1 2018-02-16 Nathan Sidwell <nathan@acm.org>
3 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
5 2018-02-16 Carl Love <cel@us.ibm.com>
7 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
8 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
9 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
10 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
11 expansion to P8V_BUILTIN_VEC_FLOAT2.
13 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
15 PR rtl-optimization/70023
16 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
17 src_regno into account.
19 2018-02-16 Carl Love <cel@us.ibm.com>
21 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
22 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
23 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
24 * config/rs6000/rs6000.c: Remove case statements for
25 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
26 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
27 and P9V_BUILTIN_VEC_VINSERT4B.
28 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
29 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
30 * config/rs6000/vsx.md:
31 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
34 2018-02-16 Carl Love <cel@us.ibm.com>
36 * config/rs6000/altivec.h: Add builtin names vec_extract4b
38 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
40 * config/rs6000/rs6000-c.c: Add the definitions for
41 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
42 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
43 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
44 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
45 definition for insert4b and define insn *insert3b_internal.
46 * doc/extend.texi: Add documentation for vec_extract4b.
48 2018-02-16 Nathan Sidwell <nathan@acm.org>
50 * doc/extend.texi (Backwards Compatibility): Mention friend
51 injection. Note for-scope is deprecated.
52 * doc/invoke.texi (-ffriend-injection): Deprecate.
54 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
56 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
57 that moved to I2, also allow destinations that are a paradoxical
58 subreg (instead of a normal reg).
60 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
63 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
66 2018-02-16 Richard Biener <rguenther@suse.de>
68 PR tree-optimization/84037
69 PR tree-optimization/84016
71 * config/i386/i386.c (ix86_builtin_vectorization_cost):
72 Adjust vec_construct for the fact we need additional higher latency
73 128bit inserts for AVX256 and AVX512 vector builds.
74 (ix86_add_stmt_cost): Scale vector construction cost for
77 2018-02-16 Richard Biener <rguenther@suse.de>
79 PR tree-optimization/84417
80 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
81 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
82 (non_rewritable_lvalue_p): Likewise, use poly-ints.
84 2018-02-16 Martin Liska <mliska@suse.cz>
87 * internal-fn.def (ASAN_CHECK): Set proper flags.
88 (ASAN_MARK): Likewise.
90 2018-02-16 Julia Koval <julia.koval@intel.com>
92 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
95 2018-02-16 Jakub Jelinek <jakub@redhat.com>
98 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
99 Use ++iter rather than iter++ for std::list iterators.
100 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
101 defer deleting them until all nodes in the forest are processed. Do
102 free even leaf nodes. Change to_process into auto_vec.
105 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
106 * vec.h (vec_default_construct): Use memset instead of placement new
107 if BROKEN_VALUE_INITIALIZATION is defined.
108 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
109 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
112 PR rtl-optimization/83723
113 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
114 * lra.c (lra_substitute_pseudo): Likewise. If true, use
115 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
117 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
119 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
121 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
123 PR rtl-optimization/81443
124 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
125 from inner REGs to paradoxical SUBREGs.
127 2018-02-16 Richard Biener <rguenther@suse.de>
129 PR tree-optimization/84399
130 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
131 For operands we can analyze at their definition make sure we can
132 analyze them at each use as well.
134 2018-02-16 Richard Biener <rguenther@suse.de>
136 PR tree-optimization/84190
137 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
138 volatile accesses if the decl isn't volatile.
140 2018-02-15 Jason Merrill <jason@redhat.com>
142 PR c++/84314 - ICE with templates and fastcall attribute.
143 * attribs.c (build_type_attribute_qual_variant): Don't clobber
144 TYPE_CANONICAL on an existing type.
146 2018-02-15 Jakub Jelinek <jakub@redhat.com>
148 PR tree-optimization/84383
149 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
150 dstoff nor call operand_equal_p if dstbase is NULL.
152 PR tree-optimization/84334
153 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
154 also a CONSTANT_CLASS_P, punt.
156 2018-02-14 Jim Wilson <jimw@sifive.com>
158 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
159 first SMALL_OPERAND check. New local min_second_step. Move assert
160 to where locals are set. Add TARGET_RVC support.
161 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
163 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
165 * doc/invoke.texi: Correct -Wformat-overflow code sample.
167 2018-02-14 Martin Sebor <msebor@redhat.com>
169 PR tree-optimization/83698
170 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
171 arrays constrain the offset range to their bounds.
172 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
173 (builtin_access::overlap): Avoid setting the size of overlap if it's
175 (maybe_diag_overlap): Also consider arrays when deciding what values
176 of offsets to include in diagnostics.
178 2018-02-14 Martin Sebor <msebor@redhat.com>
181 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
182 that correspond to the kind of a declaration.
184 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
187 * config/pa/pa.md: Load address of PIC label using the linkage table
188 if the label is nonlocal.
190 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
192 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
193 warning message if user requests -maltivec=be.
194 * doc/invoke.texi: Document deprecation of -maltivec=be.
196 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
199 * config/rs6000/rs6000-c.c: Update definitions for
200 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
201 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
203 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
206 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
207 add _get_ssp intrinsics. Remove argument from
208 __builtin_ia32_rdssp[d|q].
209 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
210 * config/i386/i386-builtin.def: Remove argument from
211 __builtin_ia32_rdssp[d|q].
212 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
213 ix86_expand_special_args_builtin for _rdssp[d|q].
214 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
215 Clear register before usage.
216 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
217 Add documentation for new _get_ssp and _inc_ssp intrinsics.
219 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
221 PR tree-optimization/84357
222 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
223 operand 1 of an ARRAY_REF too.
225 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
228 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
229 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
231 (set_of_reg): New struct.
232 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
233 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
234 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
236 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
237 Split into bitclr, bitset, bitinvert patterns if appropriate.
238 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
239 use rx_fuse_in_memory_bitop.
240 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
241 to named insn, correct maximum insn length.
243 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
246 * machmode.def: Define a complex mode for PARTIAL_INT.
247 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
249 * doc/rtl.texi: Document CSPImode.
250 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
252 (msp430_hard_regno_nregs_with_padding): Likewise.
254 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
257 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
259 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
261 PR rtl-optimization/84169
262 * combine.c (try_combine): New variable split_i2i3. Set it to true if
263 we generated a parallel as new i3 and we split that to new i2 and i3
264 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
265 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
266 those to i2, not i1. Partially rewrite this scan code.
268 2018-02-13 Jakub Jelinek <jakub@redhat.com>
271 * stor-layout.c (place_field): For variable length fields, adjust
272 offset_align afterwards not just based on the field's alignment,
273 but also on the size.
276 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
277 of exps and logs in the use_exp2 case.
279 2018-02-13 Jeff Law <law@redhat.com>
281 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
284 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
287 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
291 * common.opt (gas-loc-support, gas-locview-support): New.
292 (ginline-points, ginternal-reset-location-views): New.
293 * doc/invoke.texi: Document them. Use @itemx where intended.
294 (gvariable-location-views): Adjust.
295 * target.def (reset_location_view): New.
296 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
297 (TARGET_RESET_LOCATION_VIEW): New.
298 * doc/tm.texi: Rebuilt.
299 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
300 (dwarf2out_default_as_locview_support): New.
301 (output_asm_line_debug_info): Use option variables.
302 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
303 (output_loc_list): Likewise.
304 (add_high_low_attributes): Check option variables.
305 Don't output entry view attribute in strict mode.
306 (gen_inlined_subroutine_die): Check option variables.
307 (dwarf2out_inline_entry): Likewise.
308 (init_sections_and_labels): Likewise.
309 (dwarf2out_early_finish): Likewise.
310 (maybe_reset_location_view): New, from...
311 (dwarf2out_var_location): ... here. Call it.
312 * debug.h (dwarf2out_default_as_loc_support): Declare.
313 (dwarf2out_default_as_locview_support): Declare.
314 * hooks.c (hook_int_rtx_insn_0): New.
315 * hooks.h (hook_int_rtx_insn_0): Declare.
316 * toplev.c (process_options): Take -gas-loc-support and
317 -gas-locview-support from dwarf2out. Enable
318 -gvariable-location-views by default only with locview
319 assembler support. Enable -ginternal-reset-location-views by
320 default only if the target defines the corresponding hook.
321 Enable -ginline-points by default if location views are
322 enabled; force it disabled if statement frontiers are
324 * tree-inline.c (expand_call_inline): Check option variables.
325 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
327 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
329 PR tree-optimization/84321
330 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
331 handling. Also check whether the anti-range contains any values
332 that satisfy the mask; switch to a VR_RANGE if not.
334 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
337 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
339 2018-02-13 Martin Jambor <mjambor@suse.cz>
342 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
343 of call statements, also set location of a load to a temporary.
345 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
347 * config/rl78/rl78.c (add_vector_labels): New function.
348 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
349 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
350 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
351 which checks that no arguments are passed.
352 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
353 * doc/extend.texi: Documentation for the new attribute.
355 2018-02-13 Andreas Schwab <schwab@suse.de>
357 * config/riscv/linux.h (CPP_SPEC): Define.
359 2018-02-13 Jakub Jelinek <jakub@redhat.com>
362 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
363 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
364 OPTION_MASK_ISA_AES as first argument to def_builtin_const
365 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
366 instead of OPTION_MASK_ISA_PCLMUL as first argument to
367 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
368 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
369 temporarily for AES and PCLMUL builtins.
371 PR tree-optimization/84339
372 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
373 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
377 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
378 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
379 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
381 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
383 * omp-simd-clone.h: New file.
384 * omp-simd-clone.c: Include omp-simd-clone.h.
385 (expand_simd_clones): No longer static.
386 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
387 cgraph.h and omp-simd-clone.h.
388 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
389 (vect_recog_widen_shift_pattern): Formatting fix.
390 (vect_pattern_recog_1): Don't check optab for calls.
393 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
394 operands[2] into a REG before using gen_lowpart on it.
396 2018-02-12 Jeff Law <law@redhat.com>
399 * config/sh/sh.c (find_barrier): Consider a sibling call
402 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
403 successfully back substituting a reg.
405 2018-02-12 Richard Biener <rguenther@suse.de>
407 PR tree-optimization/84037
408 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
409 parameter, move visited init to caller.
410 (vect_slp_analyze_operations): Separate cost from validity
411 check, initialize visited once for all instances.
412 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
414 * tree-vect-stmts.c (vect_model_simple_cost): Make early
416 (vect_model_promotion_demotion_cost): Likewise.
417 (vectorizable_bswap): Guard cost modeling with !slp_node
418 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
420 (vectorizable_call): Likewise.
421 (vectorizable_conversion): Likewise.
422 (vectorizable_assignment): Likewise.
423 (vectorizable_shift): Likewise.
424 (vectorizable_operation): Likewise.
425 (vectorizable_store): Likewise.
426 (vectorizable_load): Likewise.
427 (vectorizable_condition): Likewise.
428 (vectorizable_comparison): Likewise.
430 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
433 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
434 (ASAN_MARK): Fix fnspec to account for return value, change pointer
435 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
437 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
440 * params.def (inline-min-speedup): Increase from 8 to 15.
441 (max-inline-insns-auto): Decrease from 40 to 30.
442 * ipa-split.c (consider_split): Add some buffer for function to
443 be considered inlining candidate.
444 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
447 2018-02-12 Richard Biener <rguenther@suse.de>
449 PR tree-optimization/84037
450 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
451 matched stmts if we cannot swap the non-matched ones.
453 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
455 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
456 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
457 _mm_maskz_scalef_round_ss): New intrinsics.
458 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
459 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
460 __builtin_ia32_scalefss_round): Remove.
461 (__builtin_ia32_scalefsd_mask_round,
462 __builtin_ia32_scalefss_mask_round): New intrinsics.
463 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
464 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
465 ((match_operand:VF_128 2 "<round_nimm_predicate>"
466 "<round_constraint>")): Changed to ...
467 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
468 "<round_scalar_constraint>")): ... this.
469 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
470 %0, %1, %2<round_op3>}"): Changed to ...
471 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
472 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
473 %2<round_scalar_mask_op3>}"): ... this.
474 * config/i386/subst.md (round_scalar_nimm_predicate): New.
476 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
478 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
479 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
480 (_mm_maskz_sqrt_round_ss): New intrinsics.
481 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
482 (__builtin_ia32_sqrtsd_mask_round)
483 (__builtin_ia32_sqrtss_mask_round): New builtins.
484 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
485 (__builtin_ia32_sqrtss_round): Remove.
486 (__builtin_ia32_sqrtsd_mask_round)
487 (__builtin_ia32_sqrtss_mask_round): New builtins.
488 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
489 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
490 ((match_operand:VF_128 1 "vector_operand"
491 "xBm,<round_constraint>")): Changed to ...
492 ((match_operand:VF_128 1 "vector_operand"
493 "xBm,<round_scalar_constraint>")): ... this.
494 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
495 %0, %2, %<iptr>1<round_op3>}): Changed to ...
496 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
497 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
498 %<iptr>1<round_scalar_mask_op3>}): ... this.
499 ((set_attr "prefix" "<round_prefix>")): Changed to ...
500 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
502 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
505 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
506 Cast vec_cmpeq result to correct type.
507 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
508 Cast vec_cmpgt result to correct type.
510 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
512 * final.c (final_scan_insn_1): Renamed from...
513 (final_scan_insn): ... this. New wrapper, to recover
514 seen from the outermost call in recursive ones.
515 * config/sparc/sparc.c (output_return): Drop seen from call.
516 (output_sibcall): Likewise.
517 * config/visium/visium.c (output_branch): Likewise.
519 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
521 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
524 2018-02-10 Alan Modra <amodra@gmail.com>
527 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
528 Specify LR as an input.
530 2018-02-10 Jakub Jelinek <jakub@redhat.com>
533 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
534 remove_member_access_dummy_vars): New functions.
535 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
536 lower_omp_1, execute_lower_omp): Use them.
538 PR rtl-optimization/84308
539 * shrink-wrap.c (spread_components): Release todo vector.
541 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
543 PR rtl-optimization/57193
544 * ira-color.c (struct allocno_color_data): Add member
545 conflict_allocno_hard_prefs.
546 (update_conflict_allocno_hard_prefs): New.
547 (bucket_allocno_compare_func): Add a preference based on
548 conflict_allocno_hard_prefs.
549 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
550 (color_allocnos): Remove a dead code. Initiate
551 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
553 2018-02-09 Jakub Jelinek <jakub@redhat.com>
556 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
557 constraint from =wa to wa. Avoid a subreg on the output operand,
558 instead use a pseudo and subreg it in a move.
559 (p9_xxbrd_<mode>): Changed to ...
560 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
561 (p9_xxbrd_v2df): New expander.
562 (p9_xxbrw_<mode>): Changed to ...
563 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
564 (p9_xxbrw_v4sf): New expander.
566 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
568 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
570 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
573 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
574 multiply in 32-bit mode.
575 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
576 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
579 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
581 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
582 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
583 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
584 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
586 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
589 * dwarf2out.c (is_trivial_indirect_ref): New function.
590 (dwarf2out_late_global_decl): Do not generate a location
591 attribute for variables that have a non-trivial DECL_VALUE_EXPR
592 and that are not defined in the current unit.
594 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
596 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
597 instead of a libcall for UNORDERED.
599 2018-02-09 Tamar Christina <tamar.christina@arm.com>
602 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
603 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
605 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
608 * config/s390/s390.c (s390_set_current_function): Invoke
609 s390_indirect_branch_settings also if fndecl didn't change.
611 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
613 * config/rs6000/rs6000.md (blockage): Set length to zero.
615 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
617 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
619 2018-02-09 Jakub Jelinek <jakub@redhat.com>
622 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
623 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
627 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
628 PARALLEL incoming that failed vt_get_decl_and_offset check.
631 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
632 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
634 (get_variable_section): For decls in named .bss* sections pass true as
635 second argument to bss_initializer_p.
637 2018-02-09 Marek Polacek <polacek@redhat.com>
638 Jakub Jelinek <jakub@redhat.com>
641 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
642 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
643 Sync some changes from cxx_fold_indirect_ref.
645 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
647 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
649 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
650 (BLOCK_INLINE_ENTRY_LABEL): New.
651 (dwarf2out_var_location): Disregard inline entry markers.
652 (inline_entry_data): New struct.
653 (inline_entry_data_hasher): New hashtable type.
654 (inline_entry_data_hasher::hash): New.
655 (inline_entry_data_hasher::equal): New.
656 (inline_entry_data_table): New variable.
657 (add_high_low_attributes): Add DW_AT_entry_pc and
658 DW_AT_GNU_entry_view attributes if a pending entry is found
659 in inline_entry_data_table. Add old entry_pc attribute only
660 if debug nonbinding markers are disabled.
661 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
663 (block_within_block_p, dwarf2out_inline_entry): New.
664 (dwarf2out_finish): Check that no entries remained in
665 inline_entry_data_table.
666 * final.c (reemit_insn_block_notes): Handle inline entry notes.
667 (final_scan_insn, notice_source_line): Likewise.
668 (rest_of_clean_state): Skip inline entry markers.
669 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
671 * gimple.c (gimple_build_debug_inline_entry): New.
672 * gimple.h (enum gimple_debug_subcode): Add
673 GIMPLE_DEBUG_INLINE_ENTRY.
674 (gimple_build_debug_inline_entry): Declare.
675 (gimple_debug_inline_entry_p): New.
676 (gimple_debug_nonbind_marker_p): Adjust.
677 * insn-notes.def (INLINE_ENTRY): New.
678 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
679 inline entry marker notes.
680 (print_insn): Likewise.
681 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
682 (INSN_DEBUG_MARKER_KIND): Likewise.
683 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
684 * tree-inline.c (expand_call_inline): Build and insert
685 debug_inline_entry stmt.
686 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
687 inline entry blocks early, if nonbind markers are enabled.
688 (dump_scope_block): Dump fragment info.
689 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
690 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
691 (gimple_build_debug_inline_entry): New.
692 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
693 Enable/disable inline entry points too.
694 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
695 (DEBUG_INSN): Describe inline entry markers.
697 * common.opt (gvariable-location-views): New.
698 (gvariable-location-views=incompat5): New.
699 * config.in: Rebuilt.
700 * configure: Rebuilt.
701 * configure.ac: Test assembler for view support.
702 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
703 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
704 * dwarf2out.c (var_loc_view): New typedef.
705 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
706 (dwarf2out_locviews_in_attribute): New.
707 (dwarf2out_locviews_in_loclist): New.
708 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
709 (enum dw_line_info_opcode): Add LI_adv_address.
710 (struct dw_line_info_table): Add view.
711 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
712 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
713 (zero_view_p): New variable.
714 (ZERO_VIEW_P): New macro.
715 (output_asm_line_debug_info): New.
716 (struct var_loc_node): Add view.
717 (add_AT_view_list, AT_loc_list): New.
718 (add_var_loc_to_decl): Add view param. Test it against last.
719 (new_loc_list): Add view params. Record them.
720 (AT_loc_list_ptr): Handle loc and view lists.
721 (view_list_to_loc_list_val_node): New.
722 (print_dw_val): Handle dw_val_class_view_list.
723 (size_of_die): Likewise.
724 (value_format): Likewise.
725 (loc_list_has_views): New.
726 (gen_llsym): Set vl_symbol too.
727 (maybe_gen_llsym, skip_loc_list_entry): New.
728 (dwarf2out_maybe_output_loclist_view_pair): New.
729 (output_loc_list): Output view list or entries too.
730 (output_view_list_offset): New.
731 (output_die): Handle dw_val_class_view_list.
732 (output_dwarf_version): New.
733 (output_compilation_unit_header): Use it.
734 (output_skeleton_debug_sections): Likewise.
735 (output_rnglists, output_line_info): Likewise.
736 (output_pubnames, output_aranges): Update version comments.
737 (output_one_line_info_table): Output view numbers in asm comments.
738 (dw_loc_list): Determine current endview, pass it to new_loc_list.
739 Call maybe_gen_llsym.
740 (loc_list_from_tree_1): Adjust.
741 (add_AT_location_description): Create view list attribute if
742 needed, check it's absent otherwise.
743 (convert_cfa_to_fb_loc_list): Adjust.
744 (maybe_emit_file): Call output_asm_line_debug_info for test.
745 (dwarf2out_var_location): Reset views as needed. Precompute
746 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
748 (new_line_info_table): Reset next view.
749 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
750 (dwarf2out_source_line): Likewise. Output view resets and labels to
751 the assembler, or select appropriate line info opcodes.
752 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
753 (optimize_string_length): Catch it. Adjust.
754 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
755 dw_val_class_view_list, and remove it if no longer needed.
756 (hash_loc_list): Hash view numbers.
757 (loc_list_hasher::equal): Compare them.
758 (optimize_location_lists): Check whether a view list symbol is
759 needed, and whether the locview attribute is present, and
760 whether they match. Remove the locview attribute if no longer
762 (index_location_lists): Call skip_loc_list_entry for test.
763 (dwarf2out_finish): Call output_asm_line_debug_info for test.
764 Use output_dwarf_version.
765 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
766 (struct dw_val_node): Add val_view_list.
767 * final.c (SEEN_NEXT_VIEW): New.
768 (set_next_view_needed): New.
769 (clear_next_view_needed): New.
770 (maybe_output_next_view): New.
771 (final_start_function): Rename to...
772 (final_start_function_1): ... this. Take pointer to FIRST,
773 add SEEN parameter. Emit param bindings in the initial view.
774 (final_start_function): Reintroduce SEEN-less interface.
775 (final): Rename to...
776 (final_1): ... this. Take SEEN parameter. Output final pending
777 next view at the end.
778 (final): Reintroduce seen-less interface.
779 (final_scan_insn): Output pending next view before switching
780 sections or ending a block. Mark the next view as needed when
781 outputting variable locations. Notify debug backend of section
782 changes, and of location view changes.
783 (rest_of_handle_final): Adjust.
784 * toplev.c (process_options): Autodetect value for debug variable
785 location views option. Warn on incompat5 without -gdwarf-5.
786 * doc/invoke.texi (gvariable-location-views): New.
787 (gvariable-location-views=incompat5): New.
788 (gno-variable-location-views): New.
790 2018-02-08 David Malcolm <dmalcolm@redhat.com>
792 PR tree-optimization/84136
793 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
794 that the result of find_edge is non-NULL.
796 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
799 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
800 storing integer register in SImode. Fix cost of 256 and 512
801 byte aligned SSE register store.
803 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
805 * config/i386/i386.c (ix86_multiplication_cost): Fix
806 multiplication cost for TARGET_AVX512DQ.
808 2018-02-08 Marek Polacek <polacek@redhat.com>
810 PR tree-optimization/84238
811 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
814 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
816 PR tree-optimization/84265
817 * tree-vect-stmts.c (vectorizable_store): Don't treat
818 VMAT_CONTIGUOUS accesses as grouped.
819 (vectorizable_load): Likewise.
821 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
823 PR tree-optimization/81635
824 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
825 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
826 (test_round_for_mask): New functions.
827 (wide_int_cc_tests): Call test_round_for_mask.
828 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
829 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
830 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
831 range returned by get_range_info.
833 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
836 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
837 * symtab.c: Include builtins.h
838 (symtab_node::output_to_lto_symbol_table_p): Move here
839 from lto-streamer-out.c:output_symbol_p.
840 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
841 (output_symbol_p): Move all logic to symtab.c
842 (produce_symtab): Update.
844 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
846 * config/s390/s390-opts.h (enum indirect_branch): Define.
847 * config/s390/s390-protos.h (s390_return_addr_from_memory)
848 (s390_indirect_branch_via_thunk)
849 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
850 (enum s390_indirect_branch_type): Define.
851 * config/s390/s390.c (struct s390_frame_layout, struct
852 machine_function): Remove.
853 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
854 (indirect_branch_table_label_no, indirect_branch_table_name):
856 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
857 (enum s390_indirect_branch_option): Define.
858 (s390_return_addr_from_memory): New function.
859 (s390_handle_string_attribute): New function.
860 (s390_attribute_table): Add new attribute handler.
861 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
862 (s390_indirect_branch_via_thunk): New function.
863 (s390_indirect_branch_via_inline_thunk): New function.
864 (s390_function_ok_for_sibcall): When jumping via thunk disallow
865 sibling call optimization for non z10 compiles.
866 (s390_emit_call): Force indirect branch target to be a single
867 register. Add r1 clobber for non-z10 compiles.
868 (s390_emit_epilogue): Emit return jump via return_use expander.
869 (s390_reorg): Handle JUMP_INSNs as execute targets.
870 (s390_option_override_internal): Perform validity checks for the
871 new command line options.
872 (s390_indirect_branch_attrvalue): New function.
873 (s390_indirect_branch_settings): New function.
874 (s390_set_current_function): Invoke s390_indirect_branch_settings.
875 (s390_output_indirect_thunk_function): New function.
876 (s390_code_end): Implement target hook.
877 (s390_case_values_threshold): Implement target hook.
878 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
880 * config/s390/s390.h (struct s390_frame_layout)
881 (struct machine_function): Move here from s390.c.
882 (TARGET_INDIRECT_BRANCH_NOBP_RET)
883 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
884 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
885 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
886 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
887 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
888 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
889 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
890 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
891 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
892 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
893 (mnemonic attribute): Add values which aren't recognized
895 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
896 pattern for branch conversion. Fix mnemonic attribute.
897 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
898 indirect branch via thunk if requested.
