538 wider than SImode without modification.
4539 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4540 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4541 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4542 passing and returning SVE modes.
4543 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4544 rather than GEN_INT.
4545 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4546 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4547 (aarch64_allocate_and_probe_stack_space): Likewise.
4548 (aarch64_layout_frame): Cope with polynomial offsets.
4549 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4550 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4552 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4553 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4554 poly_int64 rather than a HOST_WIDE_INT.
4555 (aarch64_get_separate_components, aarch64_process_components)
4556 (aarch64_expand_prologue, aarch64_expand_epilogue)
4557 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4558 (aarch64_anchor_offset): New function, split out from...
4559 (aarch64_legitimize_address): ...here.
4560 (aarch64_builtin_vectorization_cost): Handle polynomial
4561 TYPE_VECTOR_SUBPARTS.
4562 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4564 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4565 number of elements from the PARALLEL rather than the mode.
4566 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4567 rather than GET_MODE_BITSIZE.
4568 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4569 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4570 (aarch64_expand_vec_perm_const_1): Handle polynomial
4571 d->perm.length () and d->perm elements.
4572 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4573 Apply to_constant to d->perm elements.
4574 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4575 polynomial CONST_VECTOR_NUNITS.
4576 (aarch64_move_pointer): Take amount as a poly_int64 rather
4578 (aarch64_progress_pointer): Avoid temporary variable.
4579 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4580 the mode attribute instead of GET_MODE.
4582 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4583 Alan Hayward <alan.hayward@arm.com>
4584 David Sherwood <david.sherwood@arm.com>
4586 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4587 x exists before using it.
4588 (aarch64_add_constant_internal): Rename to...
4589 (aarch64_add_offset_1): ...this. Replace regnum with separate
4590 src and dest rtxes. Handle the case in which they're different,
4591 including when the offset is zero. Replace scratchreg with an rtx.
4592 Use 2 additions if there is no spare register into which we can
4593 move a 16-bit constant.
4594 (aarch64_add_constant): Delete.
4595 (aarch64_add_offset): Replace reg with separate src and dest
4596 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4597 Use aarch64_add_offset_1.
4598 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4599 an rtx rather than an int. Take the delta as a poly_int64
4600 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4601 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4602 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4603 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4604 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4606 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4607 aarch64_add_constant.
4609 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4611 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4612 Use scalar_float_mode.
4614 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4616 * config/aarch64/aarch64-simd.md
4617 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4618 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4619 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4620 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4621 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4622 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4623 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4624 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4625 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4626 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4628 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4631 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4632 targ_options->x_arm_arch_string is non NULL.
4634 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4636 * config/aarch64/aarch64.h
4637 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4639 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4642 * expmed.c (emit_store_flag_force): Swap if const op0
4643 and change VOIDmode to mode of op0.
4645 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4647 PR rtl-optimization/83761
4648 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4649 than bytes to mode_for_size.
4651 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4654 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4655 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4658 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4661 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4662 when in layout mode.
4663 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4664 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4667 2018-01-10 Michael Collison <michael.collison@arm.com>
4669 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4670 * config/aarch64/aarch64-option-extension.def: Add
4671 AARCH64_OPT_EXTENSION of 'fp16fml'.
4672 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4673 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4674 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4675 * config/aarch64/constraints.md (Ui7): New constraint.
4676 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4677 (VFMLA_SEL_W): Ditto.
4680 (VFMLA16_LOW): New int iterator.
4681 (VFMLA16_HIGH): Ditto.
4682 (UNSPEC_FMLAL): New unspec.
4683 (UNSPEC_FMLSL): Ditto.
4684 (UNSPEC_FMLAL2): Ditto.
4685 (UNSPEC_FMLSL2): Ditto.
4686 (f16mac): New code attribute.
4687 * config/aarch64/aarch64-simd-builtins.def
4688 (aarch64_fmlal_lowv2sf): Ditto.
4689 (aarch64_fmlsl_lowv2sf): Ditto.
4690 (aarch64_fmlalq_lowv4sf): Ditto.
4691 (aarch64_fmlslq_lowv4sf): Ditto.
4692 (aarch64_fmlal_highv2sf): Ditto.
4693 (aarch64_fmlsl_highv2sf): Ditto.
4694 (aarch64_fmlalq_highv4sf): Ditto.
4695 (aarch64_fmlslq_highv4sf): Ditto.
4696 (aarch64_fmlal_lane_lowv2sf): Ditto.
4697 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4698 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4699 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4700 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4701 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4702 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4703 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4704 (aarch64_fmlal_lane_highv2sf): Ditto.
4705 (aarch64_fmlsl_lane_highv2sf): Ditto.
4706 (aarch64_fmlal_laneq_highv2sf): Ditto.
4707 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4708 (aarch64_fmlalq_lane_highv4sf): Ditto.
4709 (aarch64_fmlsl_lane_highv4sf): Ditto.
4710 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4711 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4712 * config/aarch64/aarch64-simd.md:
4713 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4714 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4715 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4716 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4717 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4718 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4719 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4720 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4721 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4722 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4723 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4724 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4725 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4726 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4727 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4728 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4729 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4730 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4731 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4732 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4733 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4734 (vfmlsl_low_u32): Ditto.
4735 (vfmlalq_low_u32): Ditto.
4736 (vfmlslq_low_u32): Ditto.
4737 (vfmlal_high_u32): Ditto.
4738 (vfmlsl_high_u32): Ditto.
4739 (vfmlalq_high_u32): Ditto.
4740 (vfmlslq_high_u32): Ditto.
4741 (vfmlal_lane_low_u32): Ditto.
4742 (vfmlsl_lane_low_u32): Ditto.
4743 (vfmlal_laneq_low_u32): Ditto.
4744 (vfmlsl_laneq_low_u32): Ditto.
4745 (vfmlalq_lane_low_u32): Ditto.
4746 (vfmlslq_lane_low_u32): Ditto.
4747 (vfmlalq_laneq_low_u32): Ditto.
4748 (vfmlslq_laneq_low_u32): Ditto.
4749 (vfmlal_lane_high_u32): Ditto.
4750 (vfmlsl_lane_high_u32): Ditto.
4751 (vfmlal_laneq_high_u32): Ditto.
4752 (vfmlsl_laneq_high_u32): Ditto.
4753 (vfmlalq_lane_high_u32): Ditto.
4754 (vfmlslq_lane_high_u32): Ditto.
4755 (vfmlalq_laneq_high_u32): Ditto.
4756 (vfmlslq_laneq_high_u32): Ditto.
4757 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4758 (AARCH64_FL_FOR_ARCH8_4): New.
4759 (AARCH64_ISA_F16FML): New ISA flag.
4760 (TARGET_F16FML): New feature flag for fp16fml.
4761 (doc/invoke.texi): Document new fp16fml option.
4763 2018-01-10 Michael Collison <michael.collison@arm.com>
4765 * config/aarch64/aarch64-builtins.c:
4766 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4767 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4768 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4769 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4770 (AARCH64_ISA_SHA3): New ISA flag.
4771 (TARGET_SHA3): New feature flag for sha3.
4772 * config/aarch64/iterators.md (sha512_op): New int attribute.
4773 (CRYPTO_SHA512): New int iterator.
4774 (UNSPEC_SHA512H): New unspec.
4775 (UNSPEC_SHA512H2): Ditto.
4776 (UNSPEC_SHA512SU0): Ditto.
4777 (UNSPEC_SHA512SU1): Ditto.
4778 * config/aarch64/aarch64-simd-builtins.def
4779 (aarch64_crypto_sha512hqv2di): New builtin.
4780 (aarch64_crypto_sha512h2qv2di): Ditto.
4781 (aarch64_crypto_sha512su0qv2di): Ditto.
4782 (aarch64_crypto_sha512su1qv2di): Ditto.
4783 (aarch64_eor3qv8hi): Ditto.
4784 (aarch64_rax1qv2di): Ditto.
4785 (aarch64_xarqv2di): Ditto.
4786 (aarch64_bcaxqv8hi): Ditto.
4787 * config/aarch64/aarch64-simd.md:
4788 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4789 (aarch64_crypto_sha512su0qv2di): Ditto.
4790 (aarch64_crypto_sha512su1qv2di): Ditto.
4791 (aarch64_eor3qv8hi): Ditto.
4792 (aarch64_rax1qv2di): Ditto.
4793 (aarch64_xarqv2di): Ditto.
4794 (aarch64_bcaxqv8hi): Ditto.
4795 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4796 (vsha512h2q_u64): Ditto.
4797 (vsha512su0q_u64): Ditto.
4798 (vsha512su1q_u64): Ditto.
4799 (veor3q_u16): Ditto.
4800 (vrax1q_u64): Ditto.
4802 (vbcaxq_u16): Ditto.
4803 * config/arm/types.md (crypto_sha512): New type attribute.
4804 (crypto_sha3): Ditto.
4805 (doc/invoke.texi): Document new sha3 option.
4807 2018-01-10 Michael Collison <michael.collison@arm.com>
4809 * config/aarch64/aarch64-builtins.c:
4810 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4811 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4812 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4813 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4814 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4815 (AARCH64_ISA_SM4): New ISA flag.
4816 (TARGET_SM4): New feature flag for sm4.
4817 * config/aarch64/aarch64-simd-builtins.def
4818 (aarch64_sm3ss1qv4si): Ditto.
4819 (aarch64_sm3tt1aq4si): Ditto.
4820 (aarch64_sm3tt1bq4si): Ditto.
4821 (aarch64_sm3tt2aq4si): Ditto.
4822 (aarch64_sm3tt2bq4si): Ditto.
4823 (aarch64_sm3partw1qv4si): Ditto.
4824 (aarch64_sm3partw2qv4si): Ditto.
4825 (aarch64_sm4eqv4si): Ditto.
4826 (aarch64_sm4ekeyqv4si): Ditto.
4827 * config/aarch64/aarch64-simd.md:
4828 (aarch64_sm3ss1qv4si): Ditto.
