1 2018-02-16 Jakub Jelinek <jakub@redhat.com>
4 * ipa-inline.c (inline_small_functions): Fix a typo.
6 2018-02-16 Nathan Sidwell <nathan@acm.org>
8 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
10 2018-02-16 Carl Love <cel@us.ibm.com>
12 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
13 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
14 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
15 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
16 expansion to P8V_BUILTIN_VEC_FLOAT2.
18 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
20 PR rtl-optimization/70023
21 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
22 src_regno into account.
24 2018-02-16 Carl Love <cel@us.ibm.com>
26 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
27 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
28 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
29 * config/rs6000/rs6000.c: Remove case statements for
30 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
31 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
32 and P9V_BUILTIN_VEC_VINSERT4B.
33 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
34 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
35 * config/rs6000/vsx.md:
36 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
39 2018-02-16 Carl Love <cel@us.ibm.com>
41 * config/rs6000/altivec.h: Add builtin names vec_extract4b
43 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
45 * config/rs6000/rs6000-c.c: Add the definitions for
46 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
47 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
48 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
49 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
50 definition for insert4b and define insn *insert3b_internal.
51 * doc/extend.texi: Add documentation for vec_extract4b.
53 2018-02-16 Nathan Sidwell <nathan@acm.org>
55 * doc/extend.texi (Backwards Compatibility): Mention friend
56 injection. Note for-scope is deprecated.
57 * doc/invoke.texi (-ffriend-injection): Deprecate.
59 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
61 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
62 that moved to I2, also allow destinations that are a paradoxical
63 subreg (instead of a normal reg).
65 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
68 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
71 2018-02-16 Richard Biener <rguenther@suse.de>
73 PR tree-optimization/84037
74 PR tree-optimization/84016
76 * config/i386/i386.c (ix86_builtin_vectorization_cost):
77 Adjust vec_construct for the fact we need additional higher latency
78 128bit inserts for AVX256 and AVX512 vector builds.
79 (ix86_add_stmt_cost): Scale vector construction cost for
82 2018-02-16 Richard Biener <rguenther@suse.de>
84 PR tree-optimization/84417
85 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
86 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
87 (non_rewritable_lvalue_p): Likewise, use poly-ints.
89 2018-02-16 Martin Liska <mliska@suse.cz>
92 * internal-fn.def (ASAN_CHECK): Set proper flags.
93 (ASAN_MARK): Likewise.
95 2018-02-16 Julia Koval <julia.koval@intel.com>
97 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
100 2018-02-16 Jakub Jelinek <jakub@redhat.com>
103 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
104 Use ++iter rather than iter++ for std::list iterators.
105 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
106 defer deleting them until all nodes in the forest are processed. Do
107 free even leaf nodes. Change to_process into auto_vec.
110 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
111 * vec.h (vec_default_construct): Use memset instead of placement new
112 if BROKEN_VALUE_INITIALIZATION is defined.
113 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
114 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
117 PR rtl-optimization/83723
118 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
119 * lra.c (lra_substitute_pseudo): Likewise. If true, use
120 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
122 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
124 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
126 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
128 PR rtl-optimization/81443
129 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
130 from inner REGs to paradoxical SUBREGs.
132 2018-02-16 Richard Biener <rguenther@suse.de>
134 PR tree-optimization/84399
135 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
136 For operands we can analyze at their definition make sure we can
137 analyze them at each use as well.
139 2018-02-16 Richard Biener <rguenther@suse.de>
141 PR tree-optimization/84190
142 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
143 volatile accesses if the decl isn't volatile.
145 2018-02-15 Jason Merrill <jason@redhat.com>
147 PR c++/84314 - ICE with templates and fastcall attribute.
148 * attribs.c (build_type_attribute_qual_variant): Don't clobber
149 TYPE_CANONICAL on an existing type.
151 2018-02-15 Jakub Jelinek <jakub@redhat.com>
153 PR tree-optimization/84383
154 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
155 dstoff nor call operand_equal_p if dstbase is NULL.
157 PR tree-optimization/84334
158 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
159 also a CONSTANT_CLASS_P, punt.
161 2018-02-14 Jim Wilson <jimw@sifive.com>
163 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
164 first SMALL_OPERAND check. New local min_second_step. Move assert
165 to where locals are set. Add TARGET_RVC support.
166 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
168 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
170 * doc/invoke.texi: Correct -Wformat-overflow code sample.
172 2018-02-14 Martin Sebor <msebor@redhat.com>
174 PR tree-optimization/83698
175 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
176 arrays constrain the offset range to their bounds.
177 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
178 (builtin_access::overlap): Avoid setting the size of overlap if it's
180 (maybe_diag_overlap): Also consider arrays when deciding what values
181 of offsets to include in diagnostics.
183 2018-02-14 Martin Sebor <msebor@redhat.com>
186 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
187 that correspond to the kind of a declaration.