899 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
900 ("*indirect_jump"): Disable for branch conversion using out of
902 ("indirect_jump_via_thunk<mode>_z10")
903 ("indirect_jump_via_thunk<mode>")
904 ("indirect_jump_via_inlinethunk<mode>_z10")
905 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
906 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
907 ("casesi_jump_via_inlinethunk<mode>_z10")
908 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
909 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
910 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
911 ("*indirect2_jump"): Disable for branch conversion.
912 ("casesi_jump"): Turn into expander and expand patterns for branch
914 ("return_use"): New expander.
915 ("*return"): Emit return via thunk and rename it to ...
916 ("*return<mode>"): ... this one.
917 * config/s390/s390.opt: Add new options and and enum for the
920 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
922 * lra-constraints.c (match_reload): Unconditionally use
923 gen_lowpart_SUBREG, rather than selecting between that
924 and equivalent gen_rtx_SUBREG code.
926 2018-02-08 Richard Biener <rguenther@suse.de>
928 PR tree-optimization/84233
929 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
930 changed flag instead of boguously re-using phi_inserted.
932 2018-02-08 Martin Jambor <mjambor@suse.cz>
934 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
935 static local variables.
937 2018-02-08 Richard Biener <rguenther@suse.de>
939 PR tree-optimization/84278
940 * tree-vect-stmts.c (vectorizable_store): When looking for
941 smaller vector types to perform grouped strided loads/stores
942 make sure the mode is supported by the target.
943 (vectorizable_load): Likewise.
945 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
947 * config/aarch64/aarch64.c (aarch64_components_for_bb):
948 Increase LDP/STP opportunities by adding adjacent callee-saves.
950 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
952 PR rtl-optimization/84068
953 PR rtl-optimization/83459
954 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
956 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
958 PR tree-optimization/84224
959 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
960 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
963 2018-02-07 Iain Sandoe <iain@codesourcery.com>
966 * config/rs6000/altivec.md (*restore_world): Remove LR use.
967 * config/rs6000/predicates.md (restore_world_operation): Adjust op
968 count, remove one USE.
970 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
972 * doc/install.texi (Configuration): Document the
973 --with-long-double-format={ibm,ieee} PowerPC configuration
977 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
978 Convert from define_expand to be define_insn_and_split. Rework
979 float/double/_Float128 conversions to QI/HI/SImode to work with
980 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
981 conversions to QI/HImode types did a store and then a load to
982 truncate the value. For conversions to VSX registers, don't split
983 the insn, instead emit the code directly. Use the code iterator
984 any_fix to combine signed and unsigned conversions.
985 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
986 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
987 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
988 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
989 (fix_<mode>di2_hw): Likewise.
990 (fixuns_<mode>di2_hw): Likewise.
991 (fix_<mode>si2_hw): Likewise.
992 (fixuns_<mode>si2_hw): Likewise.
993 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
994 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
995 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
996 fix<uns>_trunc<SFDF:mode>si2_p8.
997 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
999 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1000 (fix<uns>_<mode>_mem): Likewise.
1001 (fctiw<u>z_<mode>_mem): Likewise.
1002 (fix<uns>_<mode>_mem): Likewise.
1003 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1004 the register allocator from doing a direct move to the GPRs to do
1005 a store, and instead use the ISA 3.0 store byte/half-word from
1006 vector register instruction. For IEEE 128-bit floating point,
1007 also optimize stores of 32-bit ints.
1008 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1010 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1012 * genextract.c (push_pathstr_operand): New function to support
1014 (walk_rtx): Call push_pathstr_operand.
1015 (print_path): Support [a-zA-Z].
1017 2018-02-07 Richard Biener <rguenther@suse.de>
1019 PR tree-optimization/84037
1020 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1021 (cse_and_gimplify_to_preheader): Declare.
1022 (vect_get_place_in_interleaving_chain): Likewise.
1023 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1025 (_loop_vec_info::~_loop_vec_info): Delete it.
1026 (cse_and_gimplify_to_preheader): New function.
1027 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1028 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1029 (vectorizable_load): Likewise. For grouped stores always base
1030 the IV on the first element.
1031 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1032 condition before gimplifying.
1034 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1036 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1037 *DIV_EXPR and *MOD_EXPR.
1039 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1042 * config/i386/i386.c (ix86_option_override_internal): Mask out
1043 the CF_SET bit when checking -fcf-protection.
1045 2018-02-07 Tom de Vries <tom@codesourcery.com>
1048 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1051 2018-02-07 Richard Biener <rguenther@suse.de>
1053 PR tree-optimization/84204
1054 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1057 PR tree-optimization/84205
1058 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1059 special-case isl_ast_op_zdiv_r.
1061 PR tree-optimization/84223
1062 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1063 Only add conditions from within the region.
1064 (gather_bbs::after_dom_children): Adjust.
1066 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1069 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1070 * config/avr/avr.md: Only post-reload split REG-REG moves if
1071 either register is REGERAL_REG_P.
1073 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1075 PR tree-optimization/84235
1076 * tree-ssa-scopedtables.c
1077 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1078 if the subtraction is performed in floating point type where NaNs are
1079 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1080 build 1. Formatting fix.
1082 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1085 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1086 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1087 and skip it regardless of bb boundaries. Use CALL_P macro,
1088 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1089 unnecessarily, formatting fix.
1091 2018-02-06 Michael Collison <michael.collison@arm.com>
1093 * config/arm/thumb2.md:
1094 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1095 (*thumb_mov_notscc): Ditto.
1097 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1100 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1101 unsigned_fix, not "s".
1103 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1105 * configure.ac (gcc_fn_eh_frame_ro): New function.
1106 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1107 correct .eh_frame permissions.
1108 * configure: Regenerate.
1110 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1112 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1115 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1117 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1118 Display warning message for -mno-speculate-indirect-jumps.
1120 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1122 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1124 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1126 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1128 PR tree-optimization/84225
1129 * tree-eh.c (find_trapping_overflow): Only call
1130 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1132 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1135 * config/i386/i386.c: Reimplement the check of possible options
1136 -mibt/-mshstk conbination. Change error messages.
1137 * doc/invoke.texi: Fix a typo: remove extra '='.
1139 2018-02-06 Marek Polacek <polacek@redhat.com>
1141 PR tree-optimization/84228
1142 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1144 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1147 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1148 emitted arch directives.
1149 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1150 __ARM_FEATURE_COPROC before changing architectures.
1152 2018-02-06 Richard Biener <rguenther@suse.de>
1154 * config/i386/i386.c (print_reg): Fix typo.
1155 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1157 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1159 * configure: Regenerate.
1161 2018-02-05 Martin Sebor <msebor@redhat.com>
1163 PR tree-optimization/83369
1164 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1167 2018-02-05 Martin Liska <mliska@suse.cz>
1169 * doc/invoke.texi: Cherry-pick upstream r323995.
1171 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1173 * ira.c (ira_init_register_move_cost): Adjust comment.
1175 2018-02-05 Martin Liska <mliska@suse.cz>
1177 PR gcov-profile/84137
1178 * doc/gcov.texi: Fix typo in documentation.
1180 2018-02-05 Martin Liska <mliska@suse.cz>
1182 PR gcov-profile/83879
1183 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1184 using dlopen functionality.
1186 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1188 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1189 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1190 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1191 _mm_maskz_range_round_ss): New intrinsics.
1192 (__builtin_ia32_rangesd128_round)
1193 (__builtin_ia32_rangess128_round): Remove.
1194 (__builtin_ia32_rangesd128_mask_round,
1195 __builtin_ia32_rangess128_mask_round): New builtins.
1196 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1197 __builtin_ia32_rangess128_round): Remove.
1198 (__builtin_ia32_rangesd128_mask_round,
1199 __builtin_ia32_rangess128_mask_round): New builtins.
1200 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1201 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1202 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1203 "<round_saeonly_constraint>")): Changed to ...
1204 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1205 "<round_saeonly_scalar_constraint>")): ... this.
1206 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1207 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1208 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1209 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1210 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1212 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1214 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1216 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1217 Remove all values except native, 8540 and 8548.
1219 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1221 * config/i386/i386.c (ix86_output_function_return): Pass
1222 INVALID_REGNUM, instead of -1, as invalid register number to
1223 indirect_thunk_name and output_indirect_thunk.
1225 2018-02-02 Julia Koval <julia.koval@intel.com>
1227 * config.gcc: Add -march=icelake.
1228 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1229 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1230 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1231 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1232 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1233 (processor_target_table): Add icelake.
1234 (ix86_option_override_internal): Handle new PTAs.
1235 (get_builtin_code_for_version): Handle icelake.
1236 (M_INTEL_COREI7_ICELAKE): New.
1237 (fold_builtin_cpu): Handle icelake.
1238 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1239 * doc/invoke.texi: Add -march=icelake.
1241 2018-02-02 Julia Koval <julia.koval@intel.com>
1243 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1244 to wide_int_bitmask.
1245 * wide-int-bitmask.h: New.
1247 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1250 * config/i386/i386.md: Replace Pmode with word_mode in
1251 builtin_setjmp_setup and builtin_longjmp to support x32.
1253 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1257 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1259 (rs6000_supported_cpu_names): New static variable.
1260 (linux_cpu_translation_table): Likewise.
1261 (elf_platform) <cpu>: Define new static variable and use it.
1262 Translate kernel AT_PLATFORM name to canonical name if needed.
1263 Error if platform name is unknown.
1265 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1268 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1270 2018-02-01 Jeff Law <law@redhat.com>
1273 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1274 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1275 the scratch if RELEASE_VIA_POP is false.
1276 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1277 If we have to save a temporary register, decrement SIZE appropriately.
1278 Pass new arguments to release_scratch_register_on_entry.
1279 (ix86_adjust_stack_and_probe): Likewise.
1280 (ix86_emit_probe_stack_range): Pass new arguments to
1281 release_scratch_register_on_entry.
1283 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1285 PR rtl-optimization/84157
1286 * combine.c (change_zero_ext): Use REG_P predicate in
1287 front of HARD_REGISTER_P predicate.
1289 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1291 * config/avr/avr.c (avr_option_override): Move disabling of
1292 -fdelete-null-pointer-checks to...
1293 * common/config/avr/avr-common.c (avr_option_optimization_table):
1296 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1298 PR tree-optimization/81635
1299 * tree-data-ref.c (split_constant_offset_1): For types that
1300 wrap on overflow, try to use range info to prove that wrapping
1303 2018-02-01 Renlin Li <renlin.li@arm.com>
1306 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1308 (aarch64_register_move_cost): Likewise.
1309 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1311 (REG_CLASS_NAMES): Likewise.
1312 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1313 TAILCALL_ADDR_REGS. Remove IP registers.
1314 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1316 2018-02-01 Richard Biener <rguenther@suse.de>
1318 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1319 for specifying RPO order and allow NULL for that.
1320 * domwalk.c (dom_walker::dom_walker): Likewise.
1321 (dom_walker::walk): Handle NULL RPO order.
1322 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1324 (rewrite_update_dom_walker): Likewise.
1325 (mark_def_dom_walker): Likewise.
1327 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1329 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1330 (aarch64_maybe_expand_sve_subreg_move): Declare.
1331 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1332 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1334 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1335 that are semantically a reverse operation.
1336 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1337 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1338 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1340 (aarch64_can_change_mode_class): For big-endian, forbid changes
1341 between two SVE modes if they have different element sizes.
1343 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1345 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1346 the TImode handling for big-endian targets.
1348 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1350 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1351 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1353 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1354 Remove BSWAP handing for big-endian targets and use the form of
1355 LD1RQ appropariate for the mode.
1357 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1359 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1360 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1363 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1366 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1367 check for operands that need to go through aarch64_sve_reload_be.
1369 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1371 PR tree-optimization/81661
1372 PR tree-optimization/84117
1373 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1374 * tree-eh.c: Include gimplify.h.
1375 (find_trapping_overflow, replace_trapping_overflow,
1376 rewrite_to_non_trapping_overflow): New functions.
1377 * tree-vect-loop.c: Include tree-eh.h.
1378 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1379 * tree-data-ref.c: Include tree-eh.h.
1380 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1382 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1384 PR rtl-optimization/84123
1385 * combine.c (change_zero_ext): Check if hard register satisfies
1386 can_change_dest_mode before calling gen_lowpart_SUBREG.
1388 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1391 * ira.c (ira_init_register_move_cost): Remove assert.
1393 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1395 PR rtl-optimization/84071
1396 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1397 * doc/tm.texi: Regenerate.
1399 2018-01-31 Richard Biener <rguenther@suse.de>
1401 PR tree-optimization/84132
1402 * tree-data-ref.c (analyze_miv_subscript): Properly
1403 check whether evolution_function_is_affine_multivariate_p
1404 before calling gcd_of_steps_may_divide_p.
1406 2018-01-31 Julia Koval <julia.koval@intel.com>
1409 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1410 * config/i386/i386.md (rdpid_rex64) New.
1411 (rdpid): Make 32bit only.
1413 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1416 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1417 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1419 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1422 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1424 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1426 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1428 PR rtl-optimization/84071
1429 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1430 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1432 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1434 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1435 (arc_attribute_table): Add 'aux' attribute.
1436 (arc_in_small_data_p): Consider aux like variables.
1437 (arc_is_aux_reg_p): New function.
1438 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1439 (arc_get_aux_arg): New function.
1440 (prepare_move_operands): Handle aux-register access.
1441 (arc_handle_aux_attribute): New function.
1442 * doc/extend.texi (ARC Variable attributes): Add subsection.
1444 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1446 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1447 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1448 (arc_attribute_table): Add 'uncached' attribute.
1449 (arc_print_operand): Print '.di' flag for uncached memory
1451 (arc_in_small_data_p): Do not consider for small data the uncached
1453 (arc_is_uncached_mem_p): New function.
1454 * config/arc/predicates.md (compact_store_memory_operand): Check
1455 for uncached memory accesses.
1456 (nonvol_nonimm_operand): Likewise.
1457 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1459 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1462 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1463 falign-loops=): Add Optimization flag.
1465 2018-01-30 Jeff Law <law@redhat.com>
1468 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1469 INT_REGISTERS_SAVED. Check it prior to calling
1470 get_scratch_register_on_entry.
1471 (ix86_adjust_stack_and_probe): Similarly.
1472 (ix86_emit_probe_stack_range): Similarly.
1473 (ix86_expand_prologue): Corresponding changes.
1475 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1478 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1479 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1481 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1484 * lra-constraints.c (curr_insn_transform): Process AND in the
1487 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1489 PR rtl-optimization/83986
1490 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1491 dependence against last_pending_memory_flush in addition to
1494 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1496 PR tree-optimization/81611
1497 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1500 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1503 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1506 2018-01-30 Richard Biener <rguenther@suse.de>
1507 Jakub Jelinek <jakub@redhat.com>
1509 PR tree-optimization/84111
1510 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1511 inner loops added during recursion, as they don't have up-to-date
1514 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1517 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1518 (can_inline_edge_by_limits_p): ... here.
1519 (can_early_inline_edge_p, check_callers,
1520 update_caller_keys, update_callee_keys, recursive_inlining,
1521 add_new_edges_to_heap, speculation_useful_p,
1522 inline_small_functions,
1523 inline_small_functions, flatten_function,
1524 inline_to_all_callers_1): Update.
1526 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1528 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1531 2018-01-30 Richard Biener <rguenther@suse.de>
1533 PR tree-optimization/83008
1534 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1535 invariant and constant vector uses in stmts when they need
1538 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1541 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1542 * configure: Regenerate.
1544 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1546 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1548 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1549 Use gen_rtx_REG rather than gen_lowpart.
1551 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1553 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1554 rather than 0 when creating partial subregs.
1556 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1558 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1561 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1564 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1565 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1566 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1567 flags. This restores the settings used before the 2017-07-24.
1568 Turning off pre increment/decrement/modify allows IVOPTS to
1569 optimize DF/SF loops where the index is an int.
1571 2018-01-29 Richard Biener <rguenther@suse.de>
1572 Kelvin Nilsen <kelvin@gcc.gnu.org>
1575 * tree-vect-stmts.c (vectorizable_call): Don't call
1576 targetm.vectorize_builtin_md_vectorized_function if callee is
1579 2018-01-22 Carl Love <cel@us.ibm.com>
1581 * doc/extend.tex: Fix typo in second arg in
1582 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1584 2018-01-29 Richard Biener <rguenther@suse.de>
1586 PR tree-optimization/84086
1587 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1588 (flush_ssaname_freelist): When SSA names were released reset
1589 the SCEV hash table.
1591 2018-01-29 Richard Biener <rguenther@suse.de>
1593 PR tree-optimization/84057
1594 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1595 removed paths when removing edges.
1597 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1599 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1600 -mfunction-return=@var{choice}.
1602 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1605 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1607 (layout::print_source_line): Likewise.
1608 (test_get_line_width_without_trailing_whitespace): Add test cases.
1610 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1613 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1616 2018-01-26 Jim Wilson <jimw@sifive.com>
1618 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1620 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1623 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1625 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1626 and CMP + SUB-immediate -> SUBS.
1628 2018-01-26 Martin Sebor <msebor@redhat.com>
1630 PR tree-optimization/83896
1631 * tree-ssa-strlen.c (get_string_len): Rename...
1632 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1633 Avoid assuming length is constant.
1634 (handle_char_store): Use HOST_WIDE_INT for string length.
1636 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1639 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1640 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1642 2018-01-26 Richard Biener <rguenther@suse.de>
1644 PR rtl-optimization/84003
1645 * dse.c (record_store): Only record redundant stores when
1646 the earlier store aliases at least all accesses the later one does.
1648 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1650 PR rtl-optimization/83985
1651 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1652 REG_CFA_RESTORE insns.
1653 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1654 REG_CFA_RESTORE insns here.
1657 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1658 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1660 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1662 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1663 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1664 (arc_init): Likewise.
1665 (arc_override_options): Likewise.
1666 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1668 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1670 * config/arc/arc.h (TARGET_DBNZ): Define.
1671 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1672 properly set the tune attribute.
1673 (dbnz): Use TARGET_DBNZ guard.
1674 * config/arc/arc.opt (mtune): Add core3 option.
1676 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1678 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1679 recognize new pic like addresses.
1680 (arc_delegitimize_address): Clean up.
1682 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1684 * config/arc/arc-arches.def: Option mrf16 valid for all
1686 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1687 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1688 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1689 * config/arc/arc-tables.opt: Regenerate.
1690 * config/arc/arc.c (arc_conditional_register_usage): Handle
1691 reduced register file case.
1692 (arc_file_start): Set must have build attributes.
1693 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1695 * config/arc/arc.opt (mrf16): Add new option.
1696 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1697 * config/arc/genmultilib.awk: Handle new mrf16 option.
1698 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1699 * config/arc/t-multilib: Regenerate.
1700 * doc/invoke.texi (ARC Options): Document mrf16 option.
1702 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1704 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1705 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1706 (arc_attribute_table): Add 'secure_call' attribute.
1707 (arc_print_operand): Print secure call operand.
1708 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1710 (arc_is_secure_call_p): New function. * config/arc/arc.md
1711 (call_i): Add support for sjli instruction.
1712 (call_value_i): Likewise.
1713 * config/arc/constraints.md (Csc): New constraint.
1715 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1716 John Eric Martin <John.Martin@emmicro-us.com>
1718 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1719 * config/arc/arc.c (_arc_jli_section): New struct.
1720 (arc_jli_section): New type.
1721 (rc_jli_sections): New static variable.
1722 (arc_handle_jli_attribute): New function.
1723 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1724 (arc_file_end): New function.
1725 (TARGET_ASM_FILE_END): Define.
1726 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1727 (arc_add_jli_section): New function.
1728 (jli_call_scan): Likewise.
1729 (arc_reorg): Call jli_call_scan.
1730 (arc_output_addsi): Remove 'S' from printing asm operand.
1731 (arc_is_jli_call_p): New function.
1732 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1734 (movhi_insn): Likewise.
1735 (movsi_insn): Likewise.
1736 (movsi_set_cc_insn): Likewise.
1737 (loadqi_update): Likewise.
1738 (load_zeroextendqisi_update): Likewise.
1739 (load_signextendqisi_update): Likewise.
1740 (loadhi_update): Likewise.
1741 (load_zeroextendhisi_update): Likewise.
1742 (load_signextendhisi_update): Likewise.
1743 (loadsi_update): Likewise.
1744 (loadsf_update): Likewise.
1745 (movsicc_insn): Likewise.
1746 (bset_insn): Likewise.
1747 (bxor_insn): Likewise.
1748 (bclr_insn): Likewise.
1749 (bmsk_insn): Likewise.
1750 (bicsi3_insn): Likewise.
1751 (cmpsi_cc_c_insn): Likewise.
1752 (movsi_ne): Likewise.
1753 (movsi_cond_exec): Likewise.
1754 (clrsbsi2): Likewise.
1765 (call_i): Remove 'S' asm letter, add jli instruction.
1766 (call_value_i): Likewise.
1767 * config/arc/arc.op (mjli-always): New option.
1768 * config/arc/constraints.md (Cji): New constraint.
1769 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1771 (subsf3_fpx): Likewise.
1772 (mulsf3_fpx): Likewise.
1773 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1775 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1777 * doc/invoke.texi (ARC): Document mjli-always option.
1779 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1781 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
1782 avoid addition with 0 and use incw and decw where possible.
1784 2018-01-26 Richard Biener <rguenther@suse.de>
1786 PR tree-optimization/81082
1787 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1788 association if it requires casting to unsigned.
1789 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1790 from fold_plusminus_mult_expr to catch important cases late when
1791 range info is available.
1793 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1795 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1796 * configure.ac (hidden_linkonce): New test.
1797 * configure: Regenerate.
1798 * config.in: Regenerate.
1800 2018-01-26 Julia Koval <julia.koval@intel.com>
1802 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1803 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1804 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1805 _mm_mask_bitshuffle_epi64_mask): Fix type.
1806 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1807 USI_FTYPE_V4DI_V4DI_USI): Remove.
1808 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1809 __builtin_ia32_vpshufbitqmb256_mask,
1810 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1811 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1812 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1814 2018-01-26 Alan Modra <amodra@gmail.com>
1817 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1818 UNSPEC_VBPERMQ. Sort other unspecs.
1820 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1822 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1824 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1827 * predict.c (drop_profile): Do not push/pop cfun; update also
1829 (handle_missing_profiles): Fix logic looking for zero profiles.
1831 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1834 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1835 on functions with #pragma omp declare simd or functions with simd
1837 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1838 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1839 Remove trailing \n from warning_at calls.
1841 2018-01-25 Tom de Vries <tom@codesourcery.com>
1844 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1845 for neutered workers.
1847 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1850 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1851 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1853 2018-01-24 Jeff Law <law@redhat.com>
1856 * i386.c (get_probe_interval): Move to earlier point.
1857 (ix86_compute_frame_layout): If -fstack-clash-protection and
1858 the frame is larger than the probe interval, then use pushes
1859 to save registers rather than reg->mem moves.
1860 (ix86_expand_prologue): Remove conditional for int_registers_saved
1863 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1866 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1867 min/max for never referenced object.
1869 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1872 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1874 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1875 attributes from DECL_ATTRIBUTES (decl) without affecting
1876 DECL_ATTRIBUTES (current_function_decl).
1877 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1878 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1880 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1882 PR tree-optimization/83979
1883 * fold-const.c (fold_comparison): Use constant_boolean_node
1884 instead of boolean_{true,false}_node.
1886 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1888 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1891 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1893 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1894 Simplify the clause that sets the length attribute.
1895 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1896 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1897 clause that sets the length attribute.
1898 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1900 2018-01-24 Tom de Vries <tom@codesourcery.com>
1903 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1904 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1905 Add strict parameter.
1906 (prevent_branch_around_nothing): Insert dummy insn between branch to
1907 label and label with no ptx insn inbetween.
1908 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1910 2018-01-24 Tom de Vries <tom@codesourcery.com>
1913 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1914 for neutered threads in warp.
1915 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1917 2018-01-24 Richard Biener <rguenther@suse.de>
1919 PR tree-optimization/83176
1920 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1923 2018-01-24 Richard Biener <rguenther@suse.de>
1925 PR tree-optimization/82819
1926 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1927 code generating pluses that are no-ops in the target precision.
1929 2018-01-24 Richard Biener <rguenther@suse.de>
1932 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1934 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1936 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1937 to merge probabilities.
1938 * predict.c (probably_never_executed): Also mark as cold functions
1939 with global 0 profile and guessed local profile.
1940 * profile-count.c (profile_probability::combine_with_count): New
1942 * profile-count.h (profile_probability::operator*,
1943 profile_probability::operator*=, profile_probability::operator/,
1944 profile_probability::operator/=): Reduce precision to adjusted
1945 and set value to guessed on contradictory divisions.
1946 (profile_probability::combine_with_freq): Remove.
1947 (profile_probability::combine_wiht_count): Declare.
1948 (profile_count::force_nonzero):: Set to adjusted.
1949 (profile_count::probability_in):: Set quality to adjusted.
1950 * tree-ssa-tail-merge.c (replace_block_by): Use
1953 2018-01-23 Andrew Waterman <andrew@sifive.com>
1954 Jim Wilson <jimw@sifive.com>
1956 * config/riscv/riscv.c (riscv_stack_boundary): New.
1957 (riscv_option_override): Set riscv_stack_boundary. Handle
1958 riscv_preferred_stack_boundary_arg.