4829 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4830 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4831 (aarch64_sm4eqv4si): Ditto.
4832 (aarch64_sm4ekeyqv4si): Ditto.
4833 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4834 (sm3part_op): Ditto.
4835 (CRYPTO_SM3TT): Ditto.
4836 (CRYPTO_SM3PART): Ditto.
4837 (UNSPEC_SM3SS1): New unspec.
4838 (UNSPEC_SM3TT1A): Ditto.
4839 (UNSPEC_SM3TT1B): Ditto.
4840 (UNSPEC_SM3TT2A): Ditto.
4841 (UNSPEC_SM3TT2B): Ditto.
4842 (UNSPEC_SM3PARTW1): Ditto.
4843 (UNSPEC_SM3PARTW2): Ditto.
4844 (UNSPEC_SM4E): Ditto.
4845 (UNSPEC_SM4EKEY): Ditto.
4846 * config/aarch64/constraints.md (Ui2): New constraint.
4847 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4848 * config/arm/types.md (crypto_sm3): New type attribute.
4849 (crypto_sm4): Ditto.
4850 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4851 (vsm3tt1aq_u32): Ditto.
4852 (vsm3tt1bq_u32): Ditto.
4853 (vsm3tt2aq_u32): Ditto.
4854 (vsm3tt2bq_u32): Ditto.
4855 (vsm3partw1q_u32): Ditto.
4856 (vsm3partw2q_u32): Ditto.
4857 (vsm4eq_u32): Ditto.
4858 (vsm4ekeyq_u32): Ditto.
4859 (doc/invoke.texi): Document new sm4 option.
4861 2018-01-10 Michael Collison <michael.collison@arm.com>
4863 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4864 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4865 (AARCH64_FL_FOR_ARCH8_4): New.
4866 (AARCH64_FL_V8_4): New flag.
4867 (doc/invoke.texi): Document new armv8.4-a option.
4869 2018-01-10 Michael Collison <michael.collison@arm.com>
4871 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4872 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4873 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4874 * config/aarch64/aarch64-option-extension.def: Add
4875 AARCH64_OPT_EXTENSION of 'sha2'.
4876 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4877 (crypto): Disable sha2 and aes if crypto disabled.
4878 (crypto): Enable aes and sha2 if enabled.
4879 (simd): Disable sha2 and aes if simd disabled.
4880 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4882 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4883 (TARGET_SHA2): New feature flag for sha2.
4884 (TARGET_AES): New feature flag for aes.
4885 * config/aarch64/aarch64-simd.md:
4886 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4887 conditional on TARGET_AES.
4888 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4889 (aarch64_crypto_sha1hsi): Make pattern conditional
4891 (aarch64_crypto_sha1hv4si): Ditto.
4892 (aarch64_be_crypto_sha1hv4si): Ditto.
4893 (aarch64_crypto_sha1su1v4si): Ditto.
4894 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4895 (aarch64_crypto_sha1su0v4si): Ditto.
4896 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4897 (aarch64_crypto_sha256su0v4si): Ditto.
4898 (aarch64_crypto_sha256su1v4si): Ditto.
4899 (doc/invoke.texi): Document new aes and sha2 options.
4901 2018-01-10 Martin Sebor <msebor@redhat.com>
4903 PR tree-optimization/83781
4904 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4907 2018-01-11 Martin Sebor <msebor@gmail.com>
4908 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4910 PR tree-optimization/83501
4911 PR tree-optimization/81703
4913 * tree-ssa-strlen.c (get_string_cst): Rename...
4914 (get_string_len): ...to this. Handle global constants.
4915 (handle_char_store): Adjust.
4917 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4918 Jim Wilson <jimw@sifive.com>
4920 * config/riscv/riscv-protos.h (riscv_output_return): New.
4921 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4922 (riscv_attribute_table, riscv_output_return),
4923 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4924 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4925 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4927 (riscv_expand_prologue): Add early return for naked function.
4928 (riscv_expand_epilogue): Likewise.
4929 (riscv_function_ok_for_sibcall): Return false for naked function.
4930 (riscv_set_current_function): New.
4931 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4932 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4933 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4934 * doc/extend.texi (RISC-V Function Attributes): New.
4936 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4938 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4939 check for 128-bit long double before checking TCmode.
4940 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4941 128-bit long doubles before checking TFmode or TCmode.
4942 (FLOAT128_IBM_P): Likewise.
4944 2018-01-10 Martin Sebor <msebor@redhat.com>
4946 PR tree-optimization/83671
4947 * builtins.c (c_strlen): Unconditionally return zero for the empty
4949 Use -Warray-bounds for warnings.
4950 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4951 for non-constant array indices with COMPONENT_REF, arrays of
4952 arrays, and pointers to arrays.
4953 (gimple_fold_builtin_strlen): Determine and set length range for
4954 non-constant character arrays.
4956 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4959 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4962 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4964 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4966 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4969 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4970 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4971 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4972 indexed_or_indirect_operand predicate.
4973 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4974 (*vsx_le_perm_load_v8hi): Likewise.
4975 (*vsx_le_perm_load_v16qi): Likewise.
4976 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4977 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4978 (*vsx_le_perm_store_v8hi): Likewise.
4979 (*vsx_le_perm_store_v16qi): Likewise.
4980 (eight unnamed splitters): Likewise.
4982 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4984 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4985 * config/rs6000/emmintrin.h: Likewise.
4986 * config/rs6000/mmintrin.h: Likewise.
4987 * config/rs6000/xmmintrin.h: Likewise.
4989 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4992 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4994 * tree.c (tree_nop_conversion): Return true for location wrapper
4996 (maybe_wrap_with_location): New function.
4997 (selftest::check_strip_nops): New function.
4998 (selftest::test_location_wrappers): New function.
4999 (selftest::tree_c_tests): Call it.
5000 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
5001 (maybe_wrap_with_location): New decl.
5002 (EXPR_LOCATION_WRAPPER_P): New macro.
5003 (location_wrapper_p): New inline function.
5004 (tree_strip_any_location_wrapper): New inline function.
5006 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
5009 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5010 stack_realign_offset for the largest alignment of stack slot
5012 (ix86_find_max_used_stack_alignment): New function.
5013 (ix86_finalize_stack_frame_flags): Use it. Set
5014 max_used_stack_alignment if we don't realign stack.
5015 * config/i386/i386.h (machine_function): Add
5016 max_used_stack_alignment.
5018 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5020 * config/arm/arm.opt (-mbranch-cost): New option.
5021 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5024 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5027 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5028 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5030 2018-01-10 Richard Biener <rguenther@suse.de>
5033 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5034 early out so it also covers the case where we have a non-NULL
5037 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5039 PR tree-optimization/83753
5040 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5041 for non-strided grouped accesses if the number of elements is 1.
5043 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5046 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5047 * i386.h (TARGET_USE_GATHER): Define.
5048 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5050 2018-01-10 Martin Liska <mliska@suse.cz>
5053 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5054 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5056 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5057 CLEANUP_NO_PARTITIONING is not set.
5059 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5061 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5062 for vectors, as a partial revert of r254296.
5063 * rtl.h (const_vec_p): Delete.
5064 (const_vec_duplicate_p): Don't test for vector CONSTs.
5065 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5066 * expmed.c (make_tree): Likewise.
5069 * common.md (E, F): Use CONSTANT_P instead of checking for
5071 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5072 checking for CONST_VECTOR.
5074 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5077 * predict.c (force_edge_cold): Handle in more sane way edges
5080 2018-01-09 Carl Love <cel@us.ibm.com>
5082 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5084 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5085 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5086 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5087 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5088 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5089 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5090 * config/rs6000/rs6000-protos.h: Add extern defition for
5091 rs6000_generate_float2_double_code.
5092 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5094 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5095 (float2_v2df): Add define_expand.
5097 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5100 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5101 op_mode in the force_to_mode call.
5103 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5105 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5106 instead of checking each element individually.
5107 (aarch64_evpc_uzp): Likewise.
5108 (aarch64_evpc_zip): Likewise.
5109 (aarch64_evpc_ext): Likewise.
5110 (aarch64_evpc_rev): Likewise.
5111 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5112 instead of checking each element individually. Return true without
5114 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5115 whether all selected elements come from the same input, instead of
5116 checking each element individually. Remove calls to gen_rtx_REG,
5117 start_sequence and end_sequence and instead assert that no rtl is
5120 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5122 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5123 order of HIGH and CONST checks.
5125 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5127 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5128 if the destination isn't an SSA_NAME.
5130 2018-01-09 Richard Biener <rguenther@suse.de>
5132 PR tree-optimization/83668
5133 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5135 (canonicalize_loop_form): ... here, renamed from ...
5136 (canonicalize_loop_closed_ssa_form): ... this and amended to
5137 swap successor edges for loop exit blocks to make us use
5138 the RPO order we need for initial schedule generation.
5140 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5142 PR tree-optimization/64811
5143 * match.pd: When optimizing comparisons with Inf, avoid
5144 introducing or losing exceptions from comparisons with NaN.
5146 2018-01-09 Martin Liska <mliska@suse.cz>
5149 * asan.c (shadow_mem_size): Add gcc_assert.
5151 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5153 Don't save registers in main().
5156 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5157 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5158 * config/avr/avr.c (avr_set_current_function): Don't error if
5159 naked, OS_task or OS_main are specified at the same time.
5160 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5162 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5164 * common/config/avr/avr-common.c (avr_option_optimization_table):
5165 Switch on -mmain-is-OS_task for optimizing compilations.
5167 2018-01-09 Richard Biener <rguenther@suse.de>
5169 PR tree-optimization/83572
5170 * graphite.c: Include cfganal.h.
5171 (graphite_transform_loops): Connect infinite loops to exit
5172 and remove fake edges at the end.
5174 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5176 * ipa-inline.c (edge_badness): Revert accidental checkin.