189 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
192 * config/pa/pa.md: Load address of PIC label using the linkage table
193 if the label is nonlocal.
195 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
197 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
198 warning message if user requests -maltivec=be.
199 * doc/invoke.texi: Document deprecation of -maltivec=be.
201 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
204 * config/rs6000/rs6000-c.c: Update definitions for
205 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
206 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
208 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
211 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
212 add _get_ssp intrinsics. Remove argument from
213 __builtin_ia32_rdssp[d|q].
214 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
215 * config/i386/i386-builtin.def: Remove argument from
216 __builtin_ia32_rdssp[d|q].
217 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
218 ix86_expand_special_args_builtin for _rdssp[d|q].
219 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
220 Clear register before usage.
221 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
222 Add documentation for new _get_ssp and _inc_ssp intrinsics.
224 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
226 PR tree-optimization/84357
227 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
228 operand 1 of an ARRAY_REF too.
230 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
233 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
234 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
236 (set_of_reg): New struct.
237 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
238 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
239 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
241 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
242 Split into bitclr, bitset, bitinvert patterns if appropriate.
243 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
244 use rx_fuse_in_memory_bitop.
245 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
246 to named insn, correct maximum insn length.
248 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
251 * machmode.def: Define a complex mode for PARTIAL_INT.
252 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
254 * doc/rtl.texi: Document CSPImode.
255 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
257 (msp430_hard_regno_nregs_with_padding): Likewise.
259 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
262 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
264 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
266 PR rtl-optimization/84169
267 * combine.c (try_combine): New variable split_i2i3. Set it to true if
268 we generated a parallel as new i3 and we split that to new i2 and i3
269 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
270 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
271 those to i2, not i1. Partially rewrite this scan code.
273 2018-02-13 Jakub Jelinek <jakub@redhat.com>
276 * stor-layout.c (place_field): For variable length fields, adjust
277 offset_align afterwards not just based on the field's alignment,
278 but also on the size.
281 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
282 of exps and logs in the use_exp2 case.
284 2018-02-13 Jeff Law <law@redhat.com>
286 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
289 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
292 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
296 * common.opt (gas-loc-support, gas-locview-support): New.
297 (ginline-points, ginternal-reset-location-views): New.
298 * doc/invoke.texi: Document them. Use @itemx where intended.
299 (gvariable-location-views): Adjust.
300 * target.def (reset_location_view): New.
301 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
302 (TARGET_RESET_LOCATION_VIEW): New.
303 * doc/tm.texi: Rebuilt.
304 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
305 (dwarf2out_default_as_locview_support): New.
306 (output_asm_line_debug_info): Use option variables.
307 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
308 (output_loc_list): Likewise.
309 (add_high_low_attributes): Check option variables.
310 Don't output entry view attribute in strict mode.
311 (gen_inlined_subroutine_die): Check option variables.
312 (dwarf2out_inline_entry): Likewise.
313 (init_sections_and_labels): Likewise.
314 (dwarf2out_early_finish): Likewise.
315 (maybe_reset_location_view): New, from...
316 (dwarf2out_var_location): ... here. Call it.
317 * debug.h (dwarf2out_default_as_loc_support): Declare.
318 (dwarf2out_default_as_locview_support): Declare.
319 * hooks.c (hook_int_rtx_insn_0): New.
320 * hooks.h (hook_int_rtx_insn_0): Declare.
321 * toplev.c (process_options): Take -gas-loc-support and
322 -gas-locview-support from dwarf2out. Enable
323 -gvariable-location-views by default only with locview
324 assembler support. Enable -ginternal-reset-location-views by
325 default only if the target defines the corresponding hook.
326 Enable -ginline-points by default if location views are
327 enabled; force it disabled if statement frontiers are
329 * tree-inline.c (expand_call_inline): Check option variables.
330 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
332 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
334 PR tree-optimization/84321
335 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
336 handling. Also check whether the anti-range contains any values
337 that satisfy the mask; switch to a VR_RANGE if not.
339 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
342 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
344 2018-02-13 Martin Jambor <mjambor@suse.cz>
347 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
348 of call statements, also set location of a load to a temporary.
350 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
352 * config/rl78/rl78.c (add_vector_labels): New function.
353 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
354 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
355 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
356 which checks that no arguments are passed.
357 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
358 * doc/extend.texi: Documentation for the new attribute.
360 2018-02-13 Andreas Schwab <schwab@suse.de>
362 * config/riscv/linux.h (CPP_SPEC): Define.
364 2018-02-13 Jakub Jelinek <jakub@redhat.com>
367 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
368 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
369 OPTION_MASK_ISA_AES as first argument to def_builtin_const
370 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
371 instead of OPTION_MASK_ISA_PCLMUL as first argument to
372 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
373 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
374 temporarily for AES and PCLMUL builtins.