1959 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1960 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1961 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1962 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1963 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1964 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1966 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1969 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1970 of struct ix86_frame.
1971 (ix86_expand_epilogue): Likewise. Add a local variable for
1972 the reg_save_offset field in struct ix86_frame.
1974 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1976 PR tree-optimization/82604
1977 * tree-loop-distribution.c (enum partition_kind): New enum item
1978 PKIND_PARTIAL_MEMSET.
1979 (partition_builtin_p): Support above new enum item.
1980 (generate_code_for_partition): Ditto.
1981 (compute_access_range): Differentiate cases that equality can be
1982 proven at all loops, the innermost loops or no loops.
1983 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1984 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1985 (finalize_partitions, distribute_loop): Don't fuse partition of
1986 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1987 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1990 2018-01-23 Martin Liska <mliska@suse.cz>
1992 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1993 order to ignore the predictor.
1994 (PRED_POLYMORPHIC_CALL): Likewise.
1995 (PRED_RECURSIVE_CALL): Likewise.
1997 2018-01-23 Martin Liska <mliska@suse.cz>
1999 * tree-profile.c (tree_profiling): Print function header to
2000 aware reader which function we are working on.
2001 * value-prof.c (gimple_find_values_to_profile): Do not print
2002 not interesting value histograms.
2004 2018-01-23 Martin Liska <mliska@suse.cz>
2006 * profile-count.h (enum profile_quality): Add
2007 profile_uninitialized as the first value. Do not number values
2008 as they are zero based.
2009 (profile_count::verify): Update sanity check.
2010 (profile_probability::verify): Likewise.
2012 2018-01-23 Nathan Sidwell <nathan@acm.org>
2014 * doc/invoke.texi (ffor-scope): Deprecate.
2016 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2018 PR tree-optimization/83510
2019 * domwalk.c (set_all_edges_as_executable): New function.
2020 (dom_walker::dom_walker): Convert bool param
2021 "skip_unreachable_blocks" to enum reachability. Move setup of
2022 edge flags to set_all_edges_as_executable and only do it when
2023 reachability is REACHABLE_BLOCKS.
2024 * domwalk.h (enum dom_walker::reachability): New enum.
2025 (dom_walker::dom_walker): Convert bool param
2026 "skip_unreachable_blocks" to enum reachability.
2027 (set_all_edges_as_executable): New decl.
2028 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2029 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2031 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2032 but converting true to REACHABLE_BLOCKS.
2033 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2035 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2036 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2037 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2039 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2040 if check_all_array_refs will be called.
2042 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2044 * tree.c (selftest::test_location_wrappers): Add more test
2047 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2049 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2050 (selftest::test_bit_in_range): Likewise.
2052 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2055 * doc/sourcebuild.texi (vect_float): Say that the selector
2056 only describes the situation when -funsafe-math-optimizations is on.
2057 (vect_float_strict): Document.
2059 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2061 PR tree-optimization/83965
2062 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2063 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2064 instead of checking only for a reduction.
2065 (vect_recog_widen_sum_pattern): Likewise.
2067 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2069 * predict.c (probably_never_executed): Only use precise profile info.
2070 (compute_function_frequency): Skip after inlining hack since we now
2071 have quality checking.
2073 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2075 * profile-count.h (profile_probability::very_unlikely,
2076 profile_probability::unlikely, profile_probability::even): Set
2077 precision to guessed.
2079 2018-01-23 Richard Biener <rguenther@suse.de>
2081 PR tree-optimization/83963
2082 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2083 Properly terminate dominator walk when crossing the exit edge not
2084 when visiting its source block.
2086 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2089 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2090 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2092 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2094 PR tree-optimization/83957
2095 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2096 semicolon after for body surrounded by braces.
2098 PR tree-optimization/83081
2099 * profile-count.h (profile_probability::split): New method.
2100 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2101 Use profile_probability::split.
2102 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2103 when splitting a single conditional jump into 2.
2105 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2107 PR tree-optimization/69452
2108 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2111 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2113 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2114 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2115 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2117 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2119 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2120 * config/rl78/rl78.md (movdi): New define_expand.
2121 * config/rl78/rl78.c (rl78_split_movdi): New function.
2123 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2126 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2128 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2129 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2130 128-bit to produce an UNSPEC move to get the double word with the
2131 signbit and then a shift directly to do signbit.
2132 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2133 implementation with a new version that just does either a direct
2134 move or a regular move. Move memory interface to separate insns.
2135 Move insns so they are next to the expander.
2136 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2137 with signbit move. Split big and little endian case.
2138 (signbit<mode>2_dm_mem_le): Likewise.
2139 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2140 (signbit<mode>2_dm2): Likewise.
2142 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2144 * config/rl78/rl78.md (anddi3): New define_expand.
2146 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2148 * config/rl78/rl78.md (umindi3): New define_expand.
2150 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2152 * config/rl78/rl78.md (smindi3): New define_expand.
2154 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2156 * config/rl78/rl78.md (smaxdi3): New define_expand.
2158 2018-01-22 Carl Love <cel@us.ibm.com>
2160 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2161 LVX_V1TI): Add macro expansion.
2162 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2163 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2164 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2165 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2166 Change check to determine if the instruction is a byte reversing
2167 entry. Fix typo in comment.
2168 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2169 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2170 Add def_builtin calls for new builtins.
2171 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2172 Add define_insn expansion.
2174 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2176 * config/rl78/rl78.md (umaxdi3): New define_expand.
2178 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2180 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2181 for non-QImode registers.
2183 2018-01-22 Richard Biener <rguenther@suse.de>
2185 PR tree-optimization/83963
2186 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2187 including the loop exit block.
2188 (scop_detection::merge_sese): Likewise.
2189 (scop_detection::add_scop): Do it here instead.
2191 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2193 * doc/sourcebuild.texi (arm_softfloat): Document.
2195 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2198 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2199 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2200 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2202 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2203 David Edelsohn <dje.gcc@gmail.com>
2206 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2207 Change "crset eq" to "crset 2".
2208 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2209 (*call_indirect_aix<mode>_nospec): Likewise.
2210 (*call_value_indirect_aix<mode>_nospec): Likewise.
2211 (*call_indirect_elfv2<mode>_nospec): Likewise.
2212 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2213 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2214 change assembly output from . to $.
2215 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2216 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2217 (*tablejump<mode>_internal1_nospec): Likewise.
2219 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2222 * config/sh/sh_optimize_sett_clrt.cc:
2223 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2225 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2227 PR tree-optimization/83940
2228 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2229 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2230 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2231 use it to pass back the definition type.
2232 (vect_check_store_rhs): Likewise rhs_dt_out.
2233 (vect_build_gather_load_calls): Add a mask_dt argument and use
2234 it instead of a call to vect_is_simple_use.
2235 (vectorizable_store): Update calls to vect_check_load_store_mask
2236 and vect_check_store_rhs. Use the dt returned by the latter instead
2237 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2238 instead of calls to vect_is_simple_use. Pass the scalar rather
2239 than the vector operand to vect_is_simple_use when handling
2240 second and subsequent copies of an rhs value.
2241 (vectorizable_load): Update calls to vect_check_load_store_mask
2242 and vect_build_gather_load_calls. Use the cached mask_dt and
2243 gs_info.offset_dt instead of calls to vect_is_simple_use.
2245 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2248 * tree-emutls.c: Include gimplify.h.
2249 (lower_emutls_2): New function.
2250 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2251 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2252 it before further processing.
2255 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2256 UINTVAL (trueop1) instead of INTVAL (op1).
2258 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2262 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2263 INCOMING_FRAME_SP_OFFSET if not defined.
2264 (scan_trace): Add ENTRY argument. If true and
2265 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2266 emit a note to adjust the CFA offset.
2267 (create_cfi_notes): Adjust scan_trace callers.
2268 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2269 INCOMING_FRAME_SP_OFFSET in the CIE.
2270 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2271 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2273 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2274 * doc/tm.texi: Regenerated.
2276 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2278 PR rtl-optimization/83147
2279 * lra-constraints.c (remove_inheritance_pseudos): Use
2280 lra_substitute_pseudo_within_insn.
2282 2018-01-19 Tom de Vries <tom@codesourcery.com>
2283 Cesar Philippidis <cesar@codesourcery.com>
2286 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2288 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2291 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2292 spaces for function labels.
2294 2018-01-19 Martin Liska <mliska@suse.cz>
2296 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2297 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2298 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2299 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2300 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2301 (PRED_CONST_RETURN): Change from 69 to 65.
2302 (PRED_NULL_RETURN): Change from 91 to 71.
2303 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2304 (PRED_LOOP_GUARD): Change from 66 to 73.
2306 2018-01-19 Martin Liska <mliska@suse.cz>
2308 * predict.c (predict_insn_def): Add new assert.
2309 (struct branch_predictor): Change type to signed integer.
2310 (test_prediction_value_range): Amend test to cover
2312 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2313 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2314 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2315 (PRED_LOOP_IV_COMPARE): Likewise.
2316 * predict.h (PROB_UNINITIALIZED): Define new constant.
2318 2018-01-19 Martin Liska <mliska@suse.cz>
2320 * predict.c (dump_prediction): Add new format for
2321 analyze_brprob.py script which is enabled with -details
2323 * profile-count.h (precise_p): New function.
2325 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2327 PR tree-optimization/83922
2328 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2329 there are no statements that need masking.
2330 (vect_active_double_reduction_p): New function.
2331 (vect_analyze_loop_operations): Use it when handling phis that
2332 are not in the loop header.
2334 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2336 PR tree-optimization/83914
2337 * tree-vect-loop.c (vectorizable_induction): Don't convert
2338 init_expr or apply the peeling adjustment for inductions
2339 that are nested within the vectorized loop.
2341 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2343 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2346 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2350 * function.h (gimplify_parameters): Add gimple_seq * argument.
2351 * function.c: Include gimple.h and options.h.
2352 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2353 for the added local temporaries if needed.
2354 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2355 if there are any parameter cleanups, wrap whole body into a
2356 try/finally with the cleanups.
2358 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2361 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2362 Use GET_MODE_CLASS for scalar floating point.
2364 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2368 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2369 Fix call of call_cgraph_insertion_hooks.
2371 2018-01-18 Martin Sebor <msebor@redhat.com>
2373 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2375 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2378 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2381 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2384 * common.opt: (-ffile-prefix-map): New option.
2385 * opts.c (common_handle_option): Defer it.
2386 * opts-global.c (handle_common_deferred_options): Handle it.
2387 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2388 * file-prefix-map.h: New file.
2389 (remap_debug_filename, add_debug_prefix_map): ...here.
2390 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2391 * final.c (debug_prefix_map, add_debug_prefix_map
2392 remap_debug_filename): Move to...
2393 * file-prefix-map.c: New file.
2394 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2395 generalize, get rid of alloca(), use strrchr() instead of strchr().
2396 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2397 Implement in terms of add_prefix_map().
2398 (remap_macro_filename, remap_debug_filename): Implement in term of
2400 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2401 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2402 * dbxout.c: Include file-prefix-map.h.
2403 * varasm.c: Likewise.
2404 * vmsdbgout.c: Likewise.
2405 * xcoffout.c: Likewise.
2406 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2407 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2408 * doc/invoke.texi (-ffile-prefix-map): Document.
2409 (-fdebug-prefix-map): Update description.
2411 2018-01-18 Martin Liska <mliska@suse.cz>
2413 * config/i386/i386.c (indirect_thunk_name): Document that also
2415 (output_indirect_thunk): Document why both instructions
2416 (pause and lfence) are generated.
2418 2018-01-18 Richard Biener <rguenther@suse.de>
2420 PR tree-optimization/83887
2421 * graphite-scop-detection.c
2422 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2423 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2424 (scop_detection::merge_sese): Re-implement with a flood-fill
2425 algorithm that properly finds a SESE region if it exists.
2427 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2430 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2431 pointer_diff optimizations use view_convert instead of convert.
2433 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2435 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2436 Generate different code for -mno-speculate-indirect-jumps.
2437 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2438 (*call_indirect_aix<mode>): Disable for
2439 -mno-speculate-indirect-jumps.
2440 (*call_indirect_aix<mode>_nospec): New define_insn.
2441 (*call_value_indirect_aix<mode>): Disable for
2442 -mno-speculate-indirect-jumps.
2443 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2444 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2445 -mno-speculate-indirect-jumps.
2446 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2448 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2450 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2451 long double type, set the flags for noting the default long double
2452 type, even if we don't pass or return a long double type.
2454 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2457 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2460 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2462 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2463 support for merge[hl].
2464 (fold_mergehl_helper): New helper function.
2465 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2466 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2467 (altivec_vmrglw_direct): Add xxmrglw insn.
2469 2018-01-17 Andrew Waterman <andrew@sifive.com>
2471 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2472 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2474 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2477 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2478 call the lto_location_cache before reading the
2479 DECL_SOURCE_LOCATION of the types.
2481 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2482 Richard Sandiford <richard.sandiford@linaro.org>
2484 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2485 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2486 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2487 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2488 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2490 * config/aarch64/constraints.md (aarch64_movti_operand):
2492 * config/aarch64/predicates.md (Uti): Add new constraint.
2494 2018-01-17 Carl Love <cel@us.ibm.com>
2496 * config/rs6000/vsx.md (define_expand xl_len_r,
2497 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2498 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2500 (define_expand, define_insn): Move the shift left from the
2501 define_insn to the define_expand for lxvl and stxvl instructions.
2502 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2503 and XL_LEN_R definitions to PURE.
2505 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2507 * config/i386/i386.c (indirect_thunk_name): Declare regno
2508 as unsigned int. Compare regno with INVALID_REGNUM.
2509 (output_indirect_thunk): Ditto.
2510 (output_indirect_thunk_function): Ditto.
2511 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2512 in the call to output_indirect_thunk_function.
2514 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2517 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2518 rather than the size of inner_type to determine the stack slot size
2519 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2521 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2524 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2527 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2529 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2530 endian Linux systems to optionally enable multilibs for selecting
2531 the long double type if the user configured an explicit type.
2532 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2533 have no long double multilibs if not defined.
2534 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2535 warn if the user used -mabi={ieee,ibm}longdouble and we built
2536 multilibs for long double.
2537 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2538 appropriate multilib option.
2539 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2541 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2542 for building long double multilibs.
2543 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2545 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2547 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2550 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2552 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2555 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2558 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2561 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2563 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2564 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2566 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2568 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2569 different rtl trees depending on TARGET_64BIT.
2570 (rs6000_gen_lvx): Likewise.
2572 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2574 * config/visium/visium.md (nop): Tweak comment.
2575 (hazard_nop): Likewise.
2577 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2579 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2580 -mspeculate-indirect-jumps.
2581 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2582 for -mno-speculate-indirect-jumps.
2583 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2584 (*call_value_indirect_elfv2<mode>): Disable for
2585 -mno-speculate-indirect-jumps.
2586 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2587 (indirect_jump): Emit different RTL for
2588 -mno-speculate-indirect-jumps.
2589 (*indirect_jump<mode>): Disable for
2590 -mno-speculate-indirect-jumps.
2591 (*indirect_jump<mode>_nospec): New define_insn.
2592 (tablejump): Emit different RTL for
2593 -mno-speculate-indirect-jumps.
2594 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2595 (tablejumpsi_nospec): New define_expand.
2596 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2597 (tablejumpdi_nospec): New define_expand.
2598 (*tablejump<mode>_internal1): Disable for
2599 -mno-speculate-indirect-jumps.
2600 (*tablejump<mode>_internal1_nospec): New define_insn.
2601 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2604 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2606 * caller-save.c (insert_save): Drop unnecessary parameter. All
2609 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2610 Richard Biener <rguenth@suse.de>
2613 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2614 return early, inline manually is_gimple_sizepos. Make sure if we
2615 call gimplify_expr we don't end up with a gimple constant.
2616 * tree.c (variably_modified_type_p): Don't return true for
2617 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2618 * gimplify.h (is_gimple_sizepos): Remove.
2620 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2622 PR tree-optimization/83857
2623 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2624 vectorizable_live_operation for pure SLP statements.
2625 (vectorizable_live_operation): Handle PHIs.
2627 2018-01-16 Richard Biener <rguenther@suse.de>
2629 PR tree-optimization/83867
2630 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2631 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2633 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2636 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2637 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2638 If off is not INTEGER_CST, issue a may not be aligned warning
2639 rather than isn't aligned. Use isn%'t rather than isn't.
2640 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2642 <case MULT_EXPR>: Improve the case when bottom and one of the
2643 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2644 operand, in that case check if the other operand is multiple of
2645 bottom divided by the INTEGER_CST operand.
2647 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2650 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2651 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2652 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2653 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2654 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2655 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2656 (pa_function_arg_size): New function.
2658 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2660 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2661 in a separate statement.
2663 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2665 PR tree-optimization/83847
2666 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2667 group gathers and scatters.
2669 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2671 PR rtl-optimization/86620
2672 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2674 PR rtl-optimization/83213
2675 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2676 to last if both are JUMP_INSNs.
2678 PR tree-optimization/83843
2679 * gimple-ssa-store-merging.c
2680 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2681 store_immediate_info for bswap/nop orig_stores.
2683 2018-01-15 Andrew Waterman <andrew@sifive.com>
2685 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2687 <UDIV>: Increase cost if !TARGET_DIV.
2689 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2691 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2692 (define_attr "cr_logical_3op"): New.
2693 (cceq_ior_compare): Adjust.
2694 (cceq_ior_compare_complement): Adjust.
2695 (*cceq_rev_compare): Adjust.
2696 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2697 (is_cracked_insn): Adjust.
2698 (insn_must_be_first_in_group): Adjust.
2699 * config/rs6000/40x.md: Adjust.
2700 * config/rs6000/440.md: Adjust.
2701 * config/rs6000/476.md: Adjust.
2702 * config/rs6000/601.md: Adjust.
2703 * config/rs6000/603.md: Adjust.
2704 * config/rs6000/6xx.md: Adjust.
2705 * config/rs6000/7450.md: Adjust.
2706 * config/rs6000/7xx.md: Adjust.
2707 * config/rs6000/8540.md: Adjust.
2708 * config/rs6000/cell.md: Adjust.
2709 * config/rs6000/e300c2c3.md: Adjust.
2710 * config/rs6000/e500mc.md: Adjust.
2711 * config/rs6000/e500mc64.md: Adjust.
2712 * config/rs6000/e5500.md: Adjust.
2713 * config/rs6000/e6500.md: Adjust.
2714 * config/rs6000/mpc.md: Adjust.
2715 * config/rs6000/power4.md: Adjust.
2716 * config/rs6000/power5.md: Adjust.
2717 * config/rs6000/power6.md: Adjust.
2718 * config/rs6000/power7.md: Adjust.
2719 * config/rs6000/power8.md: Adjust.
2720 * config/rs6000/power9.md: Adjust.
2721 * config/rs6000/rs64.md: Adjust.
2722 * config/rs6000/titan.md: Adjust.
2724 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2726 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2727 ix86_indirect_branch_register logic.
2729 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2731 * config/i386/constraints.md (Bs): Update
2732 ix86_indirect_branch_register check. Don't check
2733 ix86_indirect_branch_register with GOT_memory_operand.
2735 * config/i386/predicates.md (GOT_memory_operand): Don't check
2736 ix86_indirect_branch_register here.
2737 (GOT32_symbol_operand): Likewise.
2739 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2741 * config/i386/predicates.md (constant_call_address_operand):
2742 Rewrite ix86_indirect_branch_register logic.
2743 (sibcall_insn_operand): Likewise.
2745 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2747 * config/i386/constraints.md (Bs): Replace
2748 ix86_indirect_branch_thunk_register with
2749 ix86_indirect_branch_register.
2751 * config/i386/i386.md (indirect_jump): Likewise.
2752 (tablejump): Likewise.
2753 (*sibcall_memory): Likewise.
2754 (*sibcall_value_memory): Likewise.
2755 Peepholes of indirect call and jump via memory: Likewise.
2756 * config/i386/i386.opt: Likewise.
2757 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2758 (GOT_memory_operand): Likewise.
2759 (call_insn_operand): Likewise.
2760 (sibcall_insn_operand): Likewise.
2761 (GOT32_symbol_operand): Likewise.
2763 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2766 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2767 type rather than type addr's type points to.
2768 (expand_omp_atomic_mutex): Likewise.
2769 (expand_omp_atomic): Likewise.
2771 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2774 * config/i386/i386.c (output_indirect_thunk_function): Use
2775 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2776 for __x86_return_thunk.
2778 2018-01-15 Richard Biener <rguenther@suse.de>
2781 * expmed.c (extract_bit_field_1): Fix typo.
2783 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2786 * config/arm/iterators.md (VF): New mode iterator.
2787 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2788 Remove integer-related logic from pattern.
2789 (neon_vabd<mode>_3): Likewise.
2791 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2794 * common.opt (fstrict-overflow): No longer an alias.
2795 (fwrapv-pointer): New option.
2796 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2797 also for pointer types based on flag_wrapv_pointer.
2798 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2799 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2800 opts->x_flag_wrapv got set.
2801 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2802 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2803 POINTER_TYPE_OVERFLOW_UNDEFINED.
2804 * match.pd: Likewise in address comparison pattern.
2805 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2807 2018-01-15 Richard Biener <rguenther@suse.de>
2810 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2811 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2812 Reset type names to their identifier if their TYPE_DECL doesn't
2813 have linkage (and thus is used for ODR and devirt).
2814 (save_debug_info_for_decl): Remove.
2815 (save_debug_info_for_type): Likewise.
2816 (add_tree_to_fld_list): Adjust.
2817 * tree-pretty-print.c (dump_generic_node): Make dumping of
2818 type names more robust.
2820 2018-01-15 Richard Biener <rguenther@suse.de>
2822 * BASE-VER: Bump to 8.0.1.
2824 2018-01-14 Martin Sebor <msebor@redhat.com>
2827 * builtins.c (check_access): Avoid warning when the no-warning bit
2830 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2832 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2833 * ira-color (allocno_hard_regs_compare): Likewise.
2835 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2838 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2839 Use .pushsection/.popsection.
2841 2018-01-14 Martin Sebor <msebor@redhat.com>
2844 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2846 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2848 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2849 entry from extra_headers.
2850 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2851 extra_headers, make the list bitwise identical to the i?86-*-* one.
2853 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2855 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2856 -mcmodel=large with -mindirect-branch=thunk,
2857 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2858 -mfunction-return=thunk-extern.
2859 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2860 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2861 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2863 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2865 * config/i386/i386.c (print_reg): Print the name of the full
2866 integer register without '%'.
2867 (ix86_print_operand): Handle 'V'.
2868 * doc/extend.texi: Document 'V' modifier.
2870 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2872 * config/i386/constraints.md (Bs): Disallow memory operand for
2873 -mindirect-branch-register.
2875 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2876 (GOT_memory_operand): Likewise.
2877 (call_insn_operand): Likewise.
2878 (sibcall_insn_operand): Likewise.
2879 (GOT32_symbol_operand): Likewise.
2880 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2881 for -mindirect-branch-register.
2882 (tablejump): Likewise.
2883 (*sibcall_memory): Likewise.
2884 (*sibcall_value_memory): Likewise.
2885 Disallow peepholes of indirect call and jump via memory for
2886 -mindirect-branch-register.
2887 (*call_pop): Replace m with Bw.
2888 (*call_value_pop): Likewise.
2889 (*sibcall_pop_memory): Replace m with Bs.
2890 * config/i386/i386.opt (mindirect-branch-register): New option.
2891 * doc/invoke.texi: Document -mindirect-branch-register option.
2893 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2895 * config/i386/i386-protos.h (ix86_output_function_return): New.
2896 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2897 set function_return_type.
2898 (indirect_thunk_name): Add ret_p to indicate thunk for function
2900 (output_indirect_thunk_function): Pass false to
2901 indirect_thunk_name.
2902 (ix86_output_indirect_branch_via_reg): Likewise.
2903 (ix86_output_indirect_branch_via_push): Likewise.
2904 (output_indirect_thunk_function): Create alias for function
2905 return thunk if regno < 0.
2906 (ix86_output_function_return): New function.
2907 (ix86_handle_fndecl_attribute): Handle function_return.
2908 (ix86_attribute_table): Add function_return.
2909 * config/i386/i386.h (machine_function): Add
2910 function_return_type.
2911 * config/i386/i386.md (simple_return_internal): Use
2912 ix86_output_function_return.
2913 (simple_return_internal_long): Likewise.
2914 * config/i386/i386.opt (mfunction-return=): New option.
2915 (indirect_branch): Mention -mfunction-return=.
2916 * doc/extend.texi: Document function_return function attribute.
2917 * doc/invoke.texi: Document -mfunction-return= option.
2919 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2921 * config/i386/i386-opts.h (indirect_branch): New.
2922 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2923 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2924 with local indirect jump when converting indirect call and jump.
2925 (ix86_set_indirect_branch_type): New.
2926 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2927 (indirectlabelno): New.
2928 (indirect_thunk_needed): Likewise.
2929 (indirect_thunk_bnd_needed): Likewise.
2930 (indirect_thunks_used): Likewise.
2931 (indirect_thunks_bnd_used): Likewise.
2932 (INDIRECT_LABEL): Likewise.
2933 (indirect_thunk_name): Likewise.
2934 (output_indirect_thunk): Likewise.
2935 (output_indirect_thunk_function): Likewise.
2936 (ix86_output_indirect_branch_via_reg): Likewise.
2937 (ix86_output_indirect_branch_via_push): Likewise.