5178 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5181 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5182 symbols; not inline clones.
5184 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5187 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5188 hard registers. Formatting fixes.
5190 PR preprocessor/83722
5191 * gcc.c (try_generate_repro): Pass
5192 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5193 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5196 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5197 Kito Cheng <kito.cheng@gmail.com>
5199 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5200 (riscv_leaf_function_p): Delete.
5201 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5203 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5205 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5207 (do_ifelse): New function.
5208 (do_isel): New function.
5209 (do_sub3): New function.
5210 (do_add3): New function.
5211 (do_load_mask_compare): New function.
5212 (do_overlap_load_compare): New function.
5213 (expand_compare_loop): New function.
5214 (expand_block_compare): Call expand_compare_loop() when appropriate.
5215 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5217 (-mblock-compare-inline-loop-limit): New option.
5219 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5222 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5223 Reverse order of second and third operands in first alternative.
5224 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5225 of first and second elements in UNSPEC_VPERMR vector.
5226 (altivec_expand_vec_perm_le): Likewise.
5228 2018-01-08 Jeff Law <law@redhat.com>
5230 PR rtl-optimizatin/81308
5231 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5232 (process_switch): If group_case_labels makes a change, then set
5234 (pass_convert_switch::execute): If a switch is converted, then
5235 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5237 PR rtl-optimization/81308
5238 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5241 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5243 PR target/83663 - Revert r255946
5244 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5245 generation for cases where splatting a value is not useful.
5246 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5247 across a vec_duplicate and a paradoxical subreg forming a vector
5248 mode to a vec_concat.
5250 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5252 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5253 -march=armv8.3-a variants.
5254 * config/arm/t-multilib: Likewise.
5255 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5257 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5259 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5261 (cceq_ior_compare_complement): Give it a name so I can use it, and
5262 change boolean_or_operator predicate to boolean_operator so it can
5263 be used to generate a crand.
5264 (eqne): New code iterator.
5265 (bd/bd_neg): New code_attrs.
5266 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5267 a single define_insn.
5268 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5269 decrement (bdnzt/bdnzf/bdzt/bdzf).
5270 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5271 with the new names of the branch decrement patterns, and added the
5272 names of the branch decrement conditional patterns.
5274 2018-01-08 Richard Biener <rguenther@suse.de>
5276 PR tree-optimization/83563
5277 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5280 2018-01-08 Richard Biener <rguenther@suse.de>
5283 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5285 2018-01-08 Richard Biener <rguenther@suse.de>
5287 PR tree-optimization/83685
5288 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5289 references to abnormals.
5291 2018-01-08 Richard Biener <rguenther@suse.de>
5294 * dwarf2out.c (output_indirect_strings): Handle empty
5295 skeleton_debug_str_hash.
5296 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5298 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5300 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5301 (emit_store_direct): Likewise.
5302 (arc_trampoline_adjust_address): Likewise.
5303 (arc_asm_trampoline_template): New function.
5304 (arc_initialize_trampoline): Use asm_trampoline_template.
5305 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5306 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5307 * config/arc/arc.md (flush_icache): Delete pattern.
5309 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5311 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5312 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5315 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5318 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5319 by not USED_FOR_TARGET.
5320 (make_pass_resolve_sw_modes): Likewise.
5322 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5324 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5327 2018-01-08 Richard Biener <rguenther@suse.de>
5330 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5332 2018-01-08 Richard Biener <rguenther@suse.de>
5335 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5337 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5340 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5341 basic blocks with a small number of successors.
5342 (convert_control_dep_chain_into_preds): Improve handling of
5344 (dump_predicates): Split apart into...
5345 (dump_pred_chain): ...here...
5346 (dump_pred_info): ...and here.
5347 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5348 (can_chain_union_be_invalidated_p): Improve check for invalidation
5350 (uninit_uses_cannot_happen): Avoid unnecessary if
5351 convert_control_dep_chain_into_preds yielded nothing.
5353 2018-01-06 Martin Sebor <msebor@redhat.com>
5355 PR tree-optimization/83640
5356 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5357 subtracting negative offset from size.
5358 (builtin_access::overlap): Adjust offset bounds of the access to fall
5359 within the size of the object if possible.
5361 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5363 PR rtl-optimization/83699
5364 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5365 extract_bit_field_as_subreg to cases in which the extracted
5366 value is also a vector.
5368 * lra-constraints.c (process_alt_operands): Test for the equivalence
5369 substitutions when detecting a possible reload cycle.
5371 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5374 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5375 by default if flag_selective_schedling{,2}. Formatting fixes.
5377 PR rtl-optimization/83682
5378 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5379 if it has non-VECTOR_MODE element mode.
5380 (vec_duplicate_p): Likewise.
5383 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5384 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5386 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5389 * config/i386/i386-builtin.def
5390 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5391 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5392 Require also OPTION_MASK_ISA_AVX512F in addition to
5393 OPTION_MASK_ISA_GFNI.
5394 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5395 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5396 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5397 to OPTION_MASK_ISA_GFNI.
5398 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5399 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5400 OPTION_MASK_ISA_AVX512BW.
5401 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5402 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5403 addition to OPTION_MASK_ISA_GFNI.
5404 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5405 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5406 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5407 to OPTION_MASK_ISA_GFNI.
5408 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5409 a requirement for all ISAs rather than any of them with a few
5411 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5413 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5414 bitmasks to be enabled with 3 exceptions, instead of requiring any
5415 enabled ISA with lots of exceptions.
5416 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5417 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5418 Change avx512bw in isa attribute to avx512f.
5419 * config/i386/sgxintrin.h: Add license boilerplate.
5420 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5421 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5422 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5423 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5425 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5426 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5427 temporarily sse2 rather than sse if not enabled already.
5430 * config/i386/sse.md (VI248_VLBW): Rename to ...
5431 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5432 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5433 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5434 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5435 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5436 mode iterator instead of VI248_VLBW.
5438 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5440 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5441 (record_modified): Skip clobbers; add debug output.
5442 (param_change_prob): Use sreal frequencies.
5444 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5446 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5447 punt for user-aligned variables.
5449 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5451 * tree-chrec.c (chrec_contains_symbols): Return true for
5454 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5457 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5458 of (x|y) == x for BICS pattern.
5460 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5462 PR tree-optimization/83605
5463 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5464 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5467 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5469 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5470 * config/epiphany/rtems.h: New file.
5472 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5473 Uros Bizjak <ubizjak@gmail.com>
5476 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5477 QIreg_operand instead of register_operand predicate.
5478 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5479 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5480 comments instead of -fmitigate[-_]rop.
5482 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5485 * cgraphunit.c (symbol_table::compile): Switch to text_section
5486 before calling assembly_start debug hook.
5487 * run-rtl-passes.c (run_rtl_passes): Likewise.
5490 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5492 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5493 range_int_cst_p rather than !symbolic_range_p before calling
5494 extract_range_from_multiplicative_op_1.
5496 2018-01-04 Jeff Law <law@redhat.com>
5498 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5499 redundant test in assertion.
5501 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5503 * doc/rtl.texi: Document machine_mode wrapper classes.
5505 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5507 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5510 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5512 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5513 the VEC_PERM_EXPR fold to fail.
5515 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5518 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5519 to switched_sections.
5521 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5524 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5527 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5530 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5531 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5533 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5536 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5537 is BLKmode and bitpos not zero or mode change is needed.
5539 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5542 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5545 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5548 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5549 instead of MULT rtx. Update all corresponding splitters.
5551 (*ssub<modesuffix>): Ditto.
5553 (*cmp_sadd_di): Update split patterns.
5554 (*cmp_sadd_si): Ditto.
5555 (*cmp_sadd_sidi): Ditto.
5556 (*cmp_ssub_di): Ditto.
5557 (*cmp_ssub_si): Ditto.
5558 (*cmp_ssub_sidi): Ditto.
5559 * config/alpha/predicates.md (const23_operand): New predicate.
5560 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5561 Look for ASHIFT, not MULT inner operand.
5562 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5564 2018-01-04 Martin Liska <mliska@suse.cz>
5566 PR gcov-profile/83669
5567 * gcov.c (output_intermediate_file): Add version to intermediate
5569 * doc/gcov.texi: Document new field 'version' in intermediate
5570 file format. Fix location of '-k' option of gcov command.
5572 2018-01-04 Martin Liska <mliska@suse.cz>
5575 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5577 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5579 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5581 2018-01-03 Martin Sebor <msebor@redhat.com>
5583 PR tree-optimization/83655
5584 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5585 checking calls with invalid arguments.
5587 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5589 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5590 (vectorizable_mask_load_store): Delete.
5591 (vectorizable_call): Return false for masked loads and stores.
5592 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5593 instead of gimple_assign_rhs1.
5594 (vectorizable_load): Handle IFN_MASK_LOAD.
5595 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5597 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5599 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5601 (vectorizable_mask_load_store): ...here.
5602 (vectorizable_load): ...and here.
5604 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5606 * tree-vect-stmts.c (vect_build_all_ones_mask)
5607 (vect_build_zero_merge_argument): New functions, split out from...
5608 (vectorizable_load): ...here.
5610 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5612 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5614 (vectorizable_mask_load_store): ...here.
5615 (vectorizable_store): ...and here.
5617 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5619 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5621 (vectorizable_mask_load_store): ...here.
5623 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5625 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5626 (vect_model_store_cost): Take a vec_load_store_type instead of a
5628 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5629 (vect_model_store_cost): Take a vec_load_store_type instead of a
5631 (vectorizable_mask_load_store): Update accordingly.
5632 (vectorizable_store): Likewise.
5633 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5635 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5637 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5638 IFN_MASK_LOAD calls here rather than...
5639 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5641 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5642 Alan Hayward <alan.hayward@arm.com>
5643 David Sherwood <david.sherwood@arm.com>
5645 * expmed.c (extract_bit_field_1): For vector extracts,
5646 fall back to extract_bit_field_as_subreg if vec_extract
5649 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5650 Alan Hayward <alan.hayward@arm.com>
5651 David Sherwood <david.sherwood@arm.com>
5653 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5654 they are variable or constant sized.