376 PR tree-optimization/84339
377 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
378 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
382 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
383 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
384 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
386 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
388 * omp-simd-clone.h: New file.
389 * omp-simd-clone.c: Include omp-simd-clone.h.
390 (expand_simd_clones): No longer static.
391 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
392 cgraph.h and omp-simd-clone.h.
393 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
394 (vect_recog_widen_shift_pattern): Formatting fix.
395 (vect_pattern_recog_1): Don't check optab for calls.
398 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
399 operands[2] into a REG before using gen_lowpart on it.
401 2018-02-12 Jeff Law <law@redhat.com>
404 * config/sh/sh.c (find_barrier): Consider a sibling call
407 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
408 successfully back substituting a reg.
410 2018-02-12 Richard Biener <rguenther@suse.de>
412 PR tree-optimization/84037
413 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
414 parameter, move visited init to caller.
415 (vect_slp_analyze_operations): Separate cost from validity
416 check, initialize visited once for all instances.
417 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
419 * tree-vect-stmts.c (vect_model_simple_cost): Make early
421 (vect_model_promotion_demotion_cost): Likewise.
422 (vectorizable_bswap): Guard cost modeling with !slp_node
423 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
425 (vectorizable_call): Likewise.
426 (vectorizable_conversion): Likewise.
427 (vectorizable_assignment): Likewise.
428 (vectorizable_shift): Likewise.
429 (vectorizable_operation): Likewise.
430 (vectorizable_store): Likewise.
431 (vectorizable_load): Likewise.
432 (vectorizable_condition): Likewise.
433 (vectorizable_comparison): Likewise.
435 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
438 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
439 (ASAN_MARK): Fix fnspec to account for return value, change pointer
440 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
442 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
445 * params.def (inline-min-speedup): Increase from 8 to 15.
446 (max-inline-insns-auto): Decrease from 40 to 30.
447 * ipa-split.c (consider_split): Add some buffer for function to
448 be considered inlining candidate.
449 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
452 2018-02-12 Richard Biener <rguenther@suse.de>
454 PR tree-optimization/84037
455 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
456 matched stmts if we cannot swap the non-matched ones.
458 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
460 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
461 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
462 _mm_maskz_scalef_round_ss): New intrinsics.
463 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
464 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
465 __builtin_ia32_scalefss_round): Remove.
466 (__builtin_ia32_scalefsd_mask_round,
467 __builtin_ia32_scalefss_mask_round): New intrinsics.
468 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
469 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
470 ((match_operand:VF_128 2 "<round_nimm_predicate>"
471 "<round_constraint>")): Changed to ...
472 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
473 "<round_scalar_constraint>")): ... this.
474 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
475 %0, %1, %2<round_op3>}"): Changed to ...
476 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
477 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
478 %2<round_scalar_mask_op3>}"): ... this.
479 * config/i386/subst.md (round_scalar_nimm_predicate): New.
481 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
483 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
484 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
485 (_mm_maskz_sqrt_round_ss): New intrinsics.
486 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
487 (__builtin_ia32_sqrtsd_mask_round)
488 (__builtin_ia32_sqrtss_mask_round): New builtins.
489 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
490 (__builtin_ia32_sqrtss_round): Remove.
491 (__builtin_ia32_sqrtsd_mask_round)
492 (__builtin_ia32_sqrtss_mask_round): New builtins.
493 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
494 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
495 ((match_operand:VF_128 1 "vector_operand"
496 "xBm,<round_constraint>")): Changed to ...
497 ((match_operand:VF_128 1 "vector_operand"
498 "xBm,<round_scalar_constraint>")): ... this.
499 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
500 %0, %2, %<iptr>1<round_op3>}): Changed to ...
501 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
502 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
503 %<iptr>1<round_scalar_mask_op3>}): ... this.
504 ((set_attr "prefix" "<round_prefix>")): Changed to ...
505 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
507 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
510 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
511 Cast vec_cmpeq result to correct type.
512 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
513 Cast vec_cmpgt result to correct type.
515 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
517 * final.c (final_scan_insn_1): Renamed from...
518 (final_scan_insn): ... this. New wrapper, to recover
519 seen from the outermost call in recursive ones.
520 * config/sparc/sparc.c (output_return): Drop seen from call.
521 (output_sibcall): Likewise.
522 * config/visium/visium.c (output_branch): Likewise.
524 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
526 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
529 2018-02-10 Alan Modra <amodra@gmail.com>
532 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
533 Specify LR as an input.
535 2018-02-10 Jakub Jelinek <jakub@redhat.com>
538 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
539 remove_member_access_dummy_vars): New functions.
540 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
541 lower_omp_1, execute_lower_omp): Use them.
543 PR rtl-optimization/84308
544 * shrink-wrap.c (spread_components): Release todo vector.