2938 (ix86_output_indirect_branch): Likewise.
2939 (ix86_output_indirect_jmp): Likewise.
2940 (ix86_code_end): Call output_indirect_thunk_function if needed.
2941 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2943 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2944 (ix86_attribute_table): Add indirect_branch.
2945 * config/i386/i386.h (machine_function): Add indirect_branch_type
2946 and has_local_indirect_jump.
2947 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2949 (tablejump): Likewise.
2950 (*indirect_jump): Use ix86_output_indirect_jmp.
2951 (*tablejump_1): Likewise.
2952 (simple_return_indirect_internal): Likewise.
2953 * config/i386/i386.opt (mindirect-branch=): New option.
2954 (indirect_branch): New.
2957 (thunk-inline): Likewise.
2958 (thunk-extern): Likewise.
2959 * doc/extend.texi: Document indirect_branch function attribute.
2960 * doc/invoke.texi: Document -mindirect-branch= option.
2962 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2965 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2967 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2969 * ipa-inline.c (want_inline_small_function_p): Return false if
2970 inlining has already failed with CIF_FINAL_ERROR.
2971 (update_caller_keys): Call want_inline_small_function_p before
2973 (update_callee_keys): Likewise.
2975 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2977 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2979 (rs6000_quadword_masked_address_p): Likewise.
2980 (quad_aligned_load_p): Likewise.
2981 (quad_aligned_store_p): Likewise.
2982 (const_load_sequence_p): Add comment to describe the outer-most loop.
2983 (mimic_memory_attributes_and_flags): New function.
2984 (rs6000_gen_stvx): Likewise.
2985 (replace_swapped_aligned_store): Likewise.
2986 (rs6000_gen_lvx): Likewise.
2987 (replace_swapped_aligned_load): Likewise.
2988 (replace_swapped_load_constant): Capitalize argument name in
2989 comment describing this function.
2990 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2991 and stores that access quad-word aligned addresses and replace
2992 with stvx or lvx instructions when appropriate.
2993 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2994 New function prototype.
2995 (rs6000_quadword_masked_address_p): Likewise.
2996 (rs6000_gen_lvx): Likewise.
2997 (rs6000_gen_stvx): Likewise.
2998 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2999 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3000 when memory address is aligned.
3001 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3002 this split to select lvx instruction when memory address is aligned.
3003 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3004 instruction when memory address is aligned.
3005 (*vsx_le_perm_load_v16qi): Likewise.
3006 (four unnamed splitters): Modify to select the stvx instruction
3007 when memory is aligned.
3009 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3011 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3012 which appears in the queue multiple times.
3014 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3015 Alan Hayward <alan.hayward@arm.com>
3016 David Sherwood <david.sherwood@arm.com>
3018 * tree-vectorizer.h (vec_lower_bound): New structure.
3019 (_loop_vec_info): Add check_nonzero and lower_bounds.
3020 (LOOP_VINFO_CHECK_NONZERO): New macro.
3021 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3022 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3023 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3024 fields. Make seg_len the distance travelled, not including the
3026 (dr_direction_indicator): Declare.
3027 (dr_zero_step_indicator): Likewise.
3028 (dr_known_forward_stride_p): Likewise.
3029 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3031 (runtime_alias_check_p): Allow runtime alias checks with
3033 (operator ==): Compare access_size and align.
3034 (prune_runtime_alias_test_list): Rework for new distinction between
3035 the access_size and seg_len.
3036 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3038 (get_segment_min_max): New function.
3039 (create_intersect_range_checks): Use it.
3040 (dr_step_indicator): New function.
3041 (dr_direction_indicator): Likewise.
3042 (dr_zero_step_indicator): Likewise.
3043 (dr_known_forward_stride_p): Likewise.
3044 * tree-loop-distribution.c (data_ref_segment_size): Return
3045 DR_STEP * (niters - 1).
3046 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3048 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3049 (vect_preserves_scalar_order_p): New function, split out from...
3050 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3051 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3052 (vect_vfa_access_size): New function.
3053 (vect_vfa_align): Likewise.
3054 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3055 (dump_lower_bound): New function.
3056 (vect_check_lower_bound): Likewise.
3057 (vect_small_gap_p): Likewise.
3058 (vectorizable_with_step_bound_p): Likewise.
3059 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3060 depencies if the vectorization factor is 1. Convert the checks
3061 for nonzero steps into checks on the bounds of DR_STEP. Try using
3062 a bunds check for variable steps if the minimum required step is
3063 relatively small. Update calls to the dr_with_seg_len
3064 constructor and to vect_compile_time_alias.
3065 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3067 (vect_loop_versioning): Call it.
3068 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3070 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3072 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3073 Alan Hayward <alan.hayward@arm.com>
3074 David Sherwood <david.sherwood@arm.com>
3076 * doc/sourcebuild.texi (vect_scatter_store): Document.
3077 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3079 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3081 * genopinit.c (main): Add supports_vec_scatter_store and
3082 supports_vec_scatter_store_cached to target_optabs.
3083 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3084 IFN_MASK_SCATTER_STORE.
3085 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3087 * internal-fn.h (internal_store_fn_p): Declare.
3088 (internal_fn_stored_value_index): Likewise.
3089 * internal-fn.c (scatter_store_direct): New macro.
3090 (expand_scatter_store_optab_fn): New function.
3091 (direct_scatter_store_optab_supported_p): New macro.
3092 (internal_store_fn_p): New function.
3093 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3094 IFN_MASK_SCATTER_STORE.
3095 (internal_fn_mask_index): Likewise.
3096 (internal_fn_stored_value_index): New function.
3097 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3099 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3100 * optabs-query.c (supports_vec_scatter_store_p): New function.
3101 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3102 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3103 true for scatter stores.
3104 (vect_gather_scatter_fn_p): Handle scatter stores too.
3105 (vect_check_gather_scatter): Consider using scatter stores if
3106 supports_vec_scatter_store_p.
3107 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3109 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3110 internal_fn_stored_value_index.
3111 (check_load_store_masking): Handle scatter stores too.
3112 (vect_get_store_rhs): Make public.
3113 (vectorizable_call): Use internal_store_fn_p.
3114 (vectorizable_store): Handle scatter store internal functions.
3115 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3116 when deciding whether the end of the group has been reached.
3117 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3118 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3119 (mask_scatter_store<mode>): New insns.
3121 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3122 Alan Hayward <alan.hayward@arm.com>
3123 David Sherwood <david.sherwood@arm.com>
3125 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3126 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3127 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3129 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3130 Use vect_truncate_gather_scatter_offset if we can't treat the
3131 operation as a normal gather load or scatter store.
3132 (get_group_load_store_type): Take the gather_scatter_info
3133 as argument. Try using a gather load or scatter store for
3134 single-element groups.
3135 (get_load_store_type): Update calls to get_group_load_store_type
3136 and vect_use_strided_gather_scatters_p.
3138 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3139 Alan Hayward <alan.hayward@arm.com>
3140 David Sherwood <david.sherwood@arm.com>
3142 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3143 optional tree argument.
3144 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3146 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3147 but continue to use the current value as a fallback.
3148 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3149 to compare the updates.
3150 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3151 (get_load_store_type): Use it when handling a strided access.
3152 (vect_get_strided_load_store_ops): New function.
3153 (vect_get_data_ptr_increment): Likewise.
3154 (vectorizable_load): Handle strided gather loads. Always pass
3155 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3157 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3158 Alan Hayward <alan.hayward@arm.com>
3159 David Sherwood <david.sherwood@arm.com>
3161 * doc/md.texi (gather_load@var{m}): Document.
3162 (mask_gather_load@var{m}): Likewise.
3163 * genopinit.c (main): Add supports_vec_gather_load and
3164 supports_vec_gather_load_cached to target_optabs.
3165 * optabs-tree.c (init_tree_optimization_optabs): Use
3166 ggc_cleared_alloc to allocate target_optabs.
3167 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3168 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3170 * internal-fn.h (internal_load_fn_p): Declare.
3171 (internal_gather_scatter_fn_p): Likewise.
3172 (internal_fn_mask_index): Likewise.
3173 (internal_gather_scatter_fn_supported_p): Likewise.
3174 * internal-fn.c (gather_load_direct): New macro.
3175 (expand_gather_load_optab_fn): New function.
3176 (direct_gather_load_optab_supported_p): New macro.
3177 (direct_internal_fn_optab): New function.
3178 (internal_load_fn_p): Likewise.
3179 (internal_gather_scatter_fn_p): Likewise.
3180 (internal_fn_mask_index): Likewise.
3181 (internal_gather_scatter_fn_supported_p): Likewise.
3182 * optabs-query.c (supports_at_least_one_mode_p): New function.
3183 (supports_vec_gather_load_p): Likewise.
3184 * optabs-query.h (supports_vec_gather_load_p): Declare.
3185 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3186 and memory_type field.
3187 (NUM_PATTERNS): Bump to 15.
3188 * tree-vect-data-refs.c: Include internal-fn.h.
3189 (vect_gather_scatter_fn_p): New function.
3190 (vect_describe_gather_scatter_call): Likewise.
3191 (vect_check_gather_scatter): Try using internal functions for
3192 gather loads. Recognize existing calls to a gather load function.
3193 (vect_analyze_data_refs): Consider using gather loads if
3194 supports_vec_gather_load_p.
3195 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3196 (vect_get_gather_scatter_offset_type): Likewise.
3197 (vect_convert_mask_for_vectype): Likewise.
3198 (vect_add_conversion_to_patterm): Likewise.
3199 (vect_try_gather_scatter_pattern): Likewise.
3200 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3201 (vect_vect_recog_func_ptrs): Add it.
3202 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3203 internal_fn_mask_index and internal_gather_scatter_fn_p.
3204 (check_load_store_masking): Take the gather_scatter_info as an
3205 argument and handle gather loads.
3206 (vect_get_gather_scatter_ops): New function.
3207 (vectorizable_call): Check internal_load_fn_p.
3208 (vectorizable_load): Likewise. Handle gather load internal
3210 (vectorizable_store): Update call to check_load_store_masking.
3211 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3212 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3213 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3214 (aarch64_gather_scale_operand_d): New predicates.
3215 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3216 (mask_gather_load<mode>): New insns.
3218 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3219 Alan Hayward <alan.hayward@arm.com>
3220 David Sherwood <david.sherwood@arm.com>
3222 * optabs.def (fold_left_plus_optab): New optab.
3223 * doc/md.texi (fold_left_plus_@var{m}): Document.
3224 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3225 * internal-fn.c (fold_left_direct): Define.
3226 (expand_fold_left_optab_fn): Likewise.
3227 (direct_fold_left_optab_supported_p): Likewise.
3228 * fold-const-call.c (fold_const_fold_left): New function.
3229 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3230 * tree-parloops.c (valid_reduction_p): New function.
3231 (gather_scalar_reductions): Use it.
3232 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3233 (vect_finish_replace_stmt): Declare.
3234 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3235 (needs_fold_left_reduction_p): New function, split out from...
3236 (vect_is_simple_reduction): ...here. Accept reductions that
3237 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3238 (vect_force_simple_reduction): Also store the reduction type in
3239 the assignment's STMT_VINFO_REDUC_TYPE.
3240 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3241 (merge_with_identity): New function.
3242 (vect_expand_fold_left): Likewise.
3243 (vectorize_fold_left_reduction): Likewise.
3244 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3245 scalar phi in place for it. Check for target support and reject
3246 cases that would reassociate the operation. Defer the transform
3247 phase to vectorize_fold_left_reduction.
3248 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3249 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3250 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3252 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3254 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3255 call to ifc_temp_var.
3257 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3258 Alan Hayward <alan.hayward@arm.com>
3259 David Sherwood <david.sherwood@arm.com>
3261 * target.def (legitimize_address_displacement): Take the original
3262 offset as a poly_int.
3263 * targhooks.h (default_legitimize_address_displacement): Update
3265 * targhooks.c (default_legitimize_address_displacement): Likewise.
3266 * doc/tm.texi: Regenerate.
3267 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3268 as an argument, moving assert of ad->disp == ad->disp_term to...
3269 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3270 Try calling targetm.legitimize_address_displacement before expanding
3271 the address rather than afterwards, and adjust for the new interface.
3272 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3273 Match the new hook interface. Handle SVE addresses.
3274 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3277 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3279 * Makefile.in (OBJS): Add early-remat.o.
3280 * target.def (select_early_remat_modes): New hook.
3281 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3282 * doc/tm.texi: Regenerate.
3283 * targhooks.h (default_select_early_remat_modes): Declare.
3284 * targhooks.c (default_select_early_remat_modes): New function.
3285 * timevar.def (TV_EARLY_REMAT): New timevar.
3286 * passes.def (pass_early_remat): New pass.
3287 * tree-pass.h (make_pass_early_remat): Declare.
3288 * early-remat.c: New file.
3289 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3291 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3293 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3294 Alan Hayward <alan.hayward@arm.com>
3295 David Sherwood <david.sherwood@arm.com>
3297 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3298 vfm1 with a bound_epilog parameter.
3299 (vect_do_peeling): Update calls accordingly, and move the prologue
3300 call earlier in the function. Treat the base bound_epilog as 0 for
3301 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3302 this base when peeling for gaps.
3303 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3304 with fully-masked loops.
3305 (vect_estimate_min_profitable_iters): Handle the single peeled
3306 iteration in that case.
3308 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3309 Alan Hayward <alan.hayward@arm.com>
3310 David Sherwood <david.sherwood@arm.com>
3312 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3313 single-element interleaving even if the size is not a power of 2.
3314 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3315 accesses for single-element interleaving if the group size is
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 * doc/md.texi (fold_extract_last_@var{m}): Document.
3323 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3324 * optabs.def (fold_extract_last_optab): New optab.
3325 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3326 * internal-fn.c (fold_extract_direct): New macro.
3327 (expand_fold_extract_optab_fn): Likewise.
3328 (direct_fold_extract_optab_supported_p): Likewise.
3329 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3330 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3331 EXTRACT_LAST_REDUCTION.
3332 (get_initial_def_for_reduction): Do not create an initial vector
3333 for EXTRACT_LAST_REDUCTION reductions.
3334 (vectorizable_reduction): Leave the scalar phi in place for
3335 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3336 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3337 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3338 transform phase to vectorizable_condition.
3339 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3341 (vect_finish_stmt_generation): ...here.
3342 (vect_finish_replace_stmt): New function.
3343 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3344 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3346 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3348 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3349 Alan Hayward <alan.hayward@arm.com>
3350 David Sherwood <david.sherwood@arm.com>
3352 * doc/md.texi (extract_last_@var{m}): Document.
3353 * optabs.def (extract_last_optab): New optab.
3354 * internal-fn.def (EXTRACT_LAST): New internal function.
3355 * internal-fn.c (cond_unary_direct): New macro.
3356 (expand_cond_unary_optab_fn): Likewise.
3357 (direct_cond_unary_optab_supported_p): Likewise.
3358 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3359 loops using EXTRACT_LAST.
3360 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3361 (extract_last_<mode>): ...this optab.
3362 (vec_extract<mode><Vel>): Update accordingly.
3364 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3365 Alan Hayward <alan.hayward@arm.com>
3366 David Sherwood <david.sherwood@arm.com>
3368 * target.def (empty_mask_is_expensive): New hook.
3369 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3370 * doc/tm.texi: Regenerate.
3371 * targhooks.h (default_empty_mask_is_expensive): Declare.
3372 * targhooks.c (default_empty_mask_is_expensive): New function.
3373 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3374 if the target says that empty masks are expensive.
3375 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3377 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3379 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3380 Alan Hayward <alan.hayward@arm.com>
3381 David Sherwood <david.sherwood@arm.com>
3383 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3384 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3385 (vect_use_loop_mask_for_alignment_p): New function.
3386 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3387 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3388 niters_skip argument. Make sure that the first niters_skip elements
3389 of the first iteration are inactive.
3390 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3391 Update call to vect_set_loop_masks_directly.
3392 (get_misalign_in_elems): New function, split out from...
3393 (vect_gen_prolog_loop_niters): ...here.
3394 (vect_update_init_of_dr): Take a code argument that specifies whether
3395 the adjustment should be added or subtracted.
3396 (vect_update_init_of_drs): Likewise.
3397 (vect_prepare_for_masked_peels): New function.
3398 (vect_do_peeling): Skip prologue peeling if we're using a mask
3399 instead. Update call to vect_update_inits_of_drs.
3400 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3402 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3403 alignment. Do not include the number of peeled iterations in
3404 the minimum threshold in that case.
3405 (vectorizable_induction): Adjust the start value down by
3406 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3407 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3408 Take the number of skipped iterations into account when calculating
3410 * tree-vect-stmts.c (vect_gen_while_not): New function.
3412 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3413 Alan Hayward <alan.hayward@arm.com>
3414 David Sherwood <david.sherwood@arm.com>
3416 * doc/sourcebuild.texi (vect_fully_masked): Document.
3417 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3419 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3421 (vect_analyze_loop_2): ...here. Don't check the vectorization
3422 factor against the number of loop iterations if the loop is
3425 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3426 Alan Hayward <alan.hayward@arm.com>
3427 David Sherwood <david.sherwood@arm.com>
3429 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3430 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3431 (dump_groups): Update accordingly.
3432 (iv_use::mem_type): New member variable.
3433 (address_p): New function.
3434 (record_use): Add a mem_type argument and initialize the new
3436 (record_group_use): Add a mem_type argument. Use address_p.
3437 Remove obsolete null checks of base_object. Update call to record_use.
3438 (find_interesting_uses_op): Update call to record_group_use.
3439 (find_interesting_uses_cond): Likewise.
3440 (find_interesting_uses_address): Likewise.
3441 (get_mem_type_for_internal_fn): New function.
3442 (find_address_like_use): Likewise.
3443 (find_interesting_uses_stmt): Try find_address_like_use before
3444 calling find_interesting_uses_op.
3445 (addr_offset_valid_p): Use the iv mem_type field as the type
3446 of the addressed memory.
3447 (add_autoinc_candidates): Likewise.
3448 (get_address_cost): Likewise.
3449 (split_small_address_groups_p): Use address_p.
3450 (split_address_groups): Likewise.
3451 (add_iv_candidate_for_use): Likewise.
3452 (autoinc_possible_for_pair): Likewise.
3453 (rewrite_groups): Likewise.
3454 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3455 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3456 (get_alias_ptr_type_for_ptr_address): New function.
3457 (rewrite_use_address): Rewrite address uses in calls that were
3458 identified by find_address_like_use.
3460 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3461 Alan Hayward <alan.hayward@arm.com>
3462 David Sherwood <david.sherwood@arm.com>
3464 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3466 * gimple-expr.h (is_gimple_addressable: Likewise.
3467 * gimple-expr.c (is_gimple_address): Likewise.
3468 * internal-fn.c (expand_call_mem_ref): New function.
3469 (expand_mask_load_optab_fn): Use it.
3470 (expand_mask_store_optab_fn): Likewise.
3472 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3473 Alan Hayward <alan.hayward@arm.com>
3474 David Sherwood <david.sherwood@arm.com>
3476 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3477 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3478 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3479 (cond_umax@var{mode}): Document.
3480 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3481 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3482 (cond_umin_optab, cond_umax_optab): New optabs.
3483 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3484 (COND_IOR, COND_XOR): New internal functions.
3485 * internal-fn.h (get_conditional_internal_fn): Declare.
3486 * internal-fn.c (cond_binary_direct): New macro.
3487 (expand_cond_binary_optab_fn): Likewise.
3488 (direct_cond_binary_optab_supported_p): Likewise.
3489 (get_conditional_internal_fn): New function.
3490 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3491 Cope with reduction statements that are vectorized as calls rather
3493 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3494 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3495 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3496 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3497 (UNSPEC_COND_EOR): New unspecs.
3498 (optab): Add mappings for them.
3499 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3500 (sve_int_op, sve_fp_op): New int attributes.
3502 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3503 Alan Hayward <alan.hayward@arm.com>
3504 David Sherwood <david.sherwood@arm.com>
3506 * optabs.def (while_ult_optab): New optab.
3507 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3508 * internal-fn.def (WHILE_ULT): New internal function.
3509 * internal-fn.h (direct_internal_fn_supported_p): New override
3510 that takes two types as argument.
3511 * internal-fn.c (while_direct): New macro.
3512 (expand_while_optab_fn): New function.
3513 (convert_optab_supported_p): Likewise.
3514 (direct_while_optab_supported_p): New macro.
3515 * wide-int.h (wi::udiv_ceil): New function.
3516 * tree-vectorizer.h (rgroup_masks): New structure.
3517 (vec_loop_masks): New typedef.
3518 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3520 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3521 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3522 (vect_max_vf): New function.
3523 (slpeel_make_loop_iterate_ntimes): Delete.
3524 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3525 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3526 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3527 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3528 internal-fn.h, stor-layout.h and optabs-query.h.
3529 (vect_set_loop_mask): New function.
3530 (add_preheader_seq): Likewise.
3531 (add_header_seq): Likewise.
3532 (interleave_supported_p): Likewise.
3533 (vect_maybe_permute_loop_masks): Likewise.
3534 (vect_set_loop_masks_directly): Likewise.
3535 (vect_set_loop_condition_masked): Likewise.
3536 (vect_set_loop_condition_unmasked): New function, split out from
3537 slpeel_make_loop_iterate_ntimes.
3538 (slpeel_make_loop_iterate_ntimes): Rename to..
3539 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3540 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3541 (vect_do_peeling): Update call accordingly.
3542 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3544 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3545 mask_compare_type, can_fully_mask_p and fully_masked_p.
3546 (release_vec_loop_masks): New function.
3547 (_loop_vec_info): Use it to free the loop masks.
3548 (can_produce_all_loop_masks_p): New function.
3549 (vect_get_max_nscalars_per_iter): Likewise.
3550 (vect_verify_full_masking): Likewise.
3551 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3552 retries, and free the mask rgroups before retrying. Check loop-wide
3553 reasons for disallowing fully-masked loops. Make the final decision
3554 about whether use a fully-masked loop or not.
3555 (vect_estimate_min_profitable_iters): Do not assume that peeling
3556 for the number of iterations will be needed for fully-masked loops.
3557 (vectorizable_reduction): Disable fully-masked loops.
3558 (vectorizable_live_operation): Likewise.
3559 (vect_halve_mask_nunits): New function.
3560 (vect_double_mask_nunits): Likewise.
3561 (vect_record_loop_mask): Likewise.
3562 (vect_get_loop_mask): Likewise.
3563 (vect_transform_loop): Handle the case in which the final loop
3564 iteration might handle a partial vector. Call vect_set_loop_condition
3565 instead of slpeel_make_loop_iterate_ntimes.
3566 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3567 (check_load_store_masking): New function.
3568 (prepare_load_store_mask): Likewise.
3569 (vectorizable_store): Handle fully-masked loops.
3570 (vectorizable_load): Likewise.
3571 (supportable_widening_operation): Use vect_halve_mask_nunits for
3573 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3574 (vect_gen_while): New function.
3575 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3576 (aarch64_uqdec<mode>): New insn.
3578 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3579 Alan Hayward <alan.hayward@arm.com>
3580 David Sherwood <david.sherwood@arm.com>
3582 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3583 (reduc_xor_scal_optab): New optabs.
3584 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3585 (reduc_xor_scal_@var{m}): Document.
3586 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3587 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3589 * fold-const-call.c (fold_const_call): Handle them.
3590 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3591 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3592 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3593 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3594 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3595 (UNSPEC_XORV): New unspecs.
3596 (optab): Add entries for them.
3597 (BITWISEV): New int iterator.
3598 (bit_reduc_op): New int attributes.
3600 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3601 Alan Hayward <alan.hayward@arm.com>
3602 David Sherwood <david.sherwood@arm.com>
3604 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3605 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3606 * optabs.def (vec_shl_insert_optab): New optab.
3607 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3608 (duplicate_and_interleave): Likewise.
3609 * tree-vect-loop.c: Include internal-fn.h.
3610 (neutral_op_for_slp_reduction): New function, split out from
3611 get_initial_defs_for_reduction.
3612 (get_initial_def_for_reduction): Handle option 2 for variable-length
3613 vectors by loading the neutral value into a vector and then shifting
3614 the initial value into element 0.
3615 (get_initial_defs_for_reduction): Replace the code argument with
3616 the neutral value calculated by neutral_op_for_slp_reduction.
3617 Use gimple_build_vector for constant-length vectors.
3618 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3619 but the first group_size elements have a neutral value.
3620 Use duplicate_and_interleave otherwise.
3621 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3622 Update call to get_initial_defs_for_reduction. Handle SLP
3623 reductions for variable-length vectors by creating one vector
3624 result for each scalar result, with the elements associated
3625 with other scalar results stubbed out with the neutral value.
3626 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3627 Require IFN_VEC_SHL_INSERT for double reductions on
3628 variable-length vectors, or SLP reductions that have
3629 a neutral value. Require can_duplicate_and_interleave_p
3630 support for variable-length unchained SLP reductions if there
3631 is no neutral value, such as for MIN/MAX reductions. Also require
3632 the number of vector elements to be a multiple of the number of
3633 SLP statements when doing variable-length unchained SLP reductions.
3634 Update call to vect_create_epilog_for_reduction.
3635 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3636 and remove initial values.
3637 (duplicate_and_interleave): Make public.