5655 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5656 slots for constant-sized data.
5658 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5659 Alan Hayward <alan.hayward@arm.com>
5660 David Sherwood <david.sherwood@arm.com>
5662 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5663 handling COND_EXPRs with boolean comparisons, try to find a better
5664 basis for the mask type than the boolean itself.
5666 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5668 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5669 is calculated and how it can be overridden.
5670 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5671 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5673 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5676 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5677 Alan Hayward <alan.hayward@arm.com>
5678 David Sherwood <david.sherwood@arm.com>
5680 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5681 Remove the mode argument.
5682 (aarch64_simd_valid_immediate): Remove the mode and inverse
5684 * config/aarch64/iterators.md (bitsize): New iterator.
5685 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5686 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5687 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5688 aarch64_simd_valid_immediate.
5689 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5690 (aarch64_reg_or_bic_imm): Likewise.
5691 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5692 with an insn_type enum and msl with a modifier_type enum.
5693 Replace element_width with a scalar_mode. Change the shift
5694 to unsigned int. Add constructors for scalar_float_mode and
5695 scalar_int_mode elements.
5696 (aarch64_vect_float_const_representable_p): Delete.
5697 (aarch64_can_const_movi_rtx_p)
5698 (aarch64_simd_scalar_immediate_valid_for_move)
5699 (aarch64_simd_make_constant): Update call to
5700 aarch64_simd_valid_immediate.
5701 (aarch64_advsimd_valid_immediate_hs): New function.
5702 (aarch64_advsimd_valid_immediate): Likewise.
5703 (aarch64_simd_valid_immediate): Remove mode and inverse
5704 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5705 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5706 and aarch64_float_const_representable_p on the result.
5707 (aarch64_output_simd_mov_immediate): Remove mode argument.
5708 Update call to aarch64_simd_valid_immediate and use of
5709 simd_immediate_info.
5710 (aarch64_output_scalar_simd_mov_immediate): Update call
5713 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5714 Alan Hayward <alan.hayward@arm.com>
5715 David Sherwood <david.sherwood@arm.com>
5717 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5718 (mode_nunits): Likewise CONST_MODE_NUNITS.
5719 * machmode.def (ADJUST_NUNITS): Document.
5720 * genmodes.c (mode_data::need_nunits_adj): New field.
5721 (blank_mode): Update accordingly.
5722 (adj_nunits): New variable.
5723 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5725 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5726 listed in adj_nunits.
5727 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5728 listed in adj_nunits. Don't emit case statements for such modes.
5729 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5730 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5731 nothing if adj_nunits is nonnull.
5732 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5733 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5734 (emit_mode_fbit): Update use of print_maybe_const_decl.
5735 (emit_move_size): Likewise. Treat the array as non-const
5737 (emit_mode_adjustments): Handle adj_nunits.
5739 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5741 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5742 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5743 (VECTOR_MODES): Use it.
5744 (make_vector_modes): Take the prefix as an argument.
5746 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5747 Alan Hayward <alan.hayward@arm.com>
5748 David Sherwood <david.sherwood@arm.com>
5750 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5751 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5752 for MODE_VECTOR_BOOL.
5753 * machmode.def (VECTOR_BOOL_MODE): Document.
5754 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5755 (make_vector_bool_mode): New function.
5756 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5758 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5759 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5761 * stor-layout.c (int_mode_for_mode): Likewise.
5762 * tree.c (build_vector_type_for_mode): Likewise.
5763 * varasm.c (output_constant_pool_2): Likewise.
5764 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5765 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5766 for MODE_VECTOR_BOOL.
5767 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5768 of mode class checks.
5769 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5770 instead of a list of mode class checks.
5771 (expand_vector_scalar_condition): Likewise.
5772 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5774 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5775 Alan Hayward <alan.hayward@arm.com>
5776 David Sherwood <david.sherwood@arm.com>
5778 * machmode.h (mode_size): Change from unsigned short to
5780 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5781 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5782 or if measurement_type is not polynomial.
5783 (fixed_size_mode::includes_p): Check for constant-sized modes.
5784 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5785 return a poly_uint16 rather than an unsigned short.
5786 (emit_mode_size): Change the type of mode_size from unsigned short
5787 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5788 (emit_mode_adjustments): Cope with polynomial vector sizes.
5789 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5791 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5793 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5794 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5795 * caller-save.c (setup_save_areas): Likewise.
5796 (replace_reg_with_saved_mem): Likewise.
5797 * calls.c (emit_library_call_value_1): Likewise.
5798 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5799 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5800 (gen_lowpart_for_combine): Likewise.
5801 * convert.c (convert_to_integer_1): Likewise.
5802 * cse.c (equiv_constant, cse_insn): Likewise.
5803 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5804 (cselib_subst_to_values): Likewise.
5805 * dce.c (word_dce_process_block): Likewise.
5806 * df-problems.c (df_word_lr_mark_ref): Likewise.
5807 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5808 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5809 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5810 (rtl_for_decl_location): Likewise.
5811 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5812 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5813 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5814 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5815 (expand_expr_real_1): Likewise.
5816 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5817 (pad_below): Likewise.
5818 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5819 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5820 * ira.c (get_subreg_tracking_sizes): Likewise.
5821 * ira-build.c (ira_create_allocno_objects): Likewise.
5822 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5823 (ira_sort_regnos_for_alter_reg): Likewise.
5824 * ira-costs.c (record_operand_costs): Likewise.
5825 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5826 (resolve_simple_move): Likewise.
5827 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5828 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5829 (lra_constraints): Likewise.
5830 (CONST_POOL_OK_P): Reject variable-sized modes.
5831 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5832 (add_pseudo_to_slot, lra_spill): Likewise.
5833 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5834 * optabs-query.c (get_best_extraction_insn): Likewise.
5835 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5836 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5837 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5838 * recog.c (offsettable_address_addr_space_p): Likewise.
5839 * regcprop.c (maybe_mode_change): Likewise.
5840 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5841 * regrename.c (build_def_use): Likewise.
5842 * regstat.c (dump_reg_info): Likewise.
5843 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5844 (find_reloads, find_reloads_subreg_address): Likewise.
5845 * reload1.c (eliminate_regs_1): Likewise.
5846 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5847 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5848 (simplify_binary_operation_1, simplify_subreg): Likewise.
5849 * targhooks.c (default_function_arg_padding): Likewise.
5850 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5851 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5852 (verify_gimple_assign_ternary): Likewise.
5853 * tree-inline.c (estimate_move_cost): Likewise.
5854 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5855 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5856 (get_address_cost_ainc): Likewise.
5857 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5858 (vect_supportable_dr_alignment): Likewise.
5859 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5860 (vectorizable_reduction): Likewise.
5861 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5862 (vectorizable_operation, vectorizable_load): Likewise.
5863 * tree.c (build_same_sized_truth_vector_type): Likewise.
5864 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5865 * var-tracking.c (emit_note_insn_var_location): Likewise.
5866 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5867 (ADDR_VEC_ALIGN): Likewise.
5869 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5870 Alan Hayward <alan.hayward@arm.com>
5871 David Sherwood <david.sherwood@arm.com>
5873 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5875 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5876 or if measurement_type is polynomial.
5877 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5878 * combine.c (make_extraction): Likewise.
5879 * dse.c (find_shift_sequence): Likewise.
5880 * dwarf2out.c (mem_loc_descriptor): Likewise.
5881 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5882 (extract_bit_field, extract_low_bits): Likewise.
5883 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5884 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5885 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5886 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5887 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5888 * reload.c (find_reloads): Likewise.
5889 * reload1.c (alter_reg): Likewise.
5890 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5891 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5892 * tree-if-conv.c (predicate_mem_writes): Likewise.
5893 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5894 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5895 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5896 * valtrack.c (dead_debug_insert_temp): Likewise.
5897 * varasm.c (mergeable_constant_section): Likewise.
5898 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5900 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5901 Alan Hayward <alan.hayward@arm.com>
5902 David Sherwood <david.sherwood@arm.com>
5904 * expr.c (expand_assignment): Cope with polynomial mode sizes
5905 when assigning to a CONCAT.
5907 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5908 Alan Hayward <alan.hayward@arm.com>
5909 David Sherwood <david.sherwood@arm.com>
5911 * machmode.h (mode_precision): Change from unsigned short to
5913 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5915 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5916 or if measurement_type is not polynomial.
5917 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5918 in which the mode is already known to be a scalar_int_mode.
5919 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5920 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5922 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5923 for GET_MODE_PRECISION.
5924 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5925 for GET_MODE_PRECISION.
5926 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5928 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5929 (expand_field_assignment, make_extraction): Likewise.
5930 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5931 (get_last_value): Likewise.
5932 * convert.c (convert_to_integer_1): Likewise.
5933 * cse.c (cse_insn): Likewise.
5934 * expr.c (expand_expr_real_1): Likewise.
5935 * lra-constraints.c (simplify_operand_subreg): Likewise.
5936 * optabs-query.c (can_atomic_load_p): Likewise.
5937 * optabs.c (expand_atomic_load): Likewise.
5938 (expand_atomic_store): Likewise.
5939 * ree.c (combine_reaching_defs): Likewise.
5940 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5941 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5942 * tree.h (type_has_mode_precision_p): Likewise.
5943 * ubsan.c (instrument_si_overflow): Likewise.
5945 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5946 Alan Hayward <alan.hayward@arm.com>
5947 David Sherwood <david.sherwood@arm.com>
5949 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5950 polynomial numbers of units.
5951 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5952 (valid_vector_subparts_p): New function.
5953 (build_vector_type): Remove temporary shim and take the number
5954 of units as a poly_uint64 rather than an int.
5955 (build_opaque_vector_type): Take the number of units as a
5956 poly_uint64 rather than an int.