546 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
548 PR rtl-optimization/57193
549 * ira-color.c (struct allocno_color_data): Add member
550 conflict_allocno_hard_prefs.
551 (update_conflict_allocno_hard_prefs): New.
552 (bucket_allocno_compare_func): Add a preference based on
553 conflict_allocno_hard_prefs.
554 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
555 (color_allocnos): Remove a dead code. Initiate
556 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
558 2018-02-09 Jakub Jelinek <jakub@redhat.com>
561 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
562 constraint from =wa to wa. Avoid a subreg on the output operand,
563 instead use a pseudo and subreg it in a move.
564 (p9_xxbrd_<mode>): Changed to ...
565 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
566 (p9_xxbrd_v2df): New expander.
567 (p9_xxbrw_<mode>): Changed to ...
568 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
569 (p9_xxbrw_v4sf): New expander.
571 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
573 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
575 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
578 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
579 multiply in 32-bit mode.
580 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
581 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
584 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
586 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
587 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
588 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
589 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
591 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
594 * dwarf2out.c (is_trivial_indirect_ref): New function.
595 (dwarf2out_late_global_decl): Do not generate a location
596 attribute for variables that have a non-trivial DECL_VALUE_EXPR
597 and that are not defined in the current unit.
599 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
601 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
602 instead of a libcall for UNORDERED.
604 2018-02-09 Tamar Christina <tamar.christina@arm.com>
607 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
608 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
610 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
613 * config/s390/s390.c (s390_set_current_function): Invoke
614 s390_indirect_branch_settings also if fndecl didn't change.
616 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
618 * config/rs6000/rs6000.md (blockage): Set length to zero.
620 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
622 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
624 2018-02-09 Jakub Jelinek <jakub@redhat.com>
627 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
628 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
632 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
633 PARALLEL incoming that failed vt_get_decl_and_offset check.
636 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
637 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
639 (get_variable_section): For decls in named .bss* sections pass true as
640 second argument to bss_initializer_p.
642 2018-02-09 Marek Polacek <polacek@redhat.com>
643 Jakub Jelinek <jakub@redhat.com>
646 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
647 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
648 Sync some changes from cxx_fold_indirect_ref.
650 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
652 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
654 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
655 (BLOCK_INLINE_ENTRY_LABEL): New.
656 (dwarf2out_var_location): Disregard inline entry markers.
657 (inline_entry_data): New struct.
658 (inline_entry_data_hasher): New hashtable type.
659 (inline_entry_data_hasher::hash): New.
660 (inline_entry_data_hasher::equal): New.
661 (inline_entry_data_table): New variable.
662 (add_high_low_attributes): Add DW_AT_entry_pc and
663 DW_AT_GNU_entry_view attributes if a pending entry is found
664 in inline_entry_data_table. Add old entry_pc attribute only
665 if debug nonbinding markers are disabled.
666 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
668 (block_within_block_p, dwarf2out_inline_entry): New.
669 (dwarf2out_finish): Check that no entries remained in
670 inline_entry_data_table.
671 * final.c (reemit_insn_block_notes): Handle inline entry notes.
672 (final_scan_insn, notice_source_line): Likewise.
673 (rest_of_clean_state): Skip inline entry markers.
674 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
676 * gimple.c (gimple_build_debug_inline_entry): New.
677 * gimple.h (enum gimple_debug_subcode): Add
678 GIMPLE_DEBUG_INLINE_ENTRY.
679 (gimple_build_debug_inline_entry): Declare.
680 (gimple_debug_inline_entry_p): New.
681 (gimple_debug_nonbind_marker_p): Adjust.
682 * insn-notes.def (INLINE_ENTRY): New.
683 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
684 inline entry marker notes.
685 (print_insn): Likewise.
686 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
687 (INSN_DEBUG_MARKER_KIND): Likewise.
688 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
689 * tree-inline.c (expand_call_inline): Build and insert
690 debug_inline_entry stmt.
691 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
692 inline entry blocks early, if nonbind markers are enabled.
693 (dump_scope_block): Dump fragment info.
694 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
695 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
696 (gimple_build_debug_inline_entry): New.
697 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
698 Enable/disable inline entry points too.
699 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
700 (DEBUG_INSN): Describe inline entry markers.
702 * common.opt (gvariable-location-views): New.
703 (gvariable-location-views=incompat5): New.
704 * config.in: Rebuilt.
705 * configure: Rebuilt.
706 * configure.ac: Test assembler for view support.
707 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
708 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
709 * dwarf2out.c (var_loc_view): New typedef.
710 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
711 (dwarf2out_locviews_in_attribute): New.
712 (dwarf2out_locviews_in_loclist): New.
713 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
714 (enum dw_line_info_opcode): Add LI_adv_address.
715 (struct dw_line_info_table): Add view.
716 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
717 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
718 (zero_view_p): New variable.
719 (ZERO_VIEW_P): New macro.