3638 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3639 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3641 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3642 Alan Hayward <alan.hayward@arm.com>
3643 David Sherwood <david.sherwood@arm.com>
3645 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3646 (can_duplicate_and_interleave_p): New function.
3647 (vect_get_and_check_slp_defs): Take the vector of statements
3648 rather than just the current one. Remove excess parentheses.
3649 Restriction rejectinon of vect_constant_def and vect_external_def
3650 for variable-length vectors to boolean types, or types for which
3651 can_duplicate_and_interleave_p is false.
3652 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3653 (duplicate_and_interleave): New function.
3654 (vect_get_constant_vectors): Use gimple_build_vector for
3655 constant-length vectors and suitable variable-length constant
3656 vectors. Use duplicate_and_interleave for other variable-length
3657 vectors. Don't defer the update when inserting new statements.
3659 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3660 Alan Hayward <alan.hayward@arm.com>
3661 David Sherwood <david.sherwood@arm.com>
3663 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3664 min_profitable_iters doesn't go negative.
3666 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3667 Alan Hayward <alan.hayward@arm.com>
3668 David Sherwood <david.sherwood@arm.com>
3670 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3671 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3672 * optabs.def (vec_mask_load_lanes_optab): New optab.
3673 (vec_mask_store_lanes_optab): Likewise.
3674 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3675 (MASK_STORE_LANES): Likewise.
3676 * internal-fn.c (mask_load_lanes_direct): New macro.
3677 (mask_store_lanes_direct): Likewise.
3678 (expand_mask_load_optab_fn): Handle masked operations.
3679 (expand_mask_load_lanes_optab_fn): New macro.
3680 (expand_mask_store_optab_fn): Handle masked operations.
3681 (expand_mask_store_lanes_optab_fn): New macro.
3682 (direct_mask_load_lanes_optab_supported_p): Likewise.
3683 (direct_mask_store_lanes_optab_supported_p): Likewise.
3684 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3686 (vect_load_lanes_supported): Likewise.
3687 * tree-vect-data-refs.c (strip_conversion): New function.
3688 (can_group_stmts_p): Likewise.
3689 (vect_analyze_data_ref_accesses): Use it instead of checking
3690 for a pair of assignments.
3691 (vect_store_lanes_supported): Take a masked_p parameter.
3692 (vect_load_lanes_supported): Likewise.
3693 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3694 vect_store_lanes_supported and vect_load_lanes_supported.
3695 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3696 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3697 parameter. Don't allow gaps for masked accesses.
3698 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3699 and vect_load_lanes_supported.
3700 (get_load_store_type): Take a masked_p parameter and update
3701 call to get_group_load_store_type.
3702 (vectorizable_store): Update call to get_load_store_type.
3703 Handle IFN_MASK_STORE_LANES.
3704 (vectorizable_load): Update call to get_load_store_type.
3705 Handle IFN_MASK_LOAD_LANES.
3707 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3708 Alan Hayward <alan.hayward@arm.com>
3709 David Sherwood <david.sherwood@arm.com>
3711 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3713 * config/aarch64/aarch64-protos.h
3714 (aarch64_sve_struct_memory_operand_p): Declare.
3715 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3716 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3717 (VPRED, vpred): Handle SVE structure modes.
3718 * config/aarch64/constraints.md (Utx): New constraint.
3719 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3720 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3721 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3722 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3723 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3724 structure modes. Split into pieces after RA.
3725 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3726 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3728 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3729 SVE structure modes.
3730 (aarch64_classify_address): Likewise.
3731 (sizetochar): Move earlier in file.
3732 (aarch64_print_operand): Handle SVE register lists.
3733 (aarch64_array_mode): New function.
3734 (aarch64_sve_struct_memory_operand_p): Likewise.
3735 (TARGET_ARRAY_MODE): Redefine.
3737 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3738 Alan Hayward <alan.hayward@arm.com>
3739 David Sherwood <david.sherwood@arm.com>
3741 * target.def (array_mode): New target hook.
3742 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3743 * doc/tm.texi: Regenerate.
3744 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3745 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3746 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3748 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3751 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3752 Alan Hayward <alan.hayward@arm.com>
3753 David Sherwood <david.sherwood@arm.com>
3755 * fold-const.c (fold_binary_loc): Check the argument types
3756 rather than the result type when testing for a vector operation.
3758 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3760 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3761 * doc/tm.texi: Regenerate.
3763 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3764 Alan Hayward <alan.hayward@arm.com>
3765 David Sherwood <david.sherwood@arm.com>
3767 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3768 (sve): Document new AArch64 extension.
3769 * doc/md.texi (w): Extend the description of the AArch64
3770 constraint to include SVE vectors.
3771 (Upl, Upa): Document new AArch64 predicate constraints.
3772 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3774 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3775 (msve-vector-bits=): New option.
3776 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3777 SVE when these are disabled.
3778 (sve): New extension.
3779 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3780 modes. Adjust their number of units based on aarch64_sve_vg.
3781 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3782 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3783 aarch64_addr_query_type.
3784 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3785 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3786 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3787 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3788 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3789 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3790 (aarch64_simd_imm_zero_p): Delete.
3791 (aarch64_check_zero_based_sve_index_immediate): Declare.
3792 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3793 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3794 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3795 (aarch64_sve_float_mul_immediate_p): Likewise.
3796 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3798 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3799 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3800 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3801 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3802 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3803 (aarch64_regmode_natural_size): Likewise.
3804 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3805 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3807 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3808 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3809 for VG and the SVE predicate registers.
3810 (V_ALIASES): Add a "z"-prefixed alias.
3811 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3812 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3813 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3814 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3815 (REG_CLASS_NAMES): Add entries for them.
3816 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3817 and the predicate registers.
3818 (aarch64_sve_vg): Declare.
3819 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3820 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3821 (REGMODE_NATURAL_SIZE): Define.
3822 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3824 * config/aarch64/aarch64.c: Include cfgrtl.h.
3825 (simd_immediate_info): Add a constructor for series vectors,
3826 and an associated step field.
3827 (aarch64_sve_vg): New variable.
3828 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3829 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3830 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3831 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3832 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3833 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3834 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3835 (aarch64_get_mask_mode): New functions.
3836 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3837 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3838 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3839 predicate modes and predicate registers. Explicitly restrict
3840 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3841 to store a vector mode if it is recognized by
3842 aarch64_classify_vector_mode.
3843 (aarch64_regmode_natural_size): New function.
3844 (aarch64_hard_regno_caller_save_mode): Return the original mode
3846 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3847 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3848 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3849 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3851 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3852 does not overlap dest if the function is frame-related. Handle
3854 (aarch64_split_add_offset): New function.
3855 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3856 them aarch64_add_offset.
3857 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3858 and update call to aarch64_sub_sp.
3859 (aarch64_add_cfa_expression): New function.
3860 (aarch64_expand_prologue): Pass extra temporary registers to the
3861 functions above. Handle the case in which we need to emit new
3862 DW_CFA_expressions for registers that were originally saved
3863 relative to the stack pointer, but now have to be expressed
3864 relative to the frame pointer.
3865 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3867 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3868 IP0 and IP1 values for SVE frames.
3869 (aarch64_expand_vec_series): New function.
3870 (aarch64_expand_sve_widened_duplicate): Likewise.
3871 (aarch64_expand_sve_const_vector): Likewise.
3872 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3873 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3874 into the register, rather than emitting a SET directly.
3875 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3876 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3877 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3878 (offset_9bit_signed_scaled_p): New functions.
3879 (aarch64_replicate_bitmask_imm): New function.
3880 (aarch64_bitmask_imm): Use it.
3881 (aarch64_cannot_force_const_mem): Reject expressions involving
3882 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3883 (aarch64_classify_index): Handle SVE indices, by requiring
3884 a plain register index with a scale that matches the element size.
3885 (aarch64_classify_address): Handle SVE addresses. Assert that
3886 the mode of the address is VOIDmode or an integer mode.
3887 Update call to aarch64_classify_symbol.
3888 (aarch64_classify_symbolic_expression): Update call to
3889 aarch64_classify_symbol.
3890 (aarch64_const_vec_all_in_range_p): New function.
3891 (aarch64_print_vector_float_operand): Likewise.
3892 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3893 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3894 and the FP immediates 1.0 and 0.5.
3895 (aarch64_print_address_internal): Handle SVE addresses.
3896 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3897 (aarch64_regno_regclass): Handle predicate registers.
3898 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3900 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3901 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3902 (aarch64_convert_sve_vector_bits): New function.
3903 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3904 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3906 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3907 Handle SVE vector and predicate modes. Accept VL-based constants
3908 that need only one temporary register, and VL offsets that require
3909 no temporary registers.
3910 (aarch64_conditional_register_usage): Mark the predicate registers
3911 as fixed if SVE isn't available.
3912 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3913 Return true for SVE vector and predicate modes.
3914 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3915 rather than an unsigned int. Handle SVE modes.
3916 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3918 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3920 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3921 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3922 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3923 (aarch64_sve_float_mul_immediate_p): New functions.
3924 (aarch64_sve_valid_immediate): New function.
3925 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3926 Explicitly reject structure modes. Check for INDEX constants.
3927 Handle PTRUE and PFALSE constants.
3928 (aarch64_check_zero_based_sve_index_immediate): New function.
3929 (aarch64_simd_imm_zero_p): Delete.
3930 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3931 vector modes. Accept constants in the range of CNT[BHWD].
3932 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3933 ask for an Advanced SIMD mode.
3934 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3935 (aarch64_simd_vector_alignment): Handle SVE predicates.
3936 (aarch64_vectorize_preferred_vector_alignment): New function.
3937 (aarch64_simd_vector_alignment_reachable): Use it instead of
3939 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3940 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3942 (MAX_VECT_LEN): Delete.
3943 (expand_vec_perm_d): Add a vec_flags field.
3944 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3945 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3946 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3948 (aarch64_evpc_rev): Rename to...
3949 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3950 (aarch64_evpc_rev_global): New function.
3951 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3952 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3954 (aarch64_evpc_sve_tbl): New function.
3955 (aarch64_expand_vec_perm_const_1): Update after rename of
3956 aarch64_evpc_rev. Handle SVE permutes too, trying
3957 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3958 than aarch64_evpc_tbl.
3959 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3960 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3961 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3962 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3963 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3964 (aarch64_expand_sve_vcond): New functions.
3965 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3966 of aarch64_vector_mode_p.
3967 (aarch64_dwarf_poly_indeterminate_value): New function.
3968 (aarch64_compute_pressure_classes): Likewise.
3969 (aarch64_can_change_mode_class): Likewise.
3970 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3971 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3972 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3973 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3974 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3975 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3976 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3977 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3979 (Dn, Dl, Dr): Accept const as well as const_vector.
3980 (Dz): Likewise. Compare against CONST0_RTX.
3981 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3982 of "vector" where appropriate.
3983 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3984 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3985 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3986 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3987 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3988 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3989 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3990 (v_int_equiv): Extend to SVE modes.
3991 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3993 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3994 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3995 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3996 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3997 (SVE_COND_FP_CMP): New int iterators.
3998 (perm_hilo): Handle the new unpack unspecs.
3999 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4001 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4002 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4003 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4004 (aarch64_equality_operator, aarch64_constant_vector_operand)
4005 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4006 (aarch64_sve_nonimmediate_operand): Likewise.
4007 (aarch64_sve_general_operand): Likewise.
4008 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4009 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4010 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4011 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4012 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4013 (aarch64_sve_float_arith_immediate): Likewise.
4014 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4015 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4016 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4017 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4018 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4019 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4020 (aarch64_sve_float_arith_operand): Likewise.
4021 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4022 (aarch64_sve_float_mul_operand): Likewise.
4023 (aarch64_sve_vec_perm_operand): Likewise.
4024 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4025 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4026 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4027 as well as const_vector.
4028 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4029 in file. Use CONST0_RTX and CONSTM1_RTX.
4030 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4031 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4032 Use aarch64_simd_imm_zero.
4033 * config/aarch64/aarch64-sve.md: New file.
4034 * config/aarch64/aarch64.md: Include it.
4035 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4036 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4037 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4038 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4039 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4040 (sve): New attribute.
4041 (enabled): Disable instructions with the sve attribute unless
4043 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4044 aarch64_expand_mov_immediate.
4045 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4046 CNT[BHSD] immediates.
4047 (movti): Split CONST_POLY_INT moves into two halves.
4048 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4049 Split additions that need a temporary here if the destination
4050 is the stack pointer.
4051 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4052 (*add<mode>3_poly_1): New instruction.
4053 (set_clobber_cc): New expander.
4055 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4057 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4058 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4059 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4060 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4061 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4062 Change innermode from fixed_mode_size to machine_mode.
4063 (simplify_subreg): Update call accordingly. Handle a constant-sized
4064 subreg of a variable-length CONST_VECTOR.
4066 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4067 Alan Hayward <alan.hayward@arm.com>
4068 David Sherwood <david.sherwood@arm.com>
4070 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4071 (add_offset_to_base): New function, split out from...
4072 (create_mem_ref): ...here. When handling a scale other than 1,
4073 check first whether the address is valid without the offset.
4074 Add it into the base if so, leaving the index and scale as-is.
4076 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4079 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4080 fold_for_warn before checking if arg2 is INTEGER_CST.
4082 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4084 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4085 (store_multiple_operation): Delete.
4086 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4087 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4088 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4089 guarded by TARGET_STRING.
4090 (rs6000_output_load_multiple): Delete.
4091 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4092 OPTION_MASK_STRING / TARGET_STRING handling.
4093 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4094 (const rs6000_opt_masks) <"string">: Change mask to 0.
4095 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4096 (MASK_STRING): Delete.
4097 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4099 (load_multiple): Delete.
4106 (store_multiple): Delete.
4113 (movmemsi_8reg): Delete.
4114 (corresponding unnamed define_insn): Delete.
4115 (movmemsi_6reg): Delete.
4116 (corresponding unnamed define_insn): Delete.
4117 (movmemsi_4reg): Delete.
4118 (corresponding unnamed define_insn): Delete.
4119 (movmemsi_2reg): Delete.
4120 (corresponding unnamed define_insn): Delete.
4121 (movmemsi_1reg): Delete.
4122 (corresponding unnamed define_insn): Delete.
4123 * config/rs6000/rs6000.opt (mno-string): New.
4124 (mstring): Replace by deprecation warning stub.
4125 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4127 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4129 * regrename.c (regrename_do_replace): If replacing the same
4130 reg multiple times, try to reuse last created gen_raw_REG.
4133 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4134 main to workaround a bug in GDB.
4136 2018-01-12 Tom de Vries <tom@codesourcery.com>
4139 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4141 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4143 PR rtl-optimization/80481
4144 * ira-color.c (get_cap_member): New function.
4145 (allocnos_conflict_by_live_ranges_p): Use it.
4146 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4147 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4149 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4152 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4153 (*saddl_se_1): Ditto.
4155 (*ssubl_se_1): Ditto.
4157 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4159 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4160 rather than wi::to_widest for DR_INITs.
4161 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4162 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4163 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4165 (vect_analyze_group_access_1): Note that here.
4167 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4169 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4170 polynomial type sizes.
4172 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4174 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4175 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4176 (gimple_add_tmp_var): Likewise.
4178 2018-01-12 Martin Liska <mliska@suse.cz>
4180 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4181 (gimple_alloc_sizes): Likewise.
4182 (dump_gimple_statistics): Use PRIu64 in printf format.
4183 * gimple.h: Change uint64_t to int.
4185 2018-01-12 Martin Liska <mliska@suse.cz>
4187 * tree-core.h: Use uint64_t instead of int.
4188 * tree.c (tree_node_counts): Likewise.
4189 (tree_node_sizes): Likewise.
4190 (dump_tree_statistics): Use PRIu64 in printf format.
4192 2018-01-12 Martin Liska <mliska@suse.cz>
4194 * Makefile.in: As qsort_chk is implemented in vec.c, add
4195 vec.o to linkage of gencfn-macros.
4196 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4197 passing the info to record_node_allocation_statistics.
4198 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4200 * ggc-common.c (struct ggc_usage): Add operator== and use
4201 it in operator< and compare function.
4202 * mem-stats.h (struct mem_usage): Likewise.
4203 * vec.c (struct vec_usage): Remove operator< and compare
4204 function. Can be simply inherited.
4206 2018-01-12 Martin Jambor <mjambor@suse.cz>
4209 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4210 * tree-ssa-math-opts.c: Include domwalk.h.
4211 (convert_mult_to_fma_1): New function.
4212 (fma_transformation_info): New type.
4213 (fma_deferring_state): Likewise.
4214 (cancel_fma_deferring): New function.
4215 (result_of_phi): Likewise.
4216 (last_fma_candidate_feeds_initial_phi): Likewise.
4217 (convert_mult_to_fma): Added deferring logic, split actual
4218 transformation to convert_mult_to_fma_1.
4219 (math_opts_dom_walker): New type.
4220 (math_opts_dom_walker::after_dom_children): New method, body moved
4221 here from pass_optimize_widening_mul::execute, added deferring logic
4223 (pass_optimize_widening_mul::execute): Moved most of code to
4224 math_opts_dom_walker::after_dom_children.
4225 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4226 * config/i386/i386.c (ix86_option_override_internal): Added
4227 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4229 2018-01-12 Richard Biener <rguenther@suse.de>
4232 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4233 inline instance vars.
4235 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4238 * config/rx/rx.c (rx_is_restricted_memory_address):
4241 2018-01-12 Richard Biener <rguenther@suse.de>
4243 PR tree-optimization/80846
4244 * target.def (split_reduction): New target hook.
4245 * targhooks.c (default_split_reduction): New function.
4246 * targhooks.h (default_split_reduction): Declare.
4247 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4248 target requests first reduce vectors by combining low and high
4250 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4251 (get_vectype_for_scalar_type_and_size): Export.
4252 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4253 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4254 * doc/tm.texi: Regenerate.
4255 * config/i386/i386.c (ix86_split_reduction): Implement
4256 TARGET_VECTORIZE_SPLIT_REDUCTION.
4258 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4261 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4262 in PIC mode except for TARGET_VXWORKS_RTP.
4263 * config/sparc/sparc.c: Include cfgrtl.h.
4264 (TARGET_INIT_PIC_REG): Define.
4265 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4266 (sparc_pic_register_p): New predicate.
4267 (sparc_legitimate_address_p): Use it.
4268 (sparc_legitimize_pic_address): Likewise.
4269 (sparc_delegitimize_address): Likewise.
4270 (sparc_mode_dependent_address_p): Likewise.
4271 (gen_load_pcrel_sym): Remove 4th parameter.
4272 (load_got_register): Adjust call to above. Remove obsolete stuff.
4273 (sparc_expand_prologue): Do not call load_got_register here.
4274 (sparc_flat_expand_prologue): Likewise.
4275 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4276 (sparc_use_pseudo_pic_reg): New function.
4277 (sparc_init_pic_reg): Likewise.
4278 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4279 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4281 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4283 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4284 Add item for branch_cost.
4286 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4288 PR rtl-optimization/83565
4289 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4290 not extend the result to a larger mode for rotate operations.
4291 (num_sign_bit_copies1): Likewise.
4293 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4296 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4298 Use values-Xc.o for -pedantic.
4299 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4301 2018-01-12 Martin Liska <mliska@suse.cz>
4304 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4306 (possible_polymorphic_call_targets): Use it.
4307 (ipa_devirt): Likewise.
4309 2018-01-12 Martin Liska <mliska@suse.cz>
4311 * profile-count.h (enum profile_quality): Use 0 as invalid
4312 enum value of profile_quality.
4314 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4316 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4317 -mext-string options.
4319 2018-01-12 Richard Biener <rguenther@suse.de>
4321 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4322 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4323 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4325 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4327 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4329 * configure.ac (--with-long-double-format): Add support for the
4330 configuration option to change the default long double format on
4332 * config.gcc (powerpc*-linux*-*): Likewise.
4333 * configure: Regenerate.
4334 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4335 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4336 used without modification.
4338 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4340 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4341 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4342 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4343 MISC_BUILTIN_SPEC_BARRIER.
4344 (rs6000_init_builtins): Likewise.
4345 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4347 (speculation_barrier): New define_insn.
4348 * doc/extend.texi: Document __builtin_speculation_barrier.
4350 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4353 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4354 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4355 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4357 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4358 integral modes instead of "ss" and "sd".
4359 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4360 vectors with 32-bit and 64-bit elements.
4361 (vecdupssescalarmodesuffix): New mode attribute.
4362 (vec_dup<mode>): Use it.
4364 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4367 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4368 frame if argument is passed on stack.
4370 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4373 * ree.c (combine_reaching_defs): Optimize also
4374 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4375 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4377 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4380 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4382 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4385 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4386 after they are computed.
4388 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4390 PR tree-optimization/83695
4391 * gimple-loop-linterchange.cc
4392 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4393 reset cached scev information after interchange.
4394 (pass_linterchange::execute): Remove call to scev_reset_htab.
4396 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4398 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4399 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4400 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4401 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4402 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4403 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4404 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4405 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4406 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4407 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4408 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4409 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4410 (V_lane_reg): Likewise.
4411 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4413 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4414 (vfmal_lane_low<mode>_intrinsic,
4415 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4416 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4417 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4418 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4419 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4420 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4422 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4424 * config/arm/arm-cpus.in (fp16fml): New feature.
4425 (ALL_SIMD): Add fp16fml.
4426 (armv8.2-a): Add fp16fml as an option.
4427 (armv8.3-a): Likewise.
4428 (armv8.4-a): Add fp16fml as part of fp16.
4429 * config/arm/arm.h (TARGET_FP16FML): Define.
4430 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4432 * config/arm/arm-modes.def (V2HF): Define.
4433 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4434 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4435 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4436 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4437 vfmsl_low, vfmsl_high): New set of builtins.
4438 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4439 (vfml_op): New code attribute.
4440 (VFMLHALVES): New int iterator.
4441 (VFML, VFMLSEL): New mode attributes.
4442 (V_reg): Define mapping for V2HF.
4443 (V_hi, V_lo): New mode attributes.
4444 (VF_constraint): Likewise.
4445 (vfml_half, vfml_half_selector): New int attributes.
4446 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4448 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4449 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4451 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4452 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4453 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4454 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4456 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4457 Document new effective target and option set.
4459 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4461 * config/arm/arm-cpus.in (armv8_4): New feature.
4462 (ARMv8_4a): New fgroup.
4463 (armv8.4-a): New arch.
4464 * config/arm/arm-tables.opt: Regenerate.
4465 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4466 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4467 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4468 Add matching rules for -march=armv8.4-a and extensions.
4469 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4471 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4474 * config/rx/rx.md (BW): New mode attribute.
4475 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4477 2018-01-11 Richard Biener <rguenther@suse.de>
4479 PR tree-optimization/83435
4480 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4481 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4482 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4484 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4485 Alan Hayward <alan.hayward@arm.com>
4486 David Sherwood <david.sherwood@arm.com>
4488 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4490 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4491 (aarch64_print_address_internal): Use it to check for a zero offset.
4493 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4494 Alan Hayward <alan.hayward@arm.com>
4495 David Sherwood <david.sherwood@arm.com>
4497 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4498 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4499 Return a poly_int64 rather than a HOST_WIDE_INT.
4500 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4501 rather than a HOST_WIDE_INT.
4502 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4503 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4504 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4505 final_offset from HOST_WIDE_INT to poly_int64.
4506 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4507 to_constant when getting the number of units in an Advanced SIMD
4509 (aarch64_builtin_vectorized_function): Check for a constant number
4511 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4513 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4514 attribute instead of GET_MODE_NUNITS.
4515 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4516 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4517 GET_MODE_SIZE for fixed-size registers.
4518 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4519 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4520 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4521 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4522 (aarch64_print_operand, aarch64_print_address_internal)
4523 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4524 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4525 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4526 Handle polynomial GET_MODE_SIZE.
4527 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4528 wider than SImode without modification.
4529 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4530 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4531 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4532 passing and returning SVE modes.
4533 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4534 rather than GEN_INT.
4535 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4536 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4537 (aarch64_allocate_and_probe_stack_space): Likewise.
4538 (aarch64_layout_frame): Cope with polynomial offsets.
4539 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4540 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4542 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4543 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4544 poly_int64 rather than a HOST_WIDE_INT.
4545 (aarch64_get_separate_components, aarch64_process_components)
4546 (aarch64_expand_prologue, aarch64_expand_epilogue)
4547 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4548 (aarch64_anchor_offset): New function, split out from...
4549 (aarch64_legitimize_address): ...here.
4550 (aarch64_builtin_vectorization_cost): Handle polynomial
4551 TYPE_VECTOR_SUBPARTS.
4552 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4554 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4555 number of elements from the PARALLEL rather than the mode.
4556 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4557 rather than GET_MODE_BITSIZE.
4558 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4559 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4560 (aarch64_expand_vec_perm_const_1): Handle polynomial
4561 d->perm.length () and d->perm elements.
4562 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4563 Apply to_constant to d->perm elements.
4564 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4565 polynomial CONST_VECTOR_NUNITS.
4566 (aarch64_move_pointer): Take amount as a poly_int64 rather
4568 (aarch64_progress_pointer): Avoid temporary variable.
4569 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4570 the mode attribute instead of GET_MODE.
4572 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4573 Alan Hayward <alan.hayward@arm.com>
4574 David Sherwood <david.sherwood@arm.com>
4576 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4577 x exists before using it.
4578 (aarch64_add_constant_internal): Rename to...