5957 * tree.c (build_vector_from_ctor): Handle polynomial
5958 TYPE_VECTOR_SUBPARTS.
5959 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5960 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5961 (build_vector_from_val): If the number of units is variable,
5962 use build_vec_duplicate_cst for constant operands and
5963 VEC_DUPLICATE_EXPR otherwise.
5964 (make_vector_type): Remove temporary is_constant ().
5965 (build_vector_type, build_opaque_vector_type): Take the number of
5966 units as a poly_uint64 rather than an int.
5967 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5969 * cfgexpand.c (expand_debug_expr): Likewise.
5970 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5971 (store_constructor, expand_expr_real_1): Likewise.
5972 (const_scalar_mask_from_tree): Likewise.
5973 * fold-const-call.c (fold_const_reduction): Likewise.
5974 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5975 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5976 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5977 (fold_relational_const): Likewise.
5978 (native_interpret_vector): Likewise. Change the size from an
5979 int to an unsigned int.
5980 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5981 TYPE_VECTOR_SUBPARTS.
5982 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5983 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5984 duplicating a non-constant operand into a variable-length vector.
5985 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5986 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5987 * ipa-icf.c (sem_variable::equals): Likewise.
5988 * match.pd: Likewise.
5989 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5990 * print-tree.c (print_node): Likewise.
5991 * stor-layout.c (layout_type): Likewise.
5992 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5993 * tree-cfg.c (verify_gimple_comparison): Likewise.
5994 (verify_gimple_assign_binary): Likewise.
5995 (verify_gimple_assign_ternary): Likewise.
5996 (verify_gimple_assign_single): Likewise.
5997 * tree-pretty-print.c (dump_generic_node): Likewise.
5998 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5999 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
6000 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
6001 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
6002 (vect_shift_permute_load_chain): Likewise.
6003 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
6004 (expand_vector_condition, optimize_vector_constructor): Likewise.
6005 (lower_vec_perm, get_compute_type): Likewise.
6006 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6007 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
6008 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
6009 (vect_recog_mask_conversion_pattern): Likewise.
6010 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6011 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6012 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6013 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6014 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6015 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6016 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6017 (supportable_widening_operation): Likewise.
6018 (supportable_narrowing_operation): Likewise.
6019 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6021 * varasm.c (output_constant): Likewise.
6023 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6024 Alan Hayward <alan.hayward@arm.com>
6025 David Sherwood <david.sherwood@arm.com>
6027 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6028 so that both the length == 3 and length != 3 cases set up their
6029 own permute vectors. Add comments explaining why we know the
6030 number of elements is constant.
6031 (vect_permute_load_chain): Likewise.
6033 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6034 Alan Hayward <alan.hayward@arm.com>
6035 David Sherwood <david.sherwood@arm.com>
6037 * machmode.h (mode_nunits): Change from unsigned char to
6039 (ONLY_FIXED_SIZE_MODES): New macro.
6040 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6041 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6042 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6044 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6045 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6046 or if measurement_type is not polynomial.
6047 * genmodes.c (ZERO_COEFFS): New macro.
6048 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6050 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6051 Use ZERO_COEFFS when emitting initializers.
6052 * data-streamer.h (bp_pack_poly_value): New function.
6053 (bp_unpack_poly_value): Likewise.
6054 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6055 for GET_MODE_NUNITS.
6056 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6057 for GET_MODE_NUNITS.
6058 * tree.c (make_vector_type): Remove temporary shim and make
6059 the real function take the number of units as a poly_uint64
6061 (build_vector_type_for_mode): Handle polynomial nunits.
6062 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6063 * emit-rtl.c (const_vec_series_p_1): Likewise.
6064 (gen_rtx_CONST_VECTOR): Likewise.
6065 * fold-const.c (test_vec_duplicate_folding): Likewise.
6066 * genrecog.c (validate_pattern): Likewise.
6067 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6068 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6069 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6070 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6071 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6072 * rtlanal.c (subreg_get_info): Likewise.
6073 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6074 (vect_grouped_load_supported): Likewise.
6075 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6076 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6077 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6078 (simplify_const_unary_operation, simplify_binary_operation_1)
6079 (simplify_const_binary_operation, simplify_ternary_operation)
6080 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6081 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6082 instead of CONST_VECTOR_NUNITS.
6083 * varasm.c (output_constant_pool_2): Likewise.
6084 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6085 explicit-encoded elements in the XVEC for variable-length vectors.
6087 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6089 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6091 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6092 Alan Hayward <alan.hayward@arm.com>
6093 David Sherwood <david.sherwood@arm.com>
6095 * coretypes.h (fixed_size_mode): Declare.
6096 (fixed_size_mode_pod): New typedef.
6097 * builtins.h (target_builtins::x_apply_args_mode)
6098 (target_builtins::x_apply_result_mode): Change type to
6099 fixed_size_mode_pod.
6100 * builtins.c (apply_args_size, apply_result_size, result_vector)
6101 (expand_builtin_apply_args_1, expand_builtin_apply)
6102 (expand_builtin_return): Update accordingly.
6104 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6106 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6107 * cselib.c (cselib_hash_rtx): Likewise.
6108 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6109 CONST_VECTOR encoding.
6111 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6112 Jeff Law <law@redhat.com>
6115 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6116 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6117 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6118 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6121 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6122 explicitly probe *sp in a noreturn function if there were any callee
6123 register saves or frame pointer is needed.
6125 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6128 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6129 BLKmode for ternary, binary or unary expressions.
6132 * var-tracking.c (delete_vta_debug_insn): New inline function.
6133 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6134 insns from get_insns () to NULL instead of each bb separately.
6135 Use delete_vta_debug_insn. No longer static.
6136 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6137 delete_vta_debug_insns callers.
6138 * rtl.h (delete_vta_debug_insns): Declare.
6139 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6140 instead of variable_tracking_main.
6142 2018-01-03 Martin Sebor <msebor@redhat.com>
6144 PR tree-optimization/83603
6145 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6146 arguments past the endof the argument list in functions declared
6147 without a prototype.
6148 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6149 Avoid checking when arguments are null.
6151 2018-01-03 Martin Sebor <msebor@redhat.com>
6154 * doc/extend.texi (attribute const): Fix a typo.
6155 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6156 issuing -Wsuggest-attribute for void functions.
6158 2018-01-03 Martin Sebor <msebor@redhat.com>
6160 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6161 offset_int::from instead of wide_int::to_shwi.
6162 (maybe_diag_overlap): Remove assertion.
6163 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6164 * gimple-ssa-sprintf.c (format_directive): Same.
6165 (parse_directive): Same.
6166 (sprintf_dom_walker::compute_format_length): Same.
6167 (try_substitute_return_value): Same.
6169 2018-01-03 Jeff Law <law@redhat.com>
6172 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6173 non-constant residual for zero at runtime and avoid probing in
6174 that case. Reorganize code for trailing problem to mirror handling
6177 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6179 PR tree-optimization/83501
6180 * tree-ssa-strlen.c (get_string_cst): New.
6181 (handle_char_store): Call get_string_cst.
6183 2018-01-03 Martin Liska <mliska@suse.cz>
6185 PR tree-optimization/83593
6186 * tree-ssa-strlen.c: Include tree-cfg.h.
6187 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6188 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6189 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6191 (strlen_dom_walker::before_dom_children): Call
6192 gimple_purge_dead_eh_edges. Dump tranformation with details
6194 (strlen_dom_walker::before_dom_children): Update call by adding
6195 new argument cleanup_eh.
6196 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6198 2018-01-03 Martin Liska <mliska@suse.cz>
6201 * cif-code.def (VARIADIC_THUNK): New enum value.
6202 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6205 2018-01-03 Jan Beulich <jbeulich@suse.com>
6207 * sse.md (mov<mode>_internal): Tighten condition for when to use
6208 vmovdqu<ssescalarsize> for TI and OI modes.
6210 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6212 Update copyright years.
6214 2018-01-03 Martin Liska <mliska@suse.cz>
6217 * ipa-visibility.c (function_and_variable_visibility): Skip
6218 functions with noipa attribure.
6220 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6222 * gcc.c (process_command): Update copyright notice dates.
6223 * gcov-dump.c (print_version): Ditto.
6224 * gcov.c (print_version): Ditto.
6225 * gcov-tool.c (print_version): Ditto.
6226 * gengtype.c (create_file): Ditto.
6227 * doc/cpp.texi: Bump @copying's copyright year.
6228 * doc/cppinternals.texi: Ditto.
6229 * doc/gcc.texi: Ditto.
6230 * doc/gccint.texi: Ditto.
6231 * doc/gcov.texi: Ditto.
6232 * doc/install.texi: Ditto.
6233 * doc/invoke.texi: Ditto.
6235 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6237 * vector-builder.h (vector_builder::m_full_nelts): Change from
6238 unsigned int to poly_uint64.
6239 (vector_builder::full_nelts): Update prototype accordingly.
6240 (vector_builder::new_vector): Likewise.
6241 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6242 (vector_builder::operator ==): Likewise.
6243 (vector_builder::finalize): Likewise.
6244 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6245 Take the number of elements as a poly_uint64 rather than an
6247 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6248 from unsigned int to poly_uint64.
6249 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6250 (vec_perm_indices::new_vector): Likewise.
6251 (vec_perm_indices::length): Likewise.
6252 (vec_perm_indices::nelts_per_input): Likewise.
6253 (vec_perm_indices::input_nelts): Likewise.
6254 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6255 number of elements per input as a poly_uint64 rather than an
6256 unsigned int. Use the original encoding for variable-length
6257 vectors, rather than clamping each individual element.
6258 For the second and subsequent elements in each pattern,
6259 clamp the step and base before clamping their sum.
6260 (vec_perm_indices::series_p): Handle polynomial element counts.
6261 (vec_perm_indices::all_in_range_p): Likewise.
6262 (vec_perm_indices_to_tree): Likewise.