720 (output_asm_line_debug_info): New.
721 (struct var_loc_node): Add view.
722 (add_AT_view_list, AT_loc_list): New.
723 (add_var_loc_to_decl): Add view param. Test it against last.
724 (new_loc_list): Add view params. Record them.
725 (AT_loc_list_ptr): Handle loc and view lists.
726 (view_list_to_loc_list_val_node): New.
727 (print_dw_val): Handle dw_val_class_view_list.
728 (size_of_die): Likewise.
729 (value_format): Likewise.
730 (loc_list_has_views): New.
731 (gen_llsym): Set vl_symbol too.
732 (maybe_gen_llsym, skip_loc_list_entry): New.
733 (dwarf2out_maybe_output_loclist_view_pair): New.
734 (output_loc_list): Output view list or entries too.
735 (output_view_list_offset): New.
736 (output_die): Handle dw_val_class_view_list.
737 (output_dwarf_version): New.
738 (output_compilation_unit_header): Use it.
739 (output_skeleton_debug_sections): Likewise.
740 (output_rnglists, output_line_info): Likewise.
741 (output_pubnames, output_aranges): Update version comments.
742 (output_one_line_info_table): Output view numbers in asm comments.
743 (dw_loc_list): Determine current endview, pass it to new_loc_list.
744 Call maybe_gen_llsym.
745 (loc_list_from_tree_1): Adjust.
746 (add_AT_location_description): Create view list attribute if
747 needed, check it's absent otherwise.
748 (convert_cfa_to_fb_loc_list): Adjust.
749 (maybe_emit_file): Call output_asm_line_debug_info for test.
750 (dwarf2out_var_location): Reset views as needed. Precompute
751 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
753 (new_line_info_table): Reset next view.
754 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
755 (dwarf2out_source_line): Likewise. Output view resets and labels to
756 the assembler, or select appropriate line info opcodes.
757 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
758 (optimize_string_length): Catch it. Adjust.
759 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
760 dw_val_class_view_list, and remove it if no longer needed.
761 (hash_loc_list): Hash view numbers.
762 (loc_list_hasher::equal): Compare them.
763 (optimize_location_lists): Check whether a view list symbol is
764 needed, and whether the locview attribute is present, and
765 whether they match. Remove the locview attribute if no longer
767 (index_location_lists): Call skip_loc_list_entry for test.
768 (dwarf2out_finish): Call output_asm_line_debug_info for test.
769 Use output_dwarf_version.
770 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
771 (struct dw_val_node): Add val_view_list.
772 * final.c (SEEN_NEXT_VIEW): New.
773 (set_next_view_needed): New.
774 (clear_next_view_needed): New.
775 (maybe_output_next_view): New.
776 (final_start_function): Rename to...
777 (final_start_function_1): ... this. Take pointer to FIRST,
778 add SEEN parameter. Emit param bindings in the initial view.
779 (final_start_function): Reintroduce SEEN-less interface.
780 (final): Rename to...
781 (final_1): ... this. Take SEEN parameter. Output final pending
782 next view at the end.
783 (final): Reintroduce seen-less interface.
784 (final_scan_insn): Output pending next view before switching
785 sections or ending a block. Mark the next view as needed when
786 outputting variable locations. Notify debug backend of section
787 changes, and of location view changes.
788 (rest_of_handle_final): Adjust.
789 * toplev.c (process_options): Autodetect value for debug variable
790 location views option. Warn on incompat5 without -gdwarf-5.
791 * doc/invoke.texi (gvariable-location-views): New.
792 (gvariable-location-views=incompat5): New.
793 (gno-variable-location-views): New.
795 2018-02-08 David Malcolm <dmalcolm@redhat.com>
797 PR tree-optimization/84136
798 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
799 that the result of find_edge is non-NULL.
801 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
804 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
805 storing integer register in SImode. Fix cost of 256 and 512
806 byte aligned SSE register store.
808 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
810 * config/i386/i386.c (ix86_multiplication_cost): Fix
811 multiplication cost for TARGET_AVX512DQ.
813 2018-02-08 Marek Polacek <polacek@redhat.com>
815 PR tree-optimization/84238
816 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
819 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
821 PR tree-optimization/84265
822 * tree-vect-stmts.c (vectorizable_store): Don't treat
823 VMAT_CONTIGUOUS accesses as grouped.
824 (vectorizable_load): Likewise.
826 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
828 PR tree-optimization/81635
829 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
830 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
831 (test_round_for_mask): New functions.
832 (wide_int_cc_tests): Call test_round_for_mask.
833 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
834 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
835 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
836 range returned by get_range_info.
838 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
841 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
842 * symtab.c: Include builtins.h
843 (symtab_node::output_to_lto_symbol_table_p): Move here
844 from lto-streamer-out.c:output_symbol_p.