4579 (aarch64_add_offset_1): ...this. Replace regnum with separate
4580 src and dest rtxes. Handle the case in which they're different,
4581 including when the offset is zero. Replace scratchreg with an rtx.
4582 Use 2 additions if there is no spare register into which we can
4583 move a 16-bit constant.
4584 (aarch64_add_constant): Delete.
4585 (aarch64_add_offset): Replace reg with separate src and dest
4586 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4587 Use aarch64_add_offset_1.
4588 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4589 an rtx rather than an int. Take the delta as a poly_int64
4590 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4591 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4592 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4593 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4594 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4596 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4597 aarch64_add_constant.
4599 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4601 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4602 Use scalar_float_mode.
4604 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4606 * config/aarch64/aarch64-simd.md
4607 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4608 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4609 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4610 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4611 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4612 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4613 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4614 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4615 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4616 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4618 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4621 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4622 targ_options->x_arm_arch_string is non NULL.
4624 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4626 * config/aarch64/aarch64.h
4627 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4629 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4632 * expmed.c (emit_store_flag_force): Swap if const op0
4633 and change VOIDmode to mode of op0.
4635 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4637 PR rtl-optimization/83761
4638 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4639 than bytes to mode_for_size.
4641 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4644 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4645 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4648 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4651 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4652 when in layout mode.
4653 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4654 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4657 2018-01-10 Michael Collison <michael.collison@arm.com>
4659 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4660 * config/aarch64/aarch64-option-extension.def: Add
4661 AARCH64_OPT_EXTENSION of 'fp16fml'.
4662 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4663 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4664 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4665 * config/aarch64/constraints.md (Ui7): New constraint.
4666 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4667 (VFMLA_SEL_W): Ditto.
4670 (VFMLA16_LOW): New int iterator.
4671 (VFMLA16_HIGH): Ditto.
4672 (UNSPEC_FMLAL): New unspec.
4673 (UNSPEC_FMLSL): Ditto.
4674 (UNSPEC_FMLAL2): Ditto.
4675 (UNSPEC_FMLSL2): Ditto.
4676 (f16mac): New code attribute.
4677 * config/aarch64/aarch64-simd-builtins.def
4678 (aarch64_fmlal_lowv2sf): Ditto.
4679 (aarch64_fmlsl_lowv2sf): Ditto.
4680 (aarch64_fmlalq_lowv4sf): Ditto.
4681 (aarch64_fmlslq_lowv4sf): Ditto.
4682 (aarch64_fmlal_highv2sf): Ditto.
4683 (aarch64_fmlsl_highv2sf): Ditto.
4684 (aarch64_fmlalq_highv4sf): Ditto.
4685 (aarch64_fmlslq_highv4sf): Ditto.
4686 (aarch64_fmlal_lane_lowv2sf): Ditto.
4687 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4688 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4689 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4690 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4691 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4692 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4693 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4694 (aarch64_fmlal_lane_highv2sf): Ditto.
4695 (aarch64_fmlsl_lane_highv2sf): Ditto.
4696 (aarch64_fmlal_laneq_highv2sf): Ditto.
4697 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4698 (aarch64_fmlalq_lane_highv4sf): Ditto.
4699 (aarch64_fmlsl_lane_highv4sf): Ditto.
4700 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4701 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4702 * config/aarch64/aarch64-simd.md:
4703 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4704 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4705 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4706 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4707 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4708 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4709 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4710 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4711 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4712 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4713 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4714 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4715 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4716 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4717 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4718 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4719 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4720 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4721 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4722 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4723 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4724 (vfmlsl_low_u32): Ditto.
4725 (vfmlalq_low_u32): Ditto.
4726 (vfmlslq_low_u32): Ditto.
4727 (vfmlal_high_u32): Ditto.
4728 (vfmlsl_high_u32): Ditto.
4729 (vfmlalq_high_u32): Ditto.
4730 (vfmlslq_high_u32): Ditto.
4731 (vfmlal_lane_low_u32): Ditto.
4732 (vfmlsl_lane_low_u32): Ditto.
4733 (vfmlal_laneq_low_u32): Ditto.
4734 (vfmlsl_laneq_low_u32): Ditto.
4735 (vfmlalq_lane_low_u32): Ditto.
4736 (vfmlslq_lane_low_u32): Ditto.
4737 (vfmlalq_laneq_low_u32): Ditto.
4738 (vfmlslq_laneq_low_u32): Ditto.
4739 (vfmlal_lane_high_u32): Ditto.
4740 (vfmlsl_lane_high_u32): Ditto.
4741 (vfmlal_laneq_high_u32): Ditto.
4742 (vfmlsl_laneq_high_u32): Ditto.
4743 (vfmlalq_lane_high_u32): Ditto.
4744 (vfmlslq_lane_high_u32): Ditto.
4745 (vfmlalq_laneq_high_u32): Ditto.
4746 (vfmlslq_laneq_high_u32): Ditto.
4747 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4748 (AARCH64_FL_FOR_ARCH8_4): New.
4749 (AARCH64_ISA_F16FML): New ISA flag.
4750 (TARGET_F16FML): New feature flag for fp16fml.
4751 (doc/invoke.texi): Document new fp16fml option.
4753 2018-01-10 Michael Collison <michael.collison@arm.com>
4755 * config/aarch64/aarch64-builtins.c:
4756 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4757 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4758 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4759 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4760 (AARCH64_ISA_SHA3): New ISA flag.
4761 (TARGET_SHA3): New feature flag for sha3.
4762 * config/aarch64/iterators.md (sha512_op): New int attribute.
4763 (CRYPTO_SHA512): New int iterator.
4764 (UNSPEC_SHA512H): New unspec.
4765 (UNSPEC_SHA512H2): Ditto.
4766 (UNSPEC_SHA512SU0): Ditto.
4767 (UNSPEC_SHA512SU1): Ditto.
4768 * config/aarch64/aarch64-simd-builtins.def
4769 (aarch64_crypto_sha512hqv2di): New builtin.
4770 (aarch64_crypto_sha512h2qv2di): Ditto.
4771 (aarch64_crypto_sha512su0qv2di): Ditto.
4772 (aarch64_crypto_sha512su1qv2di): Ditto.
4773 (aarch64_eor3qv8hi): Ditto.
4774 (aarch64_rax1qv2di): Ditto.
4775 (aarch64_xarqv2di): Ditto.
4776 (aarch64_bcaxqv8hi): Ditto.
4777 * config/aarch64/aarch64-simd.md:
4778 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4779 (aarch64_crypto_sha512su0qv2di): Ditto.
4780 (aarch64_crypto_sha512su1qv2di): Ditto.
4781 (aarch64_eor3qv8hi): Ditto.
4782 (aarch64_rax1qv2di): Ditto.
4783 (aarch64_xarqv2di): Ditto.
4784 (aarch64_bcaxqv8hi): Ditto.
4785 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4786 (vsha512h2q_u64): Ditto.
4787 (vsha512su0q_u64): Ditto.
4788 (vsha512su1q_u64): Ditto.
4789 (veor3q_u16): Ditto.
4790 (vrax1q_u64): Ditto.
4792 (vbcaxq_u16): Ditto.
4793 * config/arm/types.md (crypto_sha512): New type attribute.
4794 (crypto_sha3): Ditto.
4795 (doc/invoke.texi): Document new sha3 option.
4797 2018-01-10 Michael Collison <michael.collison@arm.com>
4799 * config/aarch64/aarch64-builtins.c:
4800 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4801 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4802 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4803 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4804 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4805 (AARCH64_ISA_SM4): New ISA flag.
4806 (TARGET_SM4): New feature flag for sm4.
4807 * config/aarch64/aarch64-simd-builtins.def
4808 (aarch64_sm3ss1qv4si): Ditto.
4809 (aarch64_sm3tt1aq4si): Ditto.
4810 (aarch64_sm3tt1bq4si): Ditto.
4811 (aarch64_sm3tt2aq4si): Ditto.
4812 (aarch64_sm3tt2bq4si): Ditto.
4813 (aarch64_sm3partw1qv4si): Ditto.
4814 (aarch64_sm3partw2qv4si): Ditto.
4815 (aarch64_sm4eqv4si): Ditto.
4816 (aarch64_sm4ekeyqv4si): Ditto.
4817 * config/aarch64/aarch64-simd.md:
4818 (aarch64_sm3ss1qv4si): Ditto.
4819 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4820 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4821 (aarch64_sm4eqv4si): Ditto.
4822 (aarch64_sm4ekeyqv4si): Ditto.
4823 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4824 (sm3part_op): Ditto.
4825 (CRYPTO_SM3TT): Ditto.
4826 (CRYPTO_SM3PART): Ditto.
4827 (UNSPEC_SM3SS1): New unspec.
4828 (UNSPEC_SM3TT1A): Ditto.
4829 (UNSPEC_SM3TT1B): Ditto.
4830 (UNSPEC_SM3TT2A): Ditto.
4831 (UNSPEC_SM3TT2B): Ditto.
4832 (UNSPEC_SM3PARTW1): Ditto.
4833 (UNSPEC_SM3PARTW2): Ditto.
4834 (UNSPEC_SM4E): Ditto.
4835 (UNSPEC_SM4EKEY): Ditto.
4836 * config/aarch64/constraints.md (Ui2): New constraint.
4837 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4838 * config/arm/types.md (crypto_sm3): New type attribute.
4839 (crypto_sm4): Ditto.
4840 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4841 (vsm3tt1aq_u32): Ditto.
4842 (vsm3tt1bq_u32): Ditto.
4843 (vsm3tt2aq_u32): Ditto.
4844 (vsm3tt2bq_u32): Ditto.
4845 (vsm3partw1q_u32): Ditto.
4846 (vsm3partw2q_u32): Ditto.
4847 (vsm4eq_u32): Ditto.
4848 (vsm4ekeyq_u32): Ditto.
4849 (doc/invoke.texi): Document new sm4 option.
4851 2018-01-10 Michael Collison <michael.collison@arm.com>
4853 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4854 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4855 (AARCH64_FL_FOR_ARCH8_4): New.
4856 (AARCH64_FL_V8_4): New flag.
4857 (doc/invoke.texi): Document new armv8.4-a option.
4859 2018-01-10 Michael Collison <michael.collison@arm.com>
4861 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4862 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4863 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4864 * config/aarch64/aarch64-option-extension.def: Add
4865 AARCH64_OPT_EXTENSION of 'sha2'.
4866 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4867 (crypto): Disable sha2 and aes if crypto disabled.
4868 (crypto): Enable aes and sha2 if enabled.
4869 (simd): Disable sha2 and aes if simd disabled.
4870 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4872 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4873 (TARGET_SHA2): New feature flag for sha2.
4874 (TARGET_AES): New feature flag for aes.
4875 * config/aarch64/aarch64-simd.md:
4876 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4877 conditional on TARGET_AES.
4878 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4879 (aarch64_crypto_sha1hsi): Make pattern conditional
4881 (aarch64_crypto_sha1hv4si): Ditto.
4882 (aarch64_be_crypto_sha1hv4si): Ditto.
4883 (aarch64_crypto_sha1su1v4si): Ditto.
4884 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4885 (aarch64_crypto_sha1su0v4si): Ditto.
4886 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4887 (aarch64_crypto_sha256su0v4si): Ditto.
4888 (aarch64_crypto_sha256su1v4si): Ditto.
4889 (doc/invoke.texi): Document new aes and sha2 options.
4891 2018-01-10 Martin Sebor <msebor@redhat.com>
4893 PR tree-optimization/83781
4894 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4897 2018-01-11 Martin Sebor <msebor@gmail.com>
4898 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4900 PR tree-optimization/83501
4901 PR tree-optimization/81703
4903 * tree-ssa-strlen.c (get_string_cst): Rename...
4904 (get_string_len): ...to this. Handle global constants.
4905 (handle_char_store): Adjust.
4907 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4908 Jim Wilson <jimw@sifive.com>
4910 * config/riscv/riscv-protos.h (riscv_output_return): New.
4911 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4912 (riscv_attribute_table, riscv_output_return),
4913 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4914 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4915 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4917 (riscv_expand_prologue): Add early return for naked function.
4918 (riscv_expand_epilogue): Likewise.
4919 (riscv_function_ok_for_sibcall): Return false for naked function.
4920 (riscv_set_current_function): New.
4921 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4922 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4923 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4924 * doc/extend.texi (RISC-V Function Attributes): New.
4926 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4928 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4929 check for 128-bit long double before checking TCmode.
4930 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4931 128-bit long doubles before checking TFmode or TCmode.
4932 (FLOAT128_IBM_P): Likewise.
4934 2018-01-10 Martin Sebor <msebor@redhat.com>
4936 PR tree-optimization/83671
4937 * builtins.c (c_strlen): Unconditionally return zero for the empty
4939 Use -Warray-bounds for warnings.
4940 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4941 for non-constant array indices with COMPONENT_REF, arrays of
4942 arrays, and pointers to arrays.
4943 (gimple_fold_builtin_strlen): Determine and set length range for
4944 non-constant character arrays.
4946 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4949 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4952 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4954 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4956 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4959 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4960 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4961 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4962 indexed_or_indirect_operand predicate.
4963 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4964 (*vsx_le_perm_load_v8hi): Likewise.
4965 (*vsx_le_perm_load_v16qi): Likewise.
4966 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4967 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4968 (*vsx_le_perm_store_v8hi): Likewise.
4969 (*vsx_le_perm_store_v16qi): Likewise.
4970 (eight unnamed splitters): Likewise.
4972 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4974 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4975 * config/rs6000/emmintrin.h: Likewise.
4976 * config/rs6000/mmintrin.h: Likewise.
4977 * config/rs6000/xmmintrin.h: Likewise.
4979 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4982 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4984 * tree.c (tree_nop_conversion): Return true for location wrapper
4986 (maybe_wrap_with_location): New function.
4987 (selftest::check_strip_nops): New function.
4988 (selftest::test_location_wrappers): New function.
4989 (selftest::tree_c_tests): Call it.
4990 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4991 (maybe_wrap_with_location): New decl.
4992 (EXPR_LOCATION_WRAPPER_P): New macro.
4993 (location_wrapper_p): New inline function.
4994 (tree_strip_any_location_wrapper): New inline function.
4996 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4999 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5000 stack_realign_offset for the largest alignment of stack slot
5002 (ix86_find_max_used_stack_alignment): New function.
5003 (ix86_finalize_stack_frame_flags): Use it. Set
5004 max_used_stack_alignment if we don't realign stack.
5005 * config/i386/i386.h (machine_function): Add
5006 max_used_stack_alignment.
5008 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5010 * config/arm/arm.opt (-mbranch-cost): New option.
5011 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5014 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5017 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5018 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5020 2018-01-10 Richard Biener <rguenther@suse.de>
5023 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5024 early out so it also covers the case where we have a non-NULL
5027 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5029 PR tree-optimization/83753
5030 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5031 for non-strided grouped accesses if the number of elements is 1.
5033 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5036 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5037 * i386.h (TARGET_USE_GATHER): Define.
5038 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5040 2018-01-10 Martin Liska <mliska@suse.cz>
5043 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5044 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5046 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5047 CLEANUP_NO_PARTITIONING is not set.
5049 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5051 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5052 for vectors, as a partial revert of r254296.
5053 * rtl.h (const_vec_p): Delete.
5054 (const_vec_duplicate_p): Don't test for vector CONSTs.
5055 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5056 * expmed.c (make_tree): Likewise.
5059 * common.md (E, F): Use CONSTANT_P instead of checking for
5061 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5062 checking for CONST_VECTOR.
5064 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5067 * predict.c (force_edge_cold): Handle in more sane way edges
5070 2018-01-09 Carl Love <cel@us.ibm.com>
5072 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5074 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5075 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5076 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5077 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5078 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5079 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5080 * config/rs6000/rs6000-protos.h: Add extern defition for
5081 rs6000_generate_float2_double_code.
5082 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5084 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5085 (float2_v2df): Add define_expand.
5087 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5090 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5091 op_mode in the force_to_mode call.
5093 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5095 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5096 instead of checking each element individually.
5097 (aarch64_evpc_uzp): Likewise.
5098 (aarch64_evpc_zip): Likewise.
5099 (aarch64_evpc_ext): Likewise.
5100 (aarch64_evpc_rev): Likewise.
5101 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5102 instead of checking each element individually. Return true without
5104 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5105 whether all selected elements come from the same input, instead of
5106 checking each element individually. Remove calls to gen_rtx_REG,
5107 start_sequence and end_sequence and instead assert that no rtl is
5110 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5112 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5113 order of HIGH and CONST checks.
5115 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5117 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5118 if the destination isn't an SSA_NAME.
5120 2018-01-09 Richard Biener <rguenther@suse.de>
5122 PR tree-optimization/83668
5123 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5125 (canonicalize_loop_form): ... here, renamed from ...
5126 (canonicalize_loop_closed_ssa_form): ... this and amended to
5127 swap successor edges for loop exit blocks to make us use
5128 the RPO order we need for initial schedule generation.
5130 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5132 PR tree-optimization/64811
5133 * match.pd: When optimizing comparisons with Inf, avoid
5134 introducing or losing exceptions from comparisons with NaN.
5136 2018-01-09 Martin Liska <mliska@suse.cz>
5139 * asan.c (shadow_mem_size): Add gcc_assert.
5141 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5143 Don't save registers in main().
5146 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5147 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5148 * config/avr/avr.c (avr_set_current_function): Don't error if
5149 naked, OS_task or OS_main are specified at the same time.
5150 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5152 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5154 * common/config/avr/avr-common.c (avr_option_optimization_table):
5155 Switch on -mmain-is-OS_task for optimizing compilations.
5157 2018-01-09 Richard Biener <rguenther@suse.de>
5159 PR tree-optimization/83572
5160 * graphite.c: Include cfganal.h.
5161 (graphite_transform_loops): Connect infinite loops to exit
5162 and remove fake edges at the end.
5164 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5166 * ipa-inline.c (edge_badness): Revert accidental checkin.
5168 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5171 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5172 symbols; not inline clones.
5174 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5177 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5178 hard registers. Formatting fixes.
5180 PR preprocessor/83722
5181 * gcc.c (try_generate_repro): Pass
5182 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5183 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5186 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5187 Kito Cheng <kito.cheng@gmail.com>
5189 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5190 (riscv_leaf_function_p): Delete.
5191 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5193 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5195 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5197 (do_ifelse): New function.
5198 (do_isel): New function.
5199 (do_sub3): New function.
5200 (do_add3): New function.
5201 (do_load_mask_compare): New function.
5202 (do_overlap_load_compare): New function.
5203 (expand_compare_loop): New function.
5204 (expand_block_compare): Call expand_compare_loop() when appropriate.
5205 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5207 (-mblock-compare-inline-loop-limit): New option.
5209 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5212 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5213 Reverse order of second and third operands in first alternative.
5214 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5215 of first and second elements in UNSPEC_VPERMR vector.
5216 (altivec_expand_vec_perm_le): Likewise.
5218 2018-01-08 Jeff Law <law@redhat.com>
5220 PR rtl-optimizatin/81308
5221 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5222 (process_switch): If group_case_labels makes a change, then set
5224 (pass_convert_switch::execute): If a switch is converted, then
5225 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5227 PR rtl-optimization/81308
5228 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5231 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5233 PR target/83663 - Revert r255946
5234 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5235 generation for cases where splatting a value is not useful.
5236 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5237 across a vec_duplicate and a paradoxical subreg forming a vector
5238 mode to a vec_concat.
5240 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5242 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5243 -march=armv8.3-a variants.
5244 * config/arm/t-multilib: Likewise.
5245 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5247 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5249 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5251 (cceq_ior_compare_complement): Give it a name so I can use it, and
5252 change boolean_or_operator predicate to boolean_operator so it can
5253 be used to generate a crand.
5254 (eqne): New code iterator.
5255 (bd/bd_neg): New code_attrs.
5256 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5257 a single define_insn.
5258 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5259 decrement (bdnzt/bdnzf/bdzt/bdzf).
5260 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5261 with the new names of the branch decrement patterns, and added the
5262 names of the branch decrement conditional patterns.
5264 2018-01-08 Richard Biener <rguenther@suse.de>
5266 PR tree-optimization/83563
5267 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5270 2018-01-08 Richard Biener <rguenther@suse.de>
5273 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5275 2018-01-08 Richard Biener <rguenther@suse.de>
5277 PR tree-optimization/83685
5278 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5279 references to abnormals.
5281 2018-01-08 Richard Biener <rguenther@suse.de>
5284 * dwarf2out.c (output_indirect_strings): Handle empty
5285 skeleton_debug_str_hash.
5286 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5288 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5290 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5291 (emit_store_direct): Likewise.
5292 (arc_trampoline_adjust_address): Likewise.
5293 (arc_asm_trampoline_template): New function.
5294 (arc_initialize_trampoline): Use asm_trampoline_template.
5295 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5296 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5297 * config/arc/arc.md (flush_icache): Delete pattern.
5299 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5301 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5302 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5305 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5308 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5309 by not USED_FOR_TARGET.
5310 (make_pass_resolve_sw_modes): Likewise.
5312 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5314 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5317 2018-01-08 Richard Biener <rguenther@suse.de>
5320 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5322 2018-01-08 Richard Biener <rguenther@suse.de>
5325 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5327 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5330 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5331 basic blocks with a small number of successors.
5332 (convert_control_dep_chain_into_preds): Improve handling of
5334 (dump_predicates): Split apart into...
5335 (dump_pred_chain): ...here...
5336 (dump_pred_info): ...and here.
5337 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5338 (can_chain_union_be_invalidated_p): Improve check for invalidation
5340 (uninit_uses_cannot_happen): Avoid unnecessary if
5341 convert_control_dep_chain_into_preds yielded nothing.
5343 2018-01-06 Martin Sebor <msebor@redhat.com>
5345 PR tree-optimization/83640
5346 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5347 subtracting negative offset from size.
5348 (builtin_access::overlap): Adjust offset bounds of the access to fall
5349 within the size of the object if possible.
5351 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5353 PR rtl-optimization/83699
5354 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5355 extract_bit_field_as_subreg to cases in which the extracted
5356 value is also a vector.
5358 * lra-constraints.c (process_alt_operands): Test for the equivalence
5359 substitutions when detecting a possible reload cycle.
5361 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5364 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5365 by default if flag_selective_schedling{,2}. Formatting fixes.
5367 PR rtl-optimization/83682
5368 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5369 if it has non-VECTOR_MODE element mode.
5370 (vec_duplicate_p): Likewise.
5373 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5374 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5376 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5379 * config/i386/i386-builtin.def
5380 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5381 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5382 Require also OPTION_MASK_ISA_AVX512F in addition to
5383 OPTION_MASK_ISA_GFNI.
5384 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5385 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5386 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5387 to OPTION_MASK_ISA_GFNI.
5388 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5389 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5390 OPTION_MASK_ISA_AVX512BW.
5391 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5392 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5393 addition to OPTION_MASK_ISA_GFNI.
5394 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5395 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5396 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5397 to OPTION_MASK_ISA_GFNI.
5398 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5399 a requirement for all ISAs rather than any of them with a few
5401 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5403 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5404 bitmasks to be enabled with 3 exceptions, instead of requiring any
5405 enabled ISA with lots of exceptions.
5406 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5407 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5408 Change avx512bw in isa attribute to avx512f.
5409 * config/i386/sgxintrin.h: Add license boilerplate.
5410 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5411 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5412 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5413 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5415 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5416 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5417 temporarily sse2 rather than sse if not enabled already.
5420 * config/i386/sse.md (VI248_VLBW): Rename to ...
5421 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5422 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5423 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5424 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5425 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5426 mode iterator instead of VI248_VLBW.
5428 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5430 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5431 (record_modified): Skip clobbers; add debug output.
5432 (param_change_prob): Use sreal frequencies.
5434 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5436 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5437 punt for user-aligned variables.
5439 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5441 * tree-chrec.c (chrec_contains_symbols): Return true for
5444 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5447 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5448 of (x|y) == x for BICS pattern.
5450 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5452 PR tree-optimization/83605
5453 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5454 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5457 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5459 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5460 * config/epiphany/rtems.h: New file.
5462 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5463 Uros Bizjak <ubizjak@gmail.com>
5466 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5467 QIreg_operand instead of register_operand predicate.
5468 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5469 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5470 comments instead of -fmitigate[-_]rop.
5472 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5475 * cgraphunit.c (symbol_table::compile): Switch to text_section
5476 before calling assembly_start debug hook.
5477 * run-rtl-passes.c (run_rtl_passes): Likewise.
5480 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5482 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5483 range_int_cst_p rather than !symbolic_range_p before calling
5484 extract_range_from_multiplicative_op_1.
5486 2018-01-04 Jeff Law <law@redhat.com>
5488 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5489 redundant test in assertion.
5491 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5493 * doc/rtl.texi: Document machine_mode wrapper classes.
5495 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5497 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5500 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5502 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5503 the VEC_PERM_EXPR fold to fail.
5505 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5508 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5509 to switched_sections.
5511 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5514 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5517 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5520 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5521 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5523 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5526 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5527 is BLKmode and bitpos not zero or mode change is needed.
5529 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5532 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5535 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5538 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5539 instead of MULT rtx. Update all corresponding splitters.
5541 (*ssub<modesuffix>): Ditto.
5543 (*cmp_sadd_di): Update split patterns.
5544 (*cmp_sadd_si): Ditto.
5545 (*cmp_sadd_sidi): Ditto.
5546 (*cmp_ssub_di): Ditto.