6263 (vec_perm_indices_to_rtx): Likewise.
6264 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6265 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6266 (tree_vector_builder::new_binary_operation): Handle polynomial
6267 element counts. Return false if we need to know the number
6268 of elements at compile time.
6269 * fold-const.c (fold_vec_perm): Punt if the number of elements
6270 isn't known at compile time.
6272 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6274 * vec-perm-indices.h (vec_perm_builder): Change element type
6275 from HOST_WIDE_INT to poly_int64.
6276 (vec_perm_indices::element_type): Update accordingly.
6277 (vec_perm_indices::clamp): Handle polynomial element_types.
6278 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6279 (vec_perm_indices::all_in_range_p): Likewise.
6280 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6282 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6283 polynomial vec_perm_indices element types.
6284 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6285 * fold-const.c (fold_vec_perm): Likewise.
6286 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6287 * tree-vect-generic.c (lower_vec_perm): Likewise.
6288 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6289 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6290 element type to HOST_WIDE_INT.
6292 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6293 Alan Hayward <alan.hayward@arm.com>
6294 David Sherwood <david.sherwood@arm.com>
6296 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6297 rather than an int. Use plus_constant.
6298 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6299 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6301 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6302 Alan Hayward <alan.hayward@arm.com>
6303 David Sherwood <david.sherwood@arm.com>
6305 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6306 a HOST_WIDE_INT to a poly_int64.
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 * calls.c (load_register_parameters): Cope with polynomial
6313 mode sizes. Require a constant size for BLKmode parameters
6314 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6315 forces a parameter to be padded at the lsb end in order to
6316 fill a complete number of words, require the parameter size
6317 to be ordered wrt UNITS_PER_WORD.
6319 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6320 Alan Hayward <alan.hayward@arm.com>
6321 David Sherwood <david.sherwood@arm.com>
6323 * reload1.c (spill_stack_slot_width): Change element type
6324 from unsigned int to poly_uint64_pod.
6325 (alter_reg): Treat mode sizes as polynomial.
6327 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6328 Alan Hayward <alan.hayward@arm.com>
6329 David Sherwood <david.sherwood@arm.com>
6331 * reload.c (complex_word_subreg_p): New function.
6332 (reload_inner_reg_of_subreg, push_reload): Use it.
6334 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6335 Alan Hayward <alan.hayward@arm.com>
6336 David Sherwood <david.sherwood@arm.com>
6338 * lra-constraints.c (process_alt_operands): Reject matched
6339 operands whose sizes aren't ordered.
6340 (match_reload): Refer to this check here.
6342 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6343 Alan Hayward <alan.hayward@arm.com>
6344 David Sherwood <david.sherwood@arm.com>
6346 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6347 that the mode size is in the set {1, 2, 4, 8, 16}.
6349 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6350 Alan Hayward <alan.hayward@arm.com>
6351 David Sherwood <david.sherwood@arm.com>
6353 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6354 Use plus_constant instead of gen_rtx_PLUS.
6356 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6357 Alan Hayward <alan.hayward@arm.com>
6358 David Sherwood <david.sherwood@arm.com>
6360 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6361 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6362 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6363 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6364 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6365 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6366 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6367 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6368 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6369 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6371 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6372 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6373 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6374 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6375 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6376 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6377 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6378 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6379 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6380 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6382 * expr.c (emit_move_resolve_push): Treat the input and result
6383 of PUSH_ROUNDING as a poly_int64.
6384 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6385 (emit_push_insn): Likewise.
6386 * lra-eliminations.c (mark_not_eliminable): Likewise.
6387 * recog.c (push_operand): Likewise.
6388 * reload1.c (elimination_effects): Likewise.
6389 * rtlanal.c (nonzero_bits1): Likewise.
6390 * calls.c (store_one_arg): Likewise. Require the padding to be
6391 known at compile time.
6393 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6394 Alan Hayward <alan.hayward@arm.com>
6395 David Sherwood <david.sherwood@arm.com>
6397 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6398 Use plus_constant instead of gen_rtx_PLUS.
6400 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6401 Alan Hayward <alan.hayward@arm.com>
6402 David Sherwood <david.sherwood@arm.com>
6404 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6407 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6408 Alan Hayward <alan.hayward@arm.com>
6409 David Sherwood <david.sherwood@arm.com>
6411 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6412 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6413 via stack temporaries. Treat the mode size as polynomial too.
6415 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6416 Alan Hayward <alan.hayward@arm.com>
6417 David Sherwood <david.sherwood@arm.com>
6419 * expr.c (expand_expr_real_2): When handling conversions involving
6420 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6421 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6422 as a poly_uint64 too.
6424 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6425 Alan Hayward <alan.hayward@arm.com>
6426 David Sherwood <david.sherwood@arm.com>
6428 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6430 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6431 Alan Hayward <alan.hayward@arm.com>
6432 David Sherwood <david.sherwood@arm.com>
6434 * combine.c (can_change_dest_mode): Handle polynomial
6435 REGMODE_NATURAL_SIZE.
6436 * expmed.c (store_bit_field_1): Likewise.
6437 * expr.c (store_constructor): Likewise.
6438 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6439 and polynomial REGMODE_NATURAL_SIZE.
6440 (gen_lowpart_common): Likewise.
6441 * reginfo.c (record_subregs_of_mode): Likewise.
6442 * rtlanal.c (read_modify_subreg_p): Likewise.
6444 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6445 Alan Hayward <alan.hayward@arm.com>
6446 David Sherwood <david.sherwood@arm.com>
6448 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6449 numbers of elements.
6451 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6452 Alan Hayward <alan.hayward@arm.com>
6453 David Sherwood <david.sherwood@arm.com>
6455 * match.pd: Cope with polynomial numbers of vector elements.
6457 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6458 Alan Hayward <alan.hayward@arm.com>
6459 David Sherwood <david.sherwood@arm.com>
6461 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6462 in a POINTER_PLUS_EXPR.
6464 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6465 Alan Hayward <alan.hayward@arm.com>
6466 David Sherwood <david.sherwood@arm.com>
6468 * omp-simd-clone.c (simd_clone_subparts): New function.
6469 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6470 (ipa_simd_modify_function_body): Likewise.
6472 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6473 Alan Hayward <alan.hayward@arm.com>
6474 David Sherwood <david.sherwood@arm.com>
6476 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6477 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6478 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6479 (expand_vector_condition, vector_element): Likewise.
6480 (subparts_gt): New function.
6481 (get_compute_type): Use subparts_gt.
6482 (count_type_subparts): Delete.
6483 (expand_vector_operations_1): Use subparts_gt instead of
6484 count_type_subparts.
6486 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6487 Alan Hayward <alan.hayward@arm.com>
6488 David Sherwood <david.sherwood@arm.com>
6490 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6491 (vect_compile_time_alias): ...this new function. Do the calculation
6492 on poly_ints rather than trees.
6493 (vect_prune_runtime_alias_test_list): Update call accordingly.
6495 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6496 Alan Hayward <alan.hayward@arm.com>
6497 David Sherwood <david.sherwood@arm.com>
6499 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6501 (vect_schedule_slp_instance): Likewise.
6503 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6504 Alan Hayward <alan.hayward@arm.com>
6505 David Sherwood <david.sherwood@arm.com>
6507 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6508 constant and extern definitions for variable-length vectors.
6509 (vect_get_constant_vectors): Note that the number of units
6510 is known to be constant.
6512 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6513 Alan Hayward <alan.hayward@arm.com>
6514 David Sherwood <david.sherwood@arm.com>
6516 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6517 of units as polynomial. Choose between WIDE and NARROW based
6520 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6521 Alan Hayward <alan.hayward@arm.com>
6522 David Sherwood <david.sherwood@arm.com>
6524 * tree-vect-stmts.c (simd_clone_subparts): New function.
6525 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6527 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6528 Alan Hayward <alan.hayward@arm.com>
6529 David Sherwood <david.sherwood@arm.com>
6531 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6532 vectors as polynomial. Use build_index_vector for
6535 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6536 Alan Hayward <alan.hayward@arm.com>
6537 David Sherwood <david.sherwood@arm.com>
6539 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6540 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6541 for variable-length vectors.
6542 (vectorizable_mask_load_store): Treat the number of units as
6543 polynomial, asserting that it is constant if the condition has
6544 already been enforced.
6545 (vectorizable_store, vectorizable_load): Likewise.
6547 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6548 Alan Hayward <alan.hayward@arm.com>
6549 David Sherwood <david.sherwood@arm.com>
6551 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6552 of units as polynomial. Punt if we can't tell at compile time
6553 which vector contains the final result.
6555 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6556 Alan Hayward <alan.hayward@arm.com>
6557 David Sherwood <david.sherwood@arm.com>
6559 * tree-vect-loop.c (vectorizable_induction): Treat the number
6560 of units as polynomial. Punt on SLP inductions. Use an integer
6561 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6562 cast of such a series for variable-length floating-point
6565 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6566 Alan Hayward <alan.hayward@arm.com>
6567 David Sherwood <david.sherwood@arm.com>
6569 * tree.h (build_index_vector): Declare.
6570 * tree.c (build_index_vector): New function.
6571 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6572 of units as polynomial, forcibly converting it to a constant if
6573 vectorizable_reduction has already enforced the condition.
6574 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6575 to create a {1,2,3,...} vector.
6576 (vectorizable_reduction): Treat the number of units as polynomial.
6577 Choose vectype_in based on the largest scalar element size rather
6578 than the smallest number of units. Enforce the restrictions
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-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6586 number of units as polynomial.
6588 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6589 Alan Hayward <alan.hayward@arm.com>
6590 David Sherwood <david.sherwood@arm.com>
6592 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6593 * target.def (autovectorize_vector_sizes): Return the vector sizes
6594 by pointer, using vector_sizes rather than a bitmask.
6595 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6596 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6597 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6599 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6600 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6601 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6602 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6603 * omp-general.c (omp_max_vf): Likewise.