845 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
846 (output_symbol_p): Move all logic to symtab.c
847 (produce_symtab): Update.
849 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
851 * config/s390/s390-opts.h (enum indirect_branch): Define.
852 * config/s390/s390-protos.h (s390_return_addr_from_memory)
853 (s390_indirect_branch_via_thunk)
854 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
855 (enum s390_indirect_branch_type): Define.
856 * config/s390/s390.c (struct s390_frame_layout, struct
857 machine_function): Remove.
858 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
859 (indirect_branch_table_label_no, indirect_branch_table_name):
861 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
862 (enum s390_indirect_branch_option): Define.
863 (s390_return_addr_from_memory): New function.
864 (s390_handle_string_attribute): New function.
865 (s390_attribute_table): Add new attribute handler.
866 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
867 (s390_indirect_branch_via_thunk): New function.
868 (s390_indirect_branch_via_inline_thunk): New function.
869 (s390_function_ok_for_sibcall): When jumping via thunk disallow
870 sibling call optimization for non z10 compiles.
871 (s390_emit_call): Force indirect branch target to be a single
872 register. Add r1 clobber for non-z10 compiles.
873 (s390_emit_epilogue): Emit return jump via return_use expander.
874 (s390_reorg): Handle JUMP_INSNs as execute targets.
875 (s390_option_override_internal): Perform validity checks for the
876 new command line options.
877 (s390_indirect_branch_attrvalue): New function.
878 (s390_indirect_branch_settings): New function.
879 (s390_set_current_function): Invoke s390_indirect_branch_settings.
880 (s390_output_indirect_thunk_function): New function.
881 (s390_code_end): Implement target hook.
882 (s390_case_values_threshold): Implement target hook.
883 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
885 * config/s390/s390.h (struct s390_frame_layout)
886 (struct machine_function): Move here from s390.c.
887 (TARGET_INDIRECT_BRANCH_NOBP_RET)
888 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
889 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
890 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
891 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
892 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
893 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
894 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
895 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
896 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
897 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
898 (mnemonic attribute): Add values which aren't recognized
900 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
901 pattern for branch conversion. Fix mnemonic attribute.
902 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
903 indirect branch via thunk if requested.
904 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
905 ("*indirect_jump"): Disable for branch conversion using out of
907 ("indirect_jump_via_thunk<mode>_z10")
908 ("indirect_jump_via_thunk<mode>")
909 ("indirect_jump_via_inlinethunk<mode>_z10")
910 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
911 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
912 ("casesi_jump_via_inlinethunk<mode>_z10")
913 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
914 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
915 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
916 ("*indirect2_jump"): Disable for branch conversion.
917 ("casesi_jump"): Turn into expander and expand patterns for branch
919 ("return_use"): New expander.
920 ("*return"): Emit return via thunk and rename it to ...
921 ("*return<mode>"): ... this one.
922 * config/s390/s390.opt: Add new options and and enum for the
925 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
927 * lra-constraints.c (match_reload): Unconditionally use
928 gen_lowpart_SUBREG, rather than selecting between that
929 and equivalent gen_rtx_SUBREG code.
931 2018-02-08 Richard Biener <rguenther@suse.de>
933 PR tree-optimization/84233
934 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
935 changed flag instead of boguously re-using phi_inserted.
937 2018-02-08 Martin Jambor <mjambor@suse.cz>
939 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
940 static local variables.
942 2018-02-08 Richard Biener <rguenther@suse.de>
944 PR tree-optimization/84278
945 * tree-vect-stmts.c (vectorizable_store): When looking for
946 smaller vector types to perform grouped strided loads/stores
947 make sure the mode is supported by the target.
948 (vectorizable_load): Likewise.
950 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
952 * config/aarch64/aarch64.c (aarch64_components_for_bb):
953 Increase LDP/STP opportunities by adding adjacent callee-saves.
955 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
957 PR rtl-optimization/84068
958 PR rtl-optimization/83459
959 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
961 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
963 PR tree-optimization/84224
964 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
965 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
968 2018-02-07 Iain Sandoe <iain@codesourcery.com>
971 * config/rs6000/altivec.md (*restore_world): Remove LR use.
972 * config/rs6000/predicates.md (restore_world_operation): Adjust op
973 count, remove one USE.
975 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
977 * doc/install.texi (Configuration): Document the
978 --with-long-double-format={ibm,ieee} PowerPC configuration
982 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
983 Convert from define_expand to be define_insn_and_split. Rework
984 float/double/_Float128 conversions to QI/HI/SImode to work with
985 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
986 conversions to QI/HImode types did a store and then a load to
987 truncate the value. For conversions to VSX registers, don't split
988 the insn, instead emit the code directly. Use the code iterator
989 any_fix to combine signed and unsigned conversions.
990 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
991 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
992 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
993 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
994 (fix_<mode>di2_hw): Likewise.
995 (fixuns_<mode>di2_hw): Likewise.