5547 (*cmp_ssub_si): Ditto.
5548 (*cmp_ssub_sidi): Ditto.
5549 * config/alpha/predicates.md (const23_operand): New predicate.
5550 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5551 Look for ASHIFT, not MULT inner operand.
5552 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5554 2018-01-04 Martin Liska <mliska@suse.cz>
5556 PR gcov-profile/83669
5557 * gcov.c (output_intermediate_file): Add version to intermediate
5559 * doc/gcov.texi: Document new field 'version' in intermediate
5560 file format. Fix location of '-k' option of gcov command.
5562 2018-01-04 Martin Liska <mliska@suse.cz>
5565 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5567 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5569 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5571 2018-01-03 Martin Sebor <msebor@redhat.com>
5573 PR tree-optimization/83655
5574 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5575 checking calls with invalid arguments.
5577 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5579 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5580 (vectorizable_mask_load_store): Delete.
5581 (vectorizable_call): Return false for masked loads and stores.
5582 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5583 instead of gimple_assign_rhs1.
5584 (vectorizable_load): Handle IFN_MASK_LOAD.
5585 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5587 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5589 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5591 (vectorizable_mask_load_store): ...here.
5592 (vectorizable_load): ...and here.
5594 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5596 * tree-vect-stmts.c (vect_build_all_ones_mask)
5597 (vect_build_zero_merge_argument): New functions, split out from...
5598 (vectorizable_load): ...here.
5600 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5602 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5604 (vectorizable_mask_load_store): ...here.
5605 (vectorizable_store): ...and here.
5607 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5609 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5611 (vectorizable_mask_load_store): ...here.
5613 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5615 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5616 (vect_model_store_cost): Take a vec_load_store_type instead of a
5618 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5619 (vect_model_store_cost): Take a vec_load_store_type instead of a
5621 (vectorizable_mask_load_store): Update accordingly.
5622 (vectorizable_store): Likewise.
5623 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5625 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5627 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5628 IFN_MASK_LOAD calls here rather than...
5629 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5631 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5632 Alan Hayward <alan.hayward@arm.com>
5633 David Sherwood <david.sherwood@arm.com>
5635 * expmed.c (extract_bit_field_1): For vector extracts,
5636 fall back to extract_bit_field_as_subreg if vec_extract
5639 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5640 Alan Hayward <alan.hayward@arm.com>
5641 David Sherwood <david.sherwood@arm.com>
5643 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5644 they are variable or constant sized.
5645 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5646 slots for constant-sized data.
5648 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5649 Alan Hayward <alan.hayward@arm.com>
5650 David Sherwood <david.sherwood@arm.com>
5652 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5653 handling COND_EXPRs with boolean comparisons, try to find a better
5654 basis for the mask type than the boolean itself.
5656 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5658 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5659 is calculated and how it can be overridden.
5660 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5661 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5663 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5666 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5667 Alan Hayward <alan.hayward@arm.com>
5668 David Sherwood <david.sherwood@arm.com>
5670 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5671 Remove the mode argument.
5672 (aarch64_simd_valid_immediate): Remove the mode and inverse
5674 * config/aarch64/iterators.md (bitsize): New iterator.
5675 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5676 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5677 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5678 aarch64_simd_valid_immediate.
5679 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5680 (aarch64_reg_or_bic_imm): Likewise.
5681 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5682 with an insn_type enum and msl with a modifier_type enum.
5683 Replace element_width with a scalar_mode. Change the shift
5684 to unsigned int. Add constructors for scalar_float_mode and
5685 scalar_int_mode elements.
5686 (aarch64_vect_float_const_representable_p): Delete.
5687 (aarch64_can_const_movi_rtx_p)
5688 (aarch64_simd_scalar_immediate_valid_for_move)
5689 (aarch64_simd_make_constant): Update call to
5690 aarch64_simd_valid_immediate.
5691 (aarch64_advsimd_valid_immediate_hs): New function.
5692 (aarch64_advsimd_valid_immediate): Likewise.
5693 (aarch64_simd_valid_immediate): Remove mode and inverse
5694 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5695 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5696 and aarch64_float_const_representable_p on the result.
5697 (aarch64_output_simd_mov_immediate): Remove mode argument.
5698 Update call to aarch64_simd_valid_immediate and use of
5699 simd_immediate_info.
5700 (aarch64_output_scalar_simd_mov_immediate): Update call
5703 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5704 Alan Hayward <alan.hayward@arm.com>
5705 David Sherwood <david.sherwood@arm.com>
5707 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5708 (mode_nunits): Likewise CONST_MODE_NUNITS.
5709 * machmode.def (ADJUST_NUNITS): Document.
5710 * genmodes.c (mode_data::need_nunits_adj): New field.
5711 (blank_mode): Update accordingly.
5712 (adj_nunits): New variable.
5713 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5715 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5716 listed in adj_nunits.
5717 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5718 listed in adj_nunits. Don't emit case statements for such modes.
5719 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5720 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5721 nothing if adj_nunits is nonnull.
5722 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5723 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5724 (emit_mode_fbit): Update use of print_maybe_const_decl.
5725 (emit_move_size): Likewise. Treat the array as non-const
5727 (emit_mode_adjustments): Handle adj_nunits.
5729 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5731 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5732 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5733 (VECTOR_MODES): Use it.
5734 (make_vector_modes): Take the prefix as an argument.
5736 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5737 Alan Hayward <alan.hayward@arm.com>
5738 David Sherwood <david.sherwood@arm.com>
5740 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5741 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5742 for MODE_VECTOR_BOOL.
5743 * machmode.def (VECTOR_BOOL_MODE): Document.
5744 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5745 (make_vector_bool_mode): New function.
5746 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5748 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5749 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5751 * stor-layout.c (int_mode_for_mode): Likewise.
5752 * tree.c (build_vector_type_for_mode): Likewise.
5753 * varasm.c (output_constant_pool_2): Likewise.
5754 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5755 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5756 for MODE_VECTOR_BOOL.
5757 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5758 of mode class checks.
5759 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5760 instead of a list of mode class checks.
5761 (expand_vector_scalar_condition): Likewise.
5762 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5764 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5765 Alan Hayward <alan.hayward@arm.com>
5766 David Sherwood <david.sherwood@arm.com>
5768 * machmode.h (mode_size): Change from unsigned short to
5770 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5771 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5772 or if measurement_type is not polynomial.
5773 (fixed_size_mode::includes_p): Check for constant-sized modes.
5774 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5775 return a poly_uint16 rather than an unsigned short.
5776 (emit_mode_size): Change the type of mode_size from unsigned short
5777 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5778 (emit_mode_adjustments): Cope with polynomial vector sizes.
5779 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5781 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5783 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5784 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5785 * caller-save.c (setup_save_areas): Likewise.
5786 (replace_reg_with_saved_mem): Likewise.
5787 * calls.c (emit_library_call_value_1): Likewise.
5788 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5789 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5790 (gen_lowpart_for_combine): Likewise.
5791 * convert.c (convert_to_integer_1): Likewise.
5792 * cse.c (equiv_constant, cse_insn): Likewise.
5793 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5794 (cselib_subst_to_values): Likewise.
5795 * dce.c (word_dce_process_block): Likewise.
5796 * df-problems.c (df_word_lr_mark_ref): Likewise.
5797 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5798 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5799 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5800 (rtl_for_decl_location): Likewise.
5801 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5802 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5803 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5804 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5805 (expand_expr_real_1): Likewise.
5806 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5807 (pad_below): Likewise.
5808 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5809 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5810 * ira.c (get_subreg_tracking_sizes): Likewise.
5811 * ira-build.c (ira_create_allocno_objects): Likewise.
5812 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5813 (ira_sort_regnos_for_alter_reg): Likewise.
5814 * ira-costs.c (record_operand_costs): Likewise.
5815 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5816 (resolve_simple_move): Likewise.
5817 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5818 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5819 (lra_constraints): Likewise.
5820 (CONST_POOL_OK_P): Reject variable-sized modes.
5821 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5822 (add_pseudo_to_slot, lra_spill): Likewise.
5823 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5824 * optabs-query.c (get_best_extraction_insn): Likewise.
5825 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5826 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5827 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5828 * recog.c (offsettable_address_addr_space_p): Likewise.
5829 * regcprop.c (maybe_mode_change): Likewise.
5830 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5831 * regrename.c (build_def_use): Likewise.
5832 * regstat.c (dump_reg_info): Likewise.
5833 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5834 (find_reloads, find_reloads_subreg_address): Likewise.
5835 * reload1.c (eliminate_regs_1): Likewise.
5836 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5837 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5838 (simplify_binary_operation_1, simplify_subreg): Likewise.
5839 * targhooks.c (default_function_arg_padding): Likewise.
5840 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5841 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5842 (verify_gimple_assign_ternary): Likewise.
5843 * tree-inline.c (estimate_move_cost): Likewise.
5844 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5845 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5846 (get_address_cost_ainc): Likewise.
5847 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5848 (vect_supportable_dr_alignment): Likewise.
5849 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5850 (vectorizable_reduction): Likewise.
5851 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5852 (vectorizable_operation, vectorizable_load): Likewise.
5853 * tree.c (build_same_sized_truth_vector_type): Likewise.
5854 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5855 * var-tracking.c (emit_note_insn_var_location): Likewise.
5856 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5857 (ADDR_VEC_ALIGN): Likewise.
5859 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5860 Alan Hayward <alan.hayward@arm.com>
5861 David Sherwood <david.sherwood@arm.com>
5863 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5865 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5866 or if measurement_type is polynomial.
5867 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5868 * combine.c (make_extraction): Likewise.
5869 * dse.c (find_shift_sequence): Likewise.
5870 * dwarf2out.c (mem_loc_descriptor): Likewise.
5871 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5872 (extract_bit_field, extract_low_bits): Likewise.
5873 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5874 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5875 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5876 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5877 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5878 * reload.c (find_reloads): Likewise.
5879 * reload1.c (alter_reg): Likewise.
5880 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5881 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5882 * tree-if-conv.c (predicate_mem_writes): Likewise.
5883 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5884 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5885 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5886 * valtrack.c (dead_debug_insert_temp): Likewise.
5887 * varasm.c (mergeable_constant_section): Likewise.
5888 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5890 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5891 Alan Hayward <alan.hayward@arm.com>
5892 David Sherwood <david.sherwood@arm.com>
5894 * expr.c (expand_assignment): Cope with polynomial mode sizes
5895 when assigning to a CONCAT.
5897 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5898 Alan Hayward <alan.hayward@arm.com>
5899 David Sherwood <david.sherwood@arm.com>
5901 * machmode.h (mode_precision): Change from unsigned short to
5903 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5905 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5906 or if measurement_type is not polynomial.
5907 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5908 in which the mode is already known to be a scalar_int_mode.
5909 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5910 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5912 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5913 for GET_MODE_PRECISION.
5914 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5915 for GET_MODE_PRECISION.
5916 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5918 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5919 (expand_field_assignment, make_extraction): Likewise.
5920 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5921 (get_last_value): Likewise.
5922 * convert.c (convert_to_integer_1): Likewise.
5923 * cse.c (cse_insn): Likewise.
5924 * expr.c (expand_expr_real_1): Likewise.
5925 * lra-constraints.c (simplify_operand_subreg): Likewise.
5926 * optabs-query.c (can_atomic_load_p): Likewise.
5927 * optabs.c (expand_atomic_load): Likewise.
5928 (expand_atomic_store): Likewise.
5929 * ree.c (combine_reaching_defs): Likewise.
5930 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5931 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5932 * tree.h (type_has_mode_precision_p): Likewise.
5933 * ubsan.c (instrument_si_overflow): Likewise.
5935 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5936 Alan Hayward <alan.hayward@arm.com>
5937 David Sherwood <david.sherwood@arm.com>
5939 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5940 polynomial numbers of units.
5941 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5942 (valid_vector_subparts_p): New function.
5943 (build_vector_type): Remove temporary shim and take the number
5944 of units as a poly_uint64 rather than an int.
5945 (build_opaque_vector_type): Take the number of units as a
5946 poly_uint64 rather than an int.
5947 * tree.c (build_vector_from_ctor): Handle polynomial
5948 TYPE_VECTOR_SUBPARTS.
5949 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5950 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5951 (build_vector_from_val): If the number of units is variable,
5952 use build_vec_duplicate_cst for constant operands and
5953 VEC_DUPLICATE_EXPR otherwise.
5954 (make_vector_type): Remove temporary is_constant ().
5955 (build_vector_type, build_opaque_vector_type): Take the number of
5956 units as a poly_uint64 rather than an int.
5957 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5959 * cfgexpand.c (expand_debug_expr): Likewise.
5960 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5961 (store_constructor, expand_expr_real_1): Likewise.
5962 (const_scalar_mask_from_tree): Likewise.
5963 * fold-const-call.c (fold_const_reduction): Likewise.
5964 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5965 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5966 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5967 (fold_relational_const): Likewise.
5968 (native_interpret_vector): Likewise. Change the size from an
5969 int to an unsigned int.
5970 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5971 TYPE_VECTOR_SUBPARTS.
5972 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5973 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5974 duplicating a non-constant operand into a variable-length vector.
5975 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5976 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5977 * ipa-icf.c (sem_variable::equals): Likewise.
5978 * match.pd: Likewise.
5979 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5980 * print-tree.c (print_node): Likewise.
5981 * stor-layout.c (layout_type): Likewise.
5982 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5983 * tree-cfg.c (verify_gimple_comparison): Likewise.
5984 (verify_gimple_assign_binary): Likewise.
5985 (verify_gimple_assign_ternary): Likewise.
5986 (verify_gimple_assign_single): Likewise.
5987 * tree-pretty-print.c (dump_generic_node): Likewise.
5988 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5989 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5990 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5991 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5992 (vect_shift_permute_load_chain): Likewise.
5993 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5994 (expand_vector_condition, optimize_vector_constructor): Likewise.
5995 (lower_vec_perm, get_compute_type): Likewise.
5996 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5997 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5998 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5999 (vect_recog_mask_conversion_pattern): Likewise.
6000 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6001 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6002 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6003 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6004 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6005 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6006 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6007 (supportable_widening_operation): Likewise.
6008 (supportable_narrowing_operation): Likewise.
6009 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6011 * varasm.c (output_constant): Likewise.
6013 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6014 Alan Hayward <alan.hayward@arm.com>
6015 David Sherwood <david.sherwood@arm.com>
6017 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6018 so that both the length == 3 and length != 3 cases set up their
6019 own permute vectors. Add comments explaining why we know the
6020 number of elements is constant.
6021 (vect_permute_load_chain): 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 * machmode.h (mode_nunits): Change from unsigned char to
6029 (ONLY_FIXED_SIZE_MODES): New macro.
6030 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6031 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6032 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6034 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6035 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6036 or if measurement_type is not polynomial.
6037 * genmodes.c (ZERO_COEFFS): New macro.
6038 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6040 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6041 Use ZERO_COEFFS when emitting initializers.
6042 * data-streamer.h (bp_pack_poly_value): New function.
6043 (bp_unpack_poly_value): Likewise.
6044 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6045 for GET_MODE_NUNITS.
6046 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6047 for GET_MODE_NUNITS.
6048 * tree.c (make_vector_type): Remove temporary shim and make
6049 the real function take the number of units as a poly_uint64
6051 (build_vector_type_for_mode): Handle polynomial nunits.
6052 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6053 * emit-rtl.c (const_vec_series_p_1): Likewise.
6054 (gen_rtx_CONST_VECTOR): Likewise.
6055 * fold-const.c (test_vec_duplicate_folding): Likewise.
6056 * genrecog.c (validate_pattern): Likewise.
6057 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6058 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6059 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6060 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6061 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6062 * rtlanal.c (subreg_get_info): Likewise.
6063 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6064 (vect_grouped_load_supported): Likewise.
6065 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6066 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6067 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6068 (simplify_const_unary_operation, simplify_binary_operation_1)
6069 (simplify_const_binary_operation, simplify_ternary_operation)
6070 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6071 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6072 instead of CONST_VECTOR_NUNITS.
6073 * varasm.c (output_constant_pool_2): Likewise.
6074 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6075 explicit-encoded elements in the XVEC for variable-length vectors.
6077 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6079 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6081 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6082 Alan Hayward <alan.hayward@arm.com>
6083 David Sherwood <david.sherwood@arm.com>
6085 * coretypes.h (fixed_size_mode): Declare.
6086 (fixed_size_mode_pod): New typedef.
6087 * builtins.h (target_builtins::x_apply_args_mode)
6088 (target_builtins::x_apply_result_mode): Change type to
6089 fixed_size_mode_pod.
6090 * builtins.c (apply_args_size, apply_result_size, result_vector)
6091 (expand_builtin_apply_args_1, expand_builtin_apply)
6092 (expand_builtin_return): Update accordingly.
6094 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6096 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6097 * cselib.c (cselib_hash_rtx): Likewise.
6098 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6099 CONST_VECTOR encoding.
6101 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6102 Jeff Law <law@redhat.com>
6105 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6106 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6107 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6108 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6111 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6112 explicitly probe *sp in a noreturn function if there were any callee
6113 register saves or frame pointer is needed.
6115 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6118 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6119 BLKmode for ternary, binary or unary expressions.
6122 * var-tracking.c (delete_vta_debug_insn): New inline function.
6123 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6124 insns from get_insns () to NULL instead of each bb separately.
6125 Use delete_vta_debug_insn. No longer static.
6126 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6127 delete_vta_debug_insns callers.
6128 * rtl.h (delete_vta_debug_insns): Declare.
6129 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6130 instead of variable_tracking_main.
6132 2018-01-03 Martin Sebor <msebor@redhat.com>
6134 PR tree-optimization/83603
6135 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6136 arguments past the endof the argument list in functions declared
6137 without a prototype.
6138 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6139 Avoid checking when arguments are null.
6141 2018-01-03 Martin Sebor <msebor@redhat.com>
6144 * doc/extend.texi (attribute const): Fix a typo.
6145 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6146 issuing -Wsuggest-attribute for void functions.
6148 2018-01-03 Martin Sebor <msebor@redhat.com>
6150 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6151 offset_int::from instead of wide_int::to_shwi.
6152 (maybe_diag_overlap): Remove assertion.
6153 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6154 * gimple-ssa-sprintf.c (format_directive): Same.
6155 (parse_directive): Same.
6156 (sprintf_dom_walker::compute_format_length): Same.
6157 (try_substitute_return_value): Same.
6159 2018-01-03 Jeff Law <law@redhat.com>
6162 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6163 non-constant residual for zero at runtime and avoid probing in
6164 that case. Reorganize code for trailing problem to mirror handling
6167 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6169 PR tree-optimization/83501
6170 * tree-ssa-strlen.c (get_string_cst): New.
6171 (handle_char_store): Call get_string_cst.
6173 2018-01-03 Martin Liska <mliska@suse.cz>
6175 PR tree-optimization/83593
6176 * tree-ssa-strlen.c: Include tree-cfg.h.
6177 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6178 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6179 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6181 (strlen_dom_walker::before_dom_children): Call
6182 gimple_purge_dead_eh_edges. Dump tranformation with details
6184 (strlen_dom_walker::before_dom_children): Update call by adding
6185 new argument cleanup_eh.
6186 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6188 2018-01-03 Martin Liska <mliska@suse.cz>
6191 * cif-code.def (VARIADIC_THUNK): New enum value.
6192 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6195 2018-01-03 Jan Beulich <jbeulich@suse.com>
6197 * sse.md (mov<mode>_internal): Tighten condition for when to use
6198 vmovdqu<ssescalarsize> for TI and OI modes.
6200 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6202 Update copyright years.
6204 2018-01-03 Martin Liska <mliska@suse.cz>
6207 * ipa-visibility.c (function_and_variable_visibility): Skip
6208 functions with noipa attribure.
6210 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6212 * gcc.c (process_command): Update copyright notice dates.
6213 * gcov-dump.c (print_version): Ditto.
6214 * gcov.c (print_version): Ditto.
6215 * gcov-tool.c (print_version): Ditto.
6216 * gengtype.c (create_file): Ditto.
6217 * doc/cpp.texi: Bump @copying's copyright year.
6218 * doc/cppinternals.texi: Ditto.
6219 * doc/gcc.texi: Ditto.
6220 * doc/gccint.texi: Ditto.
6221 * doc/gcov.texi: Ditto.
6222 * doc/install.texi: Ditto.
6223 * doc/invoke.texi: Ditto.
6225 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6227 * vector-builder.h (vector_builder::m_full_nelts): Change from
6228 unsigned int to poly_uint64.
6229 (vector_builder::full_nelts): Update prototype accordingly.
6230 (vector_builder::new_vector): Likewise.
6231 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6232 (vector_builder::operator ==): Likewise.
6233 (vector_builder::finalize): Likewise.
6234 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6235 Take the number of elements as a poly_uint64 rather than an
6237 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6238 from unsigned int to poly_uint64.
6239 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6240 (vec_perm_indices::new_vector): Likewise.
6241 (vec_perm_indices::length): Likewise.
6242 (vec_perm_indices::nelts_per_input): Likewise.
6243 (vec_perm_indices::input_nelts): Likewise.
6244 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6245 number of elements per input as a poly_uint64 rather than an
6246 unsigned int. Use the original encoding for variable-length
6247 vectors, rather than clamping each individual element.
6248 For the second and subsequent elements in each pattern,
6249 clamp the step and base before clamping their sum.
6250 (vec_perm_indices::series_p): Handle polynomial element counts.
6251 (vec_perm_indices::all_in_range_p): Likewise.
6252 (vec_perm_indices_to_tree): Likewise.
6253 (vec_perm_indices_to_rtx): Likewise.
6254 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6255 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6256 (tree_vector_builder::new_binary_operation): Handle polynomial
6257 element counts. Return false if we need to know the number
6258 of elements at compile time.
6259 * fold-const.c (fold_vec_perm): Punt if the number of elements
6260 isn't known at compile time.
6262 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6264 * vec-perm-indices.h (vec_perm_builder): Change element type
6265 from HOST_WIDE_INT to poly_int64.
6266 (vec_perm_indices::element_type): Update accordingly.
6267 (vec_perm_indices::clamp): Handle polynomial element_types.
6268 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6269 (vec_perm_indices::all_in_range_p): Likewise.
6270 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6272 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6273 polynomial vec_perm_indices element types.
6274 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6275 * fold-const.c (fold_vec_perm): Likewise.
6276 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6277 * tree-vect-generic.c (lower_vec_perm): Likewise.
6278 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6279 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6280 element type to HOST_WIDE_INT.
6282 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6283 Alan Hayward <alan.hayward@arm.com>
6284 David Sherwood <david.sherwood@arm.com>
6286 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6287 rather than an int. Use plus_constant.
6288 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6289 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6291 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6292 Alan Hayward <alan.hayward@arm.com>
6293 David Sherwood <david.sherwood@arm.com>
6295 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6296 a HOST_WIDE_INT to a poly_int64.
6298 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6299 Alan Hayward <alan.hayward@arm.com>
6300 David Sherwood <david.sherwood@arm.com>
6302 * calls.c (load_register_parameters): Cope with polynomial
6303 mode sizes. Require a constant size for BLKmode parameters
6304 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6305 forces a parameter to be padded at the lsb end in order to
6306 fill a complete number of words, require the parameter size
6307 to be ordered wrt UNITS_PER_WORD.
6309 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6310 Alan Hayward <alan.hayward@arm.com>
6311 David Sherwood <david.sherwood@arm.com>
6313 * reload1.c (spill_stack_slot_width): Change element type
6314 from unsigned int to poly_uint64_pod.
6315 (alter_reg): Treat mode sizes as polynomial.
6317 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6318 Alan Hayward <alan.hayward@arm.com>
6319 David Sherwood <david.sherwood@arm.com>
6321 * reload.c (complex_word_subreg_p): New function.
6322 (reload_inner_reg_of_subreg, push_reload): Use it.
6324 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6325 Alan Hayward <alan.hayward@arm.com>
6326 David Sherwood <david.sherwood@arm.com>
6328 * lra-constraints.c (process_alt_operands): Reject matched
6329 operands whose sizes aren't ordered.
6330 (match_reload): Refer to this check here.
6332 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6333 Alan Hayward <alan.hayward@arm.com>
6334 David Sherwood <david.sherwood@arm.com>
6336 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6337 that the mode size is in the set {1, 2, 4, 8, 16}.
6339 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6340 Alan Hayward <alan.hayward@arm.com>
6341 David Sherwood <david.sherwood@arm.com>
6343 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6344 Use plus_constant instead of gen_rtx_PLUS.
6346 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6347 Alan Hayward <alan.hayward@arm.com>
6348 David Sherwood <david.sherwood@arm.com>
6350 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6351 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6352 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6353 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6354 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6355 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6356 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6357 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6358 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6359 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6361 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6362 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6363 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6364 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6365 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6366 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6367 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6368 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6369 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6370 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6372 * expr.c (emit_move_resolve_push): Treat the input and result
6373 of PUSH_ROUNDING as a poly_int64.
6374 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6375 (emit_push_insn): Likewise.
6376 * lra-eliminations.c (mark_not_eliminable): Likewise.
6377 * recog.c (push_operand): Likewise.
6378 * reload1.c (elimination_effects): Likewise.
6379 * rtlanal.c (nonzero_bits1): Likewise.
6380 * calls.c (store_one_arg): Likewise. Require the padding to be
6381 known at compile time.