6604 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6605 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6606 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6607 * tree-vect-slp.c (vect_slp_bb): Likewise.
6608 * doc/tm.texi: Regenerate.
6609 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6611 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6612 the vector size as a poly_uint64 rather than an unsigned int.
6613 (current_vector_size): Change from an unsigned int to a poly_uint64.
6614 (get_vectype_for_scalar_type): Update accordingly.
6615 * tree.h (build_truth_vector_type): Take the size and number of
6616 units as a poly_uint64 rather than an unsigned int.
6617 (build_vector_type): Add a temporary overload that takes
6618 the number of units as a poly_uint64 rather than an unsigned int.
6619 * tree.c (make_vector_type): Likewise.
6620 (build_truth_vector_type): Take the number of units as a poly_uint64
6621 rather than an unsigned int.
6623 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6624 Alan Hayward <alan.hayward@arm.com>
6625 David Sherwood <david.sherwood@arm.com>
6627 * target.def (get_mask_mode): Take the number of units and length
6628 as poly_uint64s rather than unsigned ints.
6629 * targhooks.h (default_get_mask_mode): Update accordingly.
6630 * targhooks.c (default_get_mask_mode): Likewise.
6631 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6632 * doc/tm.texi: Regenerate.
6634 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6635 Alan Hayward <alan.hayward@arm.com>
6636 David Sherwood <david.sherwood@arm.com>
6638 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6639 * omp-general.c (omp_max_vf): Likewise.
6640 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6641 (expand_omp_simd): Handle polynomial safelen.
6642 * omp-low.c (omplow_simd_context): Add a default constructor.
6643 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6644 (lower_rec_simd_input_clauses): Update accordingly.
6645 (lower_rec_input_clauses): Likewise.
6647 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6648 Alan Hayward <alan.hayward@arm.com>
6649 David Sherwood <david.sherwood@arm.com>
6651 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6652 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6653 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6654 (vect_analyze_slp_cost): Likewise.
6655 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6656 (vect_model_load_cost): Likewise.
6658 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6659 Alan Hayward <alan.hayward@arm.com>
6660 David Sherwood <david.sherwood@arm.com>
6662 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6663 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6664 from an unsigned int * to a poly_uint64_pod *.
6665 (calculate_unrolling_factor): New function.
6666 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6668 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6669 Alan Hayward <alan.hayward@arm.com>
6670 David Sherwood <david.sherwood@arm.com>
6672 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6673 from an unsigned int to a poly_uint64.
6674 (_loop_vec_info::slp_unrolling_factor): Likewise.
6675 (_loop_vec_info::vectorization_factor): Change from an int
6677 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6678 (vect_get_num_vectors): New function.
6679 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6680 (vect_get_num_copies): Use vect_get_num_vectors.
6681 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6682 to an unsigned int *.
6683 (vect_analyze_data_refs): Change min_vf from an int * to a
6685 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6686 than an unsigned HOST_WIDE_INT.
6687 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6688 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6689 to an unsigned int *.
6690 (vect_analyze_data_ref_dependences): Likewise.
6691 (vect_compute_data_ref_alignment): Handle polynomial vf.
6692 (vect_enhance_data_refs_alignment): Likewise.
6693 (vect_prune_runtime_alias_test_list): Likewise.
6694 (vect_shift_permute_load_chain): Likewise.
6695 (vect_supportable_dr_alignment): Likewise.
6696 (dependence_distance_ge_vf): Take the vectorization factor as a
6697 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6698 (vect_analyze_data_refs): Change min_vf from an int * to a
6700 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6701 vfm1 as a poly_uint64 rather than an int. Make the same change
6702 for the returned bound_scalar.
6703 (vect_gen_vector_loop_niters): Handle polynomial vf.
6704 (vect_do_peeling): Likewise. Update call to
6705 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6706 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6708 * tree-vect-loop.c (vect_determine_vectorization_factor)
6709 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6710 (vect_get_known_peeling_cost): Likewise.
6711 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6712 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6713 (vect_transform_loop): Likewise. Use the lowest possible VF when
6714 updating the upper bounds of the loop.
6715 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6717 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6718 polynomial unroll factors.
6719 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6720 (vect_make_slp_decision): Likewise.
6721 (vect_supported_load_permutation_p): Likewise, and polynomial
6723 (vect_analyze_slp_cost): Handle polynomial vf.
6724 (vect_slp_analyze_node_operations): Likewise.
6725 (vect_slp_analyze_bb_1): Likewise.
6726 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6727 than an unsigned HOST_WIDE_INT.
6728 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6729 (vectorizable_load): Handle polynomial vf.
6730 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6732 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6734 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6735 Alan Hayward <alan.hayward@arm.com>
6736 David Sherwood <david.sherwood@arm.com>
6738 * match.pd: Handle bit operations involving three constants
6739 and try to fold one pair.
6741 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6743 * tree-vect-loop-manip.c: Include gimple-fold.h.
6744 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6745 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6746 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6747 Add a path that uses a step of VF instead of 1, but disable it
6749 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6750 and niters_no_overflow parameters. Update calls to
6751 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6752 Create a new SSA name if the latter choses to use a ste other
6753 than zero, and return it via niters_vector_mult_vf_var.
6754 * tree-vect-loop.c (vect_transform_loop): Update calls to
6755 vect_do_peeling, vect_gen_vector_loop_niters and
6756 slpeel_make_loop_iterate_ntimes.
6757 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6758 (vect_gen_vector_loop_niters): Update declarations after above changes.
6760 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6762 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6763 128-bit round to integer instructions.
6764 (ceil<mode>2): Likewise.
6765 (btrunc<mode>2): Likewise.
6766 (round<mode>2): Likewise.
6768 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6770 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6771 unaligned VSX load/store on P8/P9.
6772 (expand_block_clear): Allow the use of unaligned VSX
6773 load/store on P8/P9.
6775 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6777 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6779 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6780 swap associated with both a load and a store.
6782 2018-01-02 Andrew Waterman <andrew@sifive.com>
6784 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6785 * config/riscv/riscv.md (clear_cache): Use it.
6787 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6789 * web.c: Remove out-of-date comment.
6791 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6793 * expr.c (fixup_args_size_notes): Check that any existing
6794 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6795 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6796 (emit_single_push_insn): ...here.
6798 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6800 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6801 (const_vector_encoded_nelts): New function.
6802 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6803 (const_vector_int_elt, const_vector_elt): Declare.
6804 * emit-rtl.c (const_vector_int_elt_1): New function.
6805 (const_vector_elt): Likewise.
6806 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6807 of CONST_VECTOR_ELT.
6809 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6811 * expr.c: Include rtx-vector-builder.h.
6812 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6813 directly on the tree encoding.
6814 (const_vector_from_tree): Likewise.
6815 * optabs.c: Include rtx-vector-builder.h.
6816 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6817 sequence of "u" values.
6818 * vec-perm-indices.c: Include rtx-vector-builder.h.
6819 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6820 directly on the vec_perm_indices encoding.
6822 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6824 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6825 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6826 * rtx-vector-builder.h: New file.
6827 * rtx-vector-builder.c: Likewise.
6828 * rtl.h (rtx_def::u2): Add a const_vector field.
6829 (CONST_VECTOR_NPATTERNS): New macro.
6830 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6831 (CONST_VECTOR_DUPLICATE_P): Likewise.
6832 (CONST_VECTOR_STEPPED_P): Likewise.
6833 (CONST_VECTOR_ENCODED_ELT): Likewise.
6834 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6835 (unwrap_const_vec_duplicate): Likewise.
6836 (const_vec_series_p): Check for a non-duplicated vector encoding.
6837 Say that the function only returns true for integer vectors.
6838 * emit-rtl.c: Include rtx-vector-builder.h.
6839 (gen_const_vec_duplicate_1): Delete.
6840 (gen_const_vector): Call gen_const_vec_duplicate instead of
6841 gen_const_vec_duplicate_1.
6842 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6843 (gen_const_vec_duplicate): Use rtx_vector_builder.
6844 (gen_const_vec_series): Likewise.
6845 (gen_rtx_CONST_VECTOR): Likewise.
6846 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6847 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6848 Build a new vector rather than modifying a CONST_VECTOR in-place.
6849 (handle_special_swappables): Update call accordingly.
6850 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6851 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6852 Build a new vector rather than modifying a CONST_VECTOR in-place.
6853 (handle_special_swappables): Update call accordingly.
6855 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6857 * simplify-rtx.c (simplify_const_binary_operation): Use
6858 CONST_VECTOR_ELT instead of XVECEXP.
6860 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6862 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6863 the selector elements to be different from the data elements
6864 if the selector is a VECTOR_CST.
6865 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6866 ssizetype for the selector.
6868 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6870 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6871 before testing each element individually.
6872 * tree-vect-generic.c (lower_vec_perm): Likewise.
6874 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6876 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6877 * selftest-run-tests.c (selftest::run_tests): Call it.
6878 * vector-builder.h (vector_builder::operator ==): New function.
6879 (vector_builder::operator !=): Likewise.
6880 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6881 (vec_perm_indices::all_from_input_p): New function.
6882 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6883 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6884 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6885 instead of reading the VECTOR_CST directly. Detect whether both
6886 vector inputs are the same before constructing the vec_perm_indices,
6887 and update the number of inputs argument accordingly. Use the
6888 utility functions added above. Only construct sel2 if we need to.
6890 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6892 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6893 the broadcast of the low byte.
6894 (expand_mult_highpart): Use an explicit encoding for the permutes.
6895 * optabs-query.c (can_mult_highpart_p): Likewise.
6896 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6897 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6898 (vectorizable_bswap): Likewise.
6899 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6900 explicit encoding for the power-of-2 permutes.
6901 (vect_permute_store_chain): Likewise.
6902 (vect_grouped_load_supported): Likewise.
6903 (vect_permute_load_chain): Likewise.