996 (fix_<mode>si2_hw): Likewise.
997 (fixuns_<mode>si2_hw): Likewise.
998 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
999 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1000 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
1001 fix<uns>_trunc<SFDF:mode>si2_p8.
1002 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
1004 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1005 (fix<uns>_<mode>_mem): Likewise.
1006 (fctiw<u>z_<mode>_mem): Likewise.
1007 (fix<uns>_<mode>_mem): Likewise.
1008 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1009 the register allocator from doing a direct move to the GPRs to do
1010 a store, and instead use the ISA 3.0 store byte/half-word from
1011 vector register instruction. For IEEE 128-bit floating point,
1012 also optimize stores of 32-bit ints.
1013 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1015 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1017 * genextract.c (push_pathstr_operand): New function to support
1019 (walk_rtx): Call push_pathstr_operand.
1020 (print_path): Support [a-zA-Z].
1022 2018-02-07 Richard Biener <rguenther@suse.de>
1024 PR tree-optimization/84037
1025 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1026 (cse_and_gimplify_to_preheader): Declare.
1027 (vect_get_place_in_interleaving_chain): Likewise.
1028 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1030 (_loop_vec_info::~_loop_vec_info): Delete it.
1031 (cse_and_gimplify_to_preheader): New function.
1032 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1033 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1034 (vectorizable_load): Likewise. For grouped stores always base
1035 the IV on the first element.
1036 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1037 condition before gimplifying.
1039 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1041 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1042 *DIV_EXPR and *MOD_EXPR.
1044 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1047 * config/i386/i386.c (ix86_option_override_internal): Mask out
1048 the CF_SET bit when checking -fcf-protection.
1050 2018-02-07 Tom de Vries <tom@codesourcery.com>
1053 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1056 2018-02-07 Richard Biener <rguenther@suse.de>
1058 PR tree-optimization/84204
1059 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1062 PR tree-optimization/84205
1063 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1064 special-case isl_ast_op_zdiv_r.
1066 PR tree-optimization/84223
1067 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1068 Only add conditions from within the region.
1069 (gather_bbs::after_dom_children): Adjust.
1071 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1074 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1075 * config/avr/avr.md: Only post-reload split REG-REG moves if
1076 either register is REGERAL_REG_P.
1078 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1080 PR tree-optimization/84235
1081 * tree-ssa-scopedtables.c
1082 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1083 if the subtraction is performed in floating point type where NaNs are
1084 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1085 build 1. Formatting fix.
1087 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1090 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1091 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1092 and skip it regardless of bb boundaries. Use CALL_P macro,
1093 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1094 unnecessarily, formatting fix.
1096 2018-02-06 Michael Collison <michael.collison@arm.com>
1098 * config/arm/thumb2.md:
1099 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1100 (*thumb_mov_notscc): Ditto.
1102 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1105 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1106 unsigned_fix, not "s".
1108 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1110 * configure.ac (gcc_fn_eh_frame_ro): New function.
1111 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1112 correct .eh_frame permissions.
1113 * configure: Regenerate.
1115 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1117 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1120 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1122 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1123 Display warning message for -mno-speculate-indirect-jumps.
1125 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1127 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1129 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1131 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1133 PR tree-optimization/84225
1134 * tree-eh.c (find_trapping_overflow): Only call
1135 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1137 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1140 * config/i386/i386.c: Reimplement the check of possible options
1141 -mibt/-mshstk conbination. Change error messages.
1142 * doc/invoke.texi: Fix a typo: remove extra '='.
1144 2018-02-06 Marek Polacek <polacek@redhat.com>
1146 PR tree-optimization/84228
1147 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1149 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1152 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1153 emitted arch directives.
1154 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1155 __ARM_FEATURE_COPROC before changing architectures.
1157 2018-02-06 Richard Biener <rguenther@suse.de>
1159 * config/i386/i386.c (print_reg): Fix typo.
1160 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1162 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1164 * configure: Regenerate.
1166 2018-02-05 Martin Sebor <msebor@redhat.com>
1168 PR tree-optimization/83369
1169 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1172 2018-02-05 Martin Liska <mliska@suse.cz>
1174 * doc/invoke.texi: Cherry-pick upstream r323995.
1176 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1178 * ira.c (ira_init_register_move_cost): Adjust comment.
1180 2018-02-05 Martin Liska <mliska@suse.cz>
1182 PR gcov-profile/84137
1183 * doc/gcov.texi: Fix typo in documentation.
1185 2018-02-05 Martin Liska <mliska@suse.cz>
1187 PR gcov-profile/83879
1188 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1189 using dlopen functionality.
1191 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1193 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1194 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1195 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1196 _mm_maskz_range_round_ss): New intrinsics.
1197 (__builtin_ia32_rangesd128_round)
1198 (__builtin_ia32_rangess128_round): Remove.