6383 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6384 Alan Hayward <alan.hayward@arm.com>
6385 David Sherwood <david.sherwood@arm.com>
6387 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6388 Use plus_constant instead of gen_rtx_PLUS.
6390 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6391 Alan Hayward <alan.hayward@arm.com>
6392 David Sherwood <david.sherwood@arm.com>
6394 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6397 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6398 Alan Hayward <alan.hayward@arm.com>
6399 David Sherwood <david.sherwood@arm.com>
6401 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6402 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6403 via stack temporaries. Treat the mode size as polynomial too.
6405 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6406 Alan Hayward <alan.hayward@arm.com>
6407 David Sherwood <david.sherwood@arm.com>
6409 * expr.c (expand_expr_real_2): When handling conversions involving
6410 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6411 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6412 as a poly_uint64 too.
6414 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6415 Alan Hayward <alan.hayward@arm.com>
6416 David Sherwood <david.sherwood@arm.com>
6418 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6420 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6421 Alan Hayward <alan.hayward@arm.com>
6422 David Sherwood <david.sherwood@arm.com>
6424 * combine.c (can_change_dest_mode): Handle polynomial
6425 REGMODE_NATURAL_SIZE.
6426 * expmed.c (store_bit_field_1): Likewise.
6427 * expr.c (store_constructor): Likewise.
6428 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6429 and polynomial REGMODE_NATURAL_SIZE.
6430 (gen_lowpart_common): Likewise.
6431 * reginfo.c (record_subregs_of_mode): Likewise.
6432 * rtlanal.c (read_modify_subreg_p): Likewise.
6434 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6435 Alan Hayward <alan.hayward@arm.com>
6436 David Sherwood <david.sherwood@arm.com>
6438 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6439 numbers of elements.
6441 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6442 Alan Hayward <alan.hayward@arm.com>
6443 David Sherwood <david.sherwood@arm.com>
6445 * match.pd: Cope with polynomial numbers of vector elements.
6447 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6448 Alan Hayward <alan.hayward@arm.com>
6449 David Sherwood <david.sherwood@arm.com>
6451 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6452 in a POINTER_PLUS_EXPR.
6454 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6455 Alan Hayward <alan.hayward@arm.com>
6456 David Sherwood <david.sherwood@arm.com>
6458 * omp-simd-clone.c (simd_clone_subparts): New function.
6459 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6460 (ipa_simd_modify_function_body): Likewise.
6462 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6463 Alan Hayward <alan.hayward@arm.com>
6464 David Sherwood <david.sherwood@arm.com>
6466 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6467 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6468 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6469 (expand_vector_condition, vector_element): Likewise.
6470 (subparts_gt): New function.
6471 (get_compute_type): Use subparts_gt.
6472 (count_type_subparts): Delete.
6473 (expand_vector_operations_1): Use subparts_gt instead of
6474 count_type_subparts.
6476 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6477 Alan Hayward <alan.hayward@arm.com>
6478 David Sherwood <david.sherwood@arm.com>
6480 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6481 (vect_compile_time_alias): ...this new function. Do the calculation
6482 on poly_ints rather than trees.
6483 (vect_prune_runtime_alias_test_list): Update call accordingly.
6485 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6486 Alan Hayward <alan.hayward@arm.com>
6487 David Sherwood <david.sherwood@arm.com>
6489 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6491 (vect_schedule_slp_instance): Likewise.
6493 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6494 Alan Hayward <alan.hayward@arm.com>
6495 David Sherwood <david.sherwood@arm.com>
6497 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6498 constant and extern definitions for variable-length vectors.
6499 (vect_get_constant_vectors): Note that the number of units
6500 is known to be constant.
6502 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6503 Alan Hayward <alan.hayward@arm.com>
6504 David Sherwood <david.sherwood@arm.com>
6506 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6507 of units as polynomial. Choose between WIDE and NARROW based
6510 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6511 Alan Hayward <alan.hayward@arm.com>
6512 David Sherwood <david.sherwood@arm.com>
6514 * tree-vect-stmts.c (simd_clone_subparts): New function.
6515 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6517 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6518 Alan Hayward <alan.hayward@arm.com>
6519 David Sherwood <david.sherwood@arm.com>
6521 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6522 vectors as polynomial. Use build_index_vector for
6525 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6526 Alan Hayward <alan.hayward@arm.com>
6527 David Sherwood <david.sherwood@arm.com>
6529 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6530 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6531 for variable-length vectors.
6532 (vectorizable_mask_load_store): Treat the number of units as
6533 polynomial, asserting that it is constant if the condition has
6534 already been enforced.
6535 (vectorizable_store, vectorizable_load): Likewise.
6537 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6538 Alan Hayward <alan.hayward@arm.com>
6539 David Sherwood <david.sherwood@arm.com>
6541 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6542 of units as polynomial. Punt if we can't tell at compile time
6543 which vector contains the final result.
6545 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6546 Alan Hayward <alan.hayward@arm.com>
6547 David Sherwood <david.sherwood@arm.com>
6549 * tree-vect-loop.c (vectorizable_induction): Treat the number
6550 of units as polynomial. Punt on SLP inductions. Use an integer
6551 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6552 cast of such a series for variable-length floating-point
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.h (build_index_vector): Declare.
6560 * tree.c (build_index_vector): New function.
6561 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6562 of units as polynomial, forcibly converting it to a constant if
6563 vectorizable_reduction has already enforced the condition.
6564 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6565 to create a {1,2,3,...} vector.
6566 (vectorizable_reduction): Treat the number of units as polynomial.
6567 Choose vectype_in based on the largest scalar element size rather
6568 than the smallest number of units. Enforce the restrictions
6571 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6572 Alan Hayward <alan.hayward@arm.com>
6573 David Sherwood <david.sherwood@arm.com>
6575 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6576 number of units as polynomial.
6578 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6579 Alan Hayward <alan.hayward@arm.com>
6580 David Sherwood <david.sherwood@arm.com>
6582 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6583 * target.def (autovectorize_vector_sizes): Return the vector sizes
6584 by pointer, using vector_sizes rather than a bitmask.
6585 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6586 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6587 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6589 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6590 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6591 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6592 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6593 * omp-general.c (omp_max_vf): Likewise.
6594 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6595 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6596 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6597 * tree-vect-slp.c (vect_slp_bb): Likewise.
6598 * doc/tm.texi: Regenerate.
6599 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6601 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6602 the vector size as a poly_uint64 rather than an unsigned int.
6603 (current_vector_size): Change from an unsigned int to a poly_uint64.
6604 (get_vectype_for_scalar_type): Update accordingly.
6605 * tree.h (build_truth_vector_type): Take the size and number of
6606 units as a poly_uint64 rather than an unsigned int.
6607 (build_vector_type): Add a temporary overload that takes
6608 the number of units as a poly_uint64 rather than an unsigned int.
6609 * tree.c (make_vector_type): Likewise.
6610 (build_truth_vector_type): Take the number of units as a poly_uint64
6611 rather than an unsigned int.
6613 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6614 Alan Hayward <alan.hayward@arm.com>
6615 David Sherwood <david.sherwood@arm.com>
6617 * target.def (get_mask_mode): Take the number of units and length
6618 as poly_uint64s rather than unsigned ints.
6619 * targhooks.h (default_get_mask_mode): Update accordingly.
6620 * targhooks.c (default_get_mask_mode): Likewise.
6621 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6622 * doc/tm.texi: Regenerate.
6624 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6625 Alan Hayward <alan.hayward@arm.com>
6626 David Sherwood <david.sherwood@arm.com>
6628 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6629 * omp-general.c (omp_max_vf): Likewise.
6630 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6631 (expand_omp_simd): Handle polynomial safelen.
6632 * omp-low.c (omplow_simd_context): Add a default constructor.
6633 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6634 (lower_rec_simd_input_clauses): Update accordingly.
6635 (lower_rec_input_clauses): Likewise.
6637 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6638 Alan Hayward <alan.hayward@arm.com>
6639 David Sherwood <david.sherwood@arm.com>
6641 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6642 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6643 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6644 (vect_analyze_slp_cost): Likewise.
6645 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6646 (vect_model_load_cost): Likewise.
6648 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6649 Alan Hayward <alan.hayward@arm.com>
6650 David Sherwood <david.sherwood@arm.com>
6652 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6653 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6654 from an unsigned int * to a poly_uint64_pod *.
6655 (calculate_unrolling_factor): New function.
6656 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
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-vectorizer.h (_slp_instance::unrolling_factor): Change
6663 from an unsigned int to a poly_uint64.
6664 (_loop_vec_info::slp_unrolling_factor): Likewise.
6665 (_loop_vec_info::vectorization_factor): Change from an int
6667 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6668 (vect_get_num_vectors): New function.
6669 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6670 (vect_get_num_copies): Use vect_get_num_vectors.
6671 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6672 to an unsigned int *.
6673 (vect_analyze_data_refs): Change min_vf from an int * to a
6675 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6676 than an unsigned HOST_WIDE_INT.
6677 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6678 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6679 to an unsigned int *.
6680 (vect_analyze_data_ref_dependences): Likewise.
6681 (vect_compute_data_ref_alignment): Handle polynomial vf.
6682 (vect_enhance_data_refs_alignment): Likewise.
6683 (vect_prune_runtime_alias_test_list): Likewise.
6684 (vect_shift_permute_load_chain): Likewise.
6685 (vect_supportable_dr_alignment): Likewise.
6686 (dependence_distance_ge_vf): Take the vectorization factor as a
6687 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6688 (vect_analyze_data_refs): Change min_vf from an int * to a
6690 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6691 vfm1 as a poly_uint64 rather than an int. Make the same change
6692 for the returned bound_scalar.
6693 (vect_gen_vector_loop_niters): Handle polynomial vf.
6694 (vect_do_peeling): Likewise. Update call to
6695 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6696 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6698 * tree-vect-loop.c (vect_determine_vectorization_factor)
6699 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6700 (vect_get_known_peeling_cost): Likewise.
6701 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6702 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6703 (vect_transform_loop): Likewise. Use the lowest possible VF when
6704 updating the upper bounds of the loop.
6705 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6707 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6708 polynomial unroll factors.
6709 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6710 (vect_make_slp_decision): Likewise.
6711 (vect_supported_load_permutation_p): Likewise, and polynomial
6713 (vect_analyze_slp_cost): Handle polynomial vf.
6714 (vect_slp_analyze_node_operations): Likewise.
6715 (vect_slp_analyze_bb_1): Likewise.
6716 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6717 than an unsigned HOST_WIDE_INT.
6718 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6719 (vectorizable_load): Handle polynomial vf.
6720 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6722 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6724 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6725 Alan Hayward <alan.hayward@arm.com>
6726 David Sherwood <david.sherwood@arm.com>
6728 * match.pd: Handle bit operations involving three constants
6729 and try to fold one pair.
6731 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6733 * tree-vect-loop-manip.c: Include gimple-fold.h.
6734 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6735 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6736 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6737 Add a path that uses a step of VF instead of 1, but disable it
6739 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6740 and niters_no_overflow parameters. Update calls to
6741 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6742 Create a new SSA name if the latter choses to use a ste other
6743 than zero, and return it via niters_vector_mult_vf_var.
6744 * tree-vect-loop.c (vect_transform_loop): Update calls to
6745 vect_do_peeling, vect_gen_vector_loop_niters and
6746 slpeel_make_loop_iterate_ntimes.
6747 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6748 (vect_gen_vector_loop_niters): Update declarations after above changes.
6750 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6752 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6753 128-bit round to integer instructions.
6754 (ceil<mode>2): Likewise.
6755 (btrunc<mode>2): Likewise.
6756 (round<mode>2): Likewise.
6758 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6760 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6761 unaligned VSX load/store on P8/P9.
6762 (expand_block_clear): Allow the use of unaligned VSX
6763 load/store on P8/P9.
6765 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6767 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6769 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6770 swap associated with both a load and a store.
6772 2018-01-02 Andrew Waterman <andrew@sifive.com>
6774 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6775 * config/riscv/riscv.md (clear_cache): Use it.
6777 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6779 * web.c: Remove out-of-date comment.
6781 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6783 * expr.c (fixup_args_size_notes): Check that any existing
6784 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6785 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6786 (emit_single_push_insn): ...here.
6788 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6790 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6791 (const_vector_encoded_nelts): New function.
6792 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6793 (const_vector_int_elt, const_vector_elt): Declare.
6794 * emit-rtl.c (const_vector_int_elt_1): New function.
6795 (const_vector_elt): Likewise.
6796 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6797 of CONST_VECTOR_ELT.
6799 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6801 * expr.c: Include rtx-vector-builder.h.
6802 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6803 directly on the tree encoding.
6804 (const_vector_from_tree): Likewise.
6805 * optabs.c: Include rtx-vector-builder.h.
6806 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6807 sequence of "u" values.
6808 * vec-perm-indices.c: Include rtx-vector-builder.h.
6809 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6810 directly on the vec_perm_indices encoding.
6812 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6814 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6815 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6816 * rtx-vector-builder.h: New file.
6817 * rtx-vector-builder.c: Likewise.
6818 * rtl.h (rtx_def::u2): Add a const_vector field.
6819 (CONST_VECTOR_NPATTERNS): New macro.
6820 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6821 (CONST_VECTOR_DUPLICATE_P): Likewise.
6822 (CONST_VECTOR_STEPPED_P): Likewise.
6823 (CONST_VECTOR_ENCODED_ELT): Likewise.
6824 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6825 (unwrap_const_vec_duplicate): Likewise.
6826 (const_vec_series_p): Check for a non-duplicated vector encoding.
6827 Say that the function only returns true for integer vectors.
6828 * emit-rtl.c: Include rtx-vector-builder.h.
6829 (gen_const_vec_duplicate_1): Delete.
6830 (gen_const_vector): Call gen_const_vec_duplicate instead of
6831 gen_const_vec_duplicate_1.
6832 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6833 (gen_const_vec_duplicate): Use rtx_vector_builder.
6834 (gen_const_vec_series): Likewise.
6835 (gen_rtx_CONST_VECTOR): Likewise.
6836 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6837 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6838 Build a new vector rather than modifying a CONST_VECTOR in-place.
6839 (handle_special_swappables): Update call accordingly.
6840 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6841 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6842 Build a new vector rather than modifying a CONST_VECTOR in-place.
6843 (handle_special_swappables): Update call accordingly.
6845 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6847 * simplify-rtx.c (simplify_const_binary_operation): Use
6848 CONST_VECTOR_ELT instead of XVECEXP.
6850 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6852 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6853 the selector elements to be different from the data elements
6854 if the selector is a VECTOR_CST.
6855 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6856 ssizetype for the selector.
6858 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6860 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6861 before testing each element individually.
6862 * tree-vect-generic.c (lower_vec_perm): Likewise.
6864 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6866 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6867 * selftest-run-tests.c (selftest::run_tests): Call it.
6868 * vector-builder.h (vector_builder::operator ==): New function.
6869 (vector_builder::operator !=): Likewise.
6870 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6871 (vec_perm_indices::all_from_input_p): New function.
6872 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6873 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6874 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6875 instead of reading the VECTOR_CST directly. Detect whether both
6876 vector inputs are the same before constructing the vec_perm_indices,
6877 and update the number of inputs argument accordingly. Use the
6878 utility functions added above. Only construct sel2 if we need to.
6880 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6882 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6883 the broadcast of the low byte.
6884 (expand_mult_highpart): Use an explicit encoding for the permutes.
6885 * optabs-query.c (can_mult_highpart_p): Likewise.
6886 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6887 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6888 (vectorizable_bswap): Likewise.
6889 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6890 explicit encoding for the power-of-2 permutes.
6891 (vect_permute_store_chain): Likewise.
6892 (vect_grouped_load_supported): Likewise.
6893 (vect_permute_load_chain): Likewise.
6895 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6897 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6898 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6899 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6900 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6901 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6902 (vect_gen_perm_mask_any): Likewise.
6904 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6906 * int-vector-builder.h: New file.
6907 * vec-perm-indices.h: Include int-vector-builder.h.
6908 (vec_perm_indices): Redefine as an int_vector_builder.
6909 (auto_vec_perm_indices): Delete.
6910 (vec_perm_builder): Redefine as a stand-alone class.
6911 (vec_perm_indices::vec_perm_indices): New function.
6912 (vec_perm_indices::clamp): Likewise.
6913 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6914 (vec_perm_indices::new_vector): New function.
6915 (vec_perm_indices::new_expanded_vector): Update for new
6916 vec_perm_indices class.
6917 (vec_perm_indices::rotate_inputs): New function.
6918 (vec_perm_indices::all_in_range_p): Operate directly on the
6919 encoded form, without computing elided elements.
6920 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6921 encoding. Update for new vec_perm_indices class.
6922 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6923 the given vec_perm_builder.
6924 (expand_vec_perm_var): Update vec_perm_builder constructor.
6925 (expand_mult_highpart): Use vec_perm_builder instead of
6926 auto_vec_perm_indices.
6927 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6928 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6929 or double series encoding as appropriate.
6930 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6931 vec_perm_indices instead of auto_vec_perm_indices.
6932 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6933 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6934 (vect_permute_store_chain): Likewise.
6935 (vect_grouped_load_supported): Likewise.
6936 (vect_permute_load_chain): Likewise.
6937 (vect_shift_permute_load_chain): Likewise.
6938 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6939 (vect_transform_slp_perm_load): Likewise.
6940 (vect_schedule_slp_instance): Likewise.
6941 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6942 (vectorizable_mask_load_store): Likewise.
6943 (vectorizable_bswap): Likewise.
6944 (vectorizable_store): Likewise.
6945 (vectorizable_load): Likewise.
6946 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6947 vec_perm_indices instead of auto_vec_perm_indices. Use
6948 tree_to_vec_perm_builder to read the vector from a tree.
6949 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6950 vec_perm_builder instead of a vec_perm_indices.
6951 (have_whole_vector_shift): Use vec_perm_builder and
6952 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6953 truncation to calc_vec_perm_mask_for_shift.
6954 (vect_create_epilog_for_reduction): Likewise.
6955 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6956 from auto_vec_perm_indices to vec_perm_indices.
6957 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6958 instead of changing individual elements.
6959 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6960 the vector in d.perm.
6961 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6962 from auto_vec_perm_indices to vec_perm_indices.
6963 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6964 instead of changing individual elements.
6965 (arm_vectorize_vec_perm_const): Use new_vector to install
6966 the vector in d.perm.
6967 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6968 Update vec_perm_builder constructor.
6969 (rs6000_expand_interleave): Likewise.
6970 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6971 (rs6000_expand_interleave): Likewise.
6973 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6975 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6976 to qimode could truncate the indices.
6977 * optabs.c (expand_vec_perm_var): Likewise.
6979 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6981 * Makefile.in (OBJS): Add vec-perm-indices.o.
6982 * vec-perm-indices.h: New file.
6983 * vec-perm-indices.c: Likewise.
6984 * target.h (vec_perm_indices): Replace with a forward class
6986 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6987 * optabs.h: Include vec-perm-indices.h.
6988 (expand_vec_perm): Delete.
6989 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6990 (expand_vec_perm_const): Declare.
6991 * target.def (vec_perm_const_ok): Replace with...
6992 (vec_perm_const): ...this new hook.
6993 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6994 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6995 * doc/tm.texi: Regenerate.
6996 * optabs.def (vec_perm_const): Delete.
6997 * doc/md.texi (vec_perm_const): Likewise.
6998 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6999 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7000 expand_vec_perm for constant permutation vectors. Assert that
7001 the mode of variable permutation vectors is the integer equivalent
7002 of the mode that is being permuted.
7003 * optabs-query.h (selector_fits_mode_p): Declare.
7004 * optabs-query.c: Include vec-perm-indices.h.
7005 (selector_fits_mode_p): New function.
7006 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7007 is defined, instead of checking whether the vec_perm_const_optab
7008 exists. Use targetm.vectorize.vec_perm_const instead of
7009 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7010 fit in the vector mode before using a variable permute.
7011 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7012 vec_perm_indices instead of an rtx.
7013 (expand_vec_perm): Replace with...
7014 (expand_vec_perm_const): ...this new function. Take the selector
7015 as a vec_perm_indices rather than an rtx. Also take the mode of
7016 the selector. Update call to shift_amt_for_vec_perm_mask.
7017 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7018 Use vec_perm_indices::new_expanded_vector to expand the original
7019 selector into bytes. Check whether the indices fit in the vector
7020 mode before using a variable permute.
7021 (expand_vec_perm_var): Make global.
7022 (expand_mult_highpart): Use expand_vec_perm_const.
7023 * fold-const.c: Includes vec-perm-indices.h.
7024 * tree-ssa-forwprop.c: Likewise.
7025 * tree-vect-data-refs.c: Likewise.
7026 * tree-vect-generic.c: Likewise.
7027 * tree-vect-loop.c: Likewise.
7028 * tree-vect-slp.c: Likewise.
7029 * tree-vect-stmts.c: Likewise.
7030 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7032 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7033 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7034 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7035 (aarch64_vectorize_vec_perm_const): ...this new function.
7036 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7037 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7038 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7039 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7040 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7041 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7042 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7044 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7045 check for NEON modes.
7046 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7047 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7048 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7049 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7051 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7052 the old VEC_PERM_CONST conditions.
7053 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7054 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7055 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7056 (ia64_vectorize_vec_perm_const_ok): Merge into...
7057 (ia64_vectorize_vec_perm_const): ...this new function.
7058 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7059 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7060 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7061 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7062 * config/mips/mips.c (mips_expand_vec_perm_const)
7063 (mips_vectorize_vec_perm_const_ok): Merge into...
7064 (mips_vectorize_vec_perm_const): ...this new function.
7065 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7066 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7067 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7068 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7069 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7070 (rs6000_expand_vec_perm_const): Delete.
7071 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7073 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7074 (altivec_expand_vec_perm_const_le): Take each operand individually.
7075 Operate on constant selectors rather than rtxes.
7076 (altivec_expand_vec_perm_const): Likewise. Update call to
7077 altivec_expand_vec_perm_const_le.
7078 (rs6000_expand_vec_perm_const): Delete.
7079 (rs6000_vectorize_vec_perm_const_ok): Delete.
7080 (rs6000_vectorize_vec_perm_const): New function.
7081 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7082 an element count and rtx array.
7083 (rs6000_expand_extract_even): Update call accordingly.
7084 (rs6000_expand_interleave): Likewise.
7085 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7086 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7087 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7088 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7089 (rs6000_expand_vec_perm_const): Delete.
7090 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7091 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7092 (altivec_expand_vec_perm_const_le): Take each operand individually.
7093 Operate on constant selectors rather than rtxes.
7094 (altivec_expand_vec_perm_const): Likewise. Update call to
7095 altivec_expand_vec_perm_const_le.
7096 (rs6000_expand_vec_perm_const): Delete.
7097 (rs6000_vectorize_vec_perm_const_ok): Delete.
7098 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7099 reference to the SPE evmerge intructions.
7100 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7101 an element count and rtx array.
7102 (rs6000_expand_extract_even): Update call accordingly.
7103 (rs6000_expand_interleave): Likewise.
7104 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7105 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7107 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7109 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7111 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7112 vector mode and that that mode matches the mode of the data
7114 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7115 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7116 directly using expand_vec_perm_1 when forcing selectors into
7118 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7120 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7122 * optabs-query.h (can_vec_perm_p): Delete.
7123 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7124 * optabs-query.c (can_vec_perm_p): Split into...
7125 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7126 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7127 particular selector is valid.
7128 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7129 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7130 (vect_grouped_load_supported): Likewise.
7131 (vect_shift_permute_load_chain): Likewise.
7132 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7133 (vect_transform_slp_perm_load): Likewise.
7134 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7135 (vectorizable_bswap): Likewise.
7136 (vect_gen_perm_mask_checked): Likewise.
7137 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7138 implementations of variable permutation vectors into account
7139 when deciding which selector to use.
7140 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7141 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7142 with a false third argument.
7143 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7144 to test whether the constant selector is valid and can_vec_perm_var_p
7145 to test whether a variable selector is valid.
7147 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7149 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7150 * optabs-query.c (can_vec_perm_p): Likewise.
7151 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7152 instead of vec_perm_indices.
7153 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7154 (vect_gen_perm_mask_checked): Likewise,
7155 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7156 (vect_gen_perm_mask_checked): Likewise,
7158 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7160 * optabs-query.h (qimode_for_vec_perm): Declare.
7161 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7162 (qimode_for_vec_perm): ...this new function.
7163 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7165 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7167 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7168 does not have a conditional at the top.
7170 2018-01-02 Richard Biener <rguenther@suse.de>
7172 * ipa-inline.c (big_speedup_p): Fix expression.
7174 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7177 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7180 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7184 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7185 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7186 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7187 cond_taken_branch_cost 3->4.
7189 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7191 PR tree-optimization/83581
7192 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7193 TODO_cleanup_cfg if any changes have been made.
7196 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7197 convert_modes if target mode has the right side, but different mode
7201 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7202 last argument when extracting from CONCAT. If either from_real or
7203 from_imag is NULL, use expansion through memory. If result is not
7204 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7205 the parts directly to inner mode, if even that fails, use expansion
7209 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7210 check for bswap in mode rather than HImode and use that in expand_unop
7213 Copyright (C) 2018 Free Software Foundation, Inc.
7215 Copying and distribution of this file, with or without modification,
7216 are permitted in any medium without royalty provided the copyright
7217 notice and this notice are preserved.