6905 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6907 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6908 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6909 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6910 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6911 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6912 (vect_gen_perm_mask_any): Likewise.
6914 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6916 * int-vector-builder.h: New file.
6917 * vec-perm-indices.h: Include int-vector-builder.h.
6918 (vec_perm_indices): Redefine as an int_vector_builder.
6919 (auto_vec_perm_indices): Delete.
6920 (vec_perm_builder): Redefine as a stand-alone class.
6921 (vec_perm_indices::vec_perm_indices): New function.
6922 (vec_perm_indices::clamp): Likewise.
6923 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6924 (vec_perm_indices::new_vector): New function.
6925 (vec_perm_indices::new_expanded_vector): Update for new
6926 vec_perm_indices class.
6927 (vec_perm_indices::rotate_inputs): New function.
6928 (vec_perm_indices::all_in_range_p): Operate directly on the
6929 encoded form, without computing elided elements.
6930 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6931 encoding. Update for new vec_perm_indices class.
6932 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6933 the given vec_perm_builder.
6934 (expand_vec_perm_var): Update vec_perm_builder constructor.
6935 (expand_mult_highpart): Use vec_perm_builder instead of
6936 auto_vec_perm_indices.
6937 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6938 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6939 or double series encoding as appropriate.
6940 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6941 vec_perm_indices instead of auto_vec_perm_indices.
6942 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6943 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6944 (vect_permute_store_chain): Likewise.
6945 (vect_grouped_load_supported): Likewise.
6946 (vect_permute_load_chain): Likewise.
6947 (vect_shift_permute_load_chain): Likewise.
6948 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6949 (vect_transform_slp_perm_load): Likewise.
6950 (vect_schedule_slp_instance): Likewise.
6951 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6952 (vectorizable_mask_load_store): Likewise.
6953 (vectorizable_bswap): Likewise.
6954 (vectorizable_store): Likewise.
6955 (vectorizable_load): Likewise.
6956 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6957 vec_perm_indices instead of auto_vec_perm_indices. Use
6958 tree_to_vec_perm_builder to read the vector from a tree.
6959 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6960 vec_perm_builder instead of a vec_perm_indices.
6961 (have_whole_vector_shift): Use vec_perm_builder and
6962 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6963 truncation to calc_vec_perm_mask_for_shift.
6964 (vect_create_epilog_for_reduction): Likewise.
6965 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6966 from auto_vec_perm_indices to vec_perm_indices.
6967 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6968 instead of changing individual elements.
6969 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6970 the vector in d.perm.
6971 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6972 from auto_vec_perm_indices to vec_perm_indices.
6973 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6974 instead of changing individual elements.
6975 (arm_vectorize_vec_perm_const): Use new_vector to install
6976 the vector in d.perm.
6977 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6978 Update vec_perm_builder constructor.
6979 (rs6000_expand_interleave): Likewise.
6980 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6981 (rs6000_expand_interleave): Likewise.
6983 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6985 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6986 to qimode could truncate the indices.
6987 * optabs.c (expand_vec_perm_var): Likewise.
6989 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6991 * Makefile.in (OBJS): Add vec-perm-indices.o.
6992 * vec-perm-indices.h: New file.
6993 * vec-perm-indices.c: Likewise.
6994 * target.h (vec_perm_indices): Replace with a forward class
6996 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6997 * optabs.h: Include vec-perm-indices.h.
6998 (expand_vec_perm): Delete.
6999 (selector_fits_mode_p, expand_vec_perm_var): Declare.
7000 (expand_vec_perm_const): Declare.
7001 * target.def (vec_perm_const_ok): Replace with...
7002 (vec_perm_const): ...this new hook.
7003 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
7004 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
7005 * doc/tm.texi: Regenerate.
7006 * optabs.def (vec_perm_const): Delete.
7007 * doc/md.texi (vec_perm_const): Likewise.
7008 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
7009 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7010 expand_vec_perm for constant permutation vectors. Assert that
7011 the mode of variable permutation vectors is the integer equivalent
7012 of the mode that is being permuted.
7013 * optabs-query.h (selector_fits_mode_p): Declare.
7014 * optabs-query.c: Include vec-perm-indices.h.
7015 (selector_fits_mode_p): New function.
7016 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7017 is defined, instead of checking whether the vec_perm_const_optab
7018 exists. Use targetm.vectorize.vec_perm_const instead of
7019 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7020 fit in the vector mode before using a variable permute.
7021 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7022 vec_perm_indices instead of an rtx.
7023 (expand_vec_perm): Replace with...
7024 (expand_vec_perm_const): ...this new function. Take the selector
7025 as a vec_perm_indices rather than an rtx. Also take the mode of
7026 the selector. Update call to shift_amt_for_vec_perm_mask.
7027 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7028 Use vec_perm_indices::new_expanded_vector to expand the original
7029 selector into bytes. Check whether the indices fit in the vector
7030 mode before using a variable permute.
7031 (expand_vec_perm_var): Make global.
7032 (expand_mult_highpart): Use expand_vec_perm_const.
7033 * fold-const.c: Includes vec-perm-indices.h.
7034 * tree-ssa-forwprop.c: Likewise.
7035 * tree-vect-data-refs.c: Likewise.
7036 * tree-vect-generic.c: Likewise.
7037 * tree-vect-loop.c: Likewise.
7038 * tree-vect-slp.c: Likewise.
7039 * tree-vect-stmts.c: Likewise.
7040 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7042 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7043 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7044 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7045 (aarch64_vectorize_vec_perm_const): ...this new function.
7046 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7047 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7048 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7049 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7050 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7051 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7052 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7054 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7055 check for NEON modes.
7056 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7057 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7058 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7059 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7061 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7062 the old VEC_PERM_CONST conditions.
7063 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7064 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7065 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7066 (ia64_vectorize_vec_perm_const_ok): Merge into...
7067 (ia64_vectorize_vec_perm_const): ...this new function.
7068 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7069 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7070 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7071 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7072 * config/mips/mips.c (mips_expand_vec_perm_const)
7073 (mips_vectorize_vec_perm_const_ok): Merge into...
7074 (mips_vectorize_vec_perm_const): ...this new function.
7075 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7076 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7077 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7078 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7079 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7080 (rs6000_expand_vec_perm_const): Delete.
7081 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7083 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7084 (altivec_expand_vec_perm_const_le): Take each operand individually.
7085 Operate on constant selectors rather than rtxes.
7086 (altivec_expand_vec_perm_const): Likewise. Update call to
7087 altivec_expand_vec_perm_const_le.
7088 (rs6000_expand_vec_perm_const): Delete.
7089 (rs6000_vectorize_vec_perm_const_ok): Delete.
7090 (rs6000_vectorize_vec_perm_const): New function.
7091 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7092 an element count and rtx array.
7093 (rs6000_expand_extract_even): Update call accordingly.
7094 (rs6000_expand_interleave): Likewise.
7095 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7096 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7097 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7098 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7099 (rs6000_expand_vec_perm_const): Delete.
7100 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7101 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7102 (altivec_expand_vec_perm_const_le): Take each operand individually.
7103 Operate on constant selectors rather than rtxes.
7104 (altivec_expand_vec_perm_const): Likewise. Update call to
7105 altivec_expand_vec_perm_const_le.
7106 (rs6000_expand_vec_perm_const): Delete.
7107 (rs6000_vectorize_vec_perm_const_ok): Delete.
7108 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7109 reference to the SPE evmerge intructions.
7110 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7111 an element count and rtx array.
7112 (rs6000_expand_extract_even): Update call accordingly.
7113 (rs6000_expand_interleave): Likewise.
7114 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7115 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7117 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7119 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7121 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7122 vector mode and that that mode matches the mode of the data
7124 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7125 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7126 directly using expand_vec_perm_1 when forcing selectors into
7128 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7130 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7132 * optabs-query.h (can_vec_perm_p): Delete.
7133 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7134 * optabs-query.c (can_vec_perm_p): Split into...
7135 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7136 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7137 particular selector is valid.
7138 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7139 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7140 (vect_grouped_load_supported): Likewise.
7141 (vect_shift_permute_load_chain): Likewise.
7142 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7143 (vect_transform_slp_perm_load): Likewise.
7144 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7145 (vectorizable_bswap): Likewise.
7146 (vect_gen_perm_mask_checked): Likewise.
7147 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7148 implementations of variable permutation vectors into account
7149 when deciding which selector to use.
7150 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7151 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7152 with a false third argument.
7153 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7154 to test whether the constant selector is valid and can_vec_perm_var_p
7155 to test whether a variable selector is valid.
7157 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7159 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7160 * optabs-query.c (can_vec_perm_p): Likewise.
7161 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7162 instead of vec_perm_indices.
7163 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7164 (vect_gen_perm_mask_checked): Likewise,
7165 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7166 (vect_gen_perm_mask_checked): Likewise,
7168 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7170 * optabs-query.h (qimode_for_vec_perm): Declare.
7171 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7172 (qimode_for_vec_perm): ...this new function.
7173 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7175 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7177 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7178 does not have a conditional at the top.
7180 2018-01-02 Richard Biener <rguenther@suse.de>
7182 * ipa-inline.c (big_speedup_p): Fix expression.
7184 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7187 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7190 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7194 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7195 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7196 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7197 cond_taken_branch_cost 3->4.
7199 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7201 PR tree-optimization/83581
7202 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7203 TODO_cleanup_cfg if any changes have been made.
7206 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7207 convert_modes if target mode has the right side, but different mode
7211 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7212 last argument when extracting from CONCAT. If either from_real or
7213 from_imag is NULL, use expansion through memory. If result is not
7214 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7215 the parts directly to inner mode, if even that fails, use expansion
7219 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7220 check for bswap in mode rather than HImode and use that in expand_unop
7223 Copyright (C) 2018 Free Software Foundation, Inc.
7225 Copying and distribution of this file, with or without modification,
7226 are permitted in any medium without royalty provided the copyright
7227 notice and this notice are preserved.