1199 (__builtin_ia32_rangesd128_mask_round,
1200 __builtin_ia32_rangess128_mask_round): New builtins.
1201 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1202 __builtin_ia32_rangess128_round): Remove.
1203 (__builtin_ia32_rangesd128_mask_round,
1204 __builtin_ia32_rangess128_mask_round): New builtins.
1205 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1206 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1207 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1208 "<round_saeonly_constraint>")): Changed to ...
1209 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1210 "<round_saeonly_scalar_constraint>")): ... this.
1211 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1212 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1213 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1214 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1215 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1217 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1219 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1221 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1222 Remove all values except native, 8540 and 8548.
1224 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1226 * config/i386/i386.c (ix86_output_function_return): Pass
1227 INVALID_REGNUM, instead of -1, as invalid register number to
1228 indirect_thunk_name and output_indirect_thunk.
1230 2018-02-02 Julia Koval <julia.koval@intel.com>
1232 * config.gcc: Add -march=icelake.
1233 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1234 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1235 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1236 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1237 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1238 (processor_target_table): Add icelake.
1239 (ix86_option_override_internal): Handle new PTAs.
1240 (get_builtin_code_for_version): Handle icelake.
1241 (M_INTEL_COREI7_ICELAKE): New.
1242 (fold_builtin_cpu): Handle icelake.
1243 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1244 * doc/invoke.texi: Add -march=icelake.
1246 2018-02-02 Julia Koval <julia.koval@intel.com>
1248 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1249 to wide_int_bitmask.
1250 * wide-int-bitmask.h: New.
1252 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1255 * config/i386/i386.md: Replace Pmode with word_mode in
1256 builtin_setjmp_setup and builtin_longjmp to support x32.
1258 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1262 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1264 (rs6000_supported_cpu_names): New static variable.
1265 (linux_cpu_translation_table): Likewise.
1266 (elf_platform) <cpu>: Define new static variable and use it.
1267 Translate kernel AT_PLATFORM name to canonical name if needed.
1268 Error if platform name is unknown.
1270 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1273 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1275 2018-02-01 Jeff Law <law@redhat.com>
1278 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1279 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1280 the scratch if RELEASE_VIA_POP is false.
1281 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1282 If we have to save a temporary register, decrement SIZE appropriately.
1283 Pass new arguments to release_scratch_register_on_entry.
1284 (ix86_adjust_stack_and_probe): Likewise.
1285 (ix86_emit_probe_stack_range): Pass new arguments to
1286 release_scratch_register_on_entry.
1288 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1290 PR rtl-optimization/84157
1291 * combine.c (change_zero_ext): Use REG_P predicate in
1292 front of HARD_REGISTER_P predicate.
1294 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1296 * config/avr/avr.c (avr_option_override): Move disabling of
1297 -fdelete-null-pointer-checks to...
1298 * common/config/avr/avr-common.c (avr_option_optimization_table):
1301 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1303 PR tree-optimization/81635
1304 * tree-data-ref.c (split_constant_offset_1): For types that
1305 wrap on overflow, try to use range info to prove that wrapping
1308 2018-02-01 Renlin Li <renlin.li@arm.com>
1311 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1313 (aarch64_register_move_cost): Likewise.
1314 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1316 (REG_CLASS_NAMES): Likewise.
1317 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1318 TAILCALL_ADDR_REGS. Remove IP registers.
1319 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1321 2018-02-01 Richard Biener <rguenther@suse.de>
1323 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1324 for specifying RPO order and allow NULL for that.
1325 * domwalk.c (dom_walker::dom_walker): Likewise.
1326 (dom_walker::walk): Handle NULL RPO order.
1327 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1329 (rewrite_update_dom_walker): Likewise.
1330 (mark_def_dom_walker): Likewise.
1332 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1334 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1335 (aarch64_maybe_expand_sve_subreg_move): Declare.
1336 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1337 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1339 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1340 that are semantically a reverse operation.
1341 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1342 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1343 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1345 (aarch64_can_change_mode_class): For big-endian, forbid changes
1346 between two SVE modes if they have different element sizes.
1348 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1350 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1351 the TImode handling for big-endian targets.
1353 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1355 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1356 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1358 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1359 Remove BSWAP handing for big-endian targets and use the form of
1360 LD1RQ appropariate for the mode.
1362 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1364 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1365 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1368 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1371 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1372 check for operands that need to go through aarch64_sve_reload_be.
1374 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1376 PR tree-optimization/81661
1377 PR tree-optimization/84117
1378 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1379 * tree-eh.c: Include gimplify.h.
1380 (find_trapping_overflow, replace_trapping_overflow,
1381 rewrite_to_non_trapping_overflow): New functions.
1382 * tree-vect-loop.c: Include tree-eh.h.
1383 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1384 * tree-data-ref.c: Include tree-eh.h.
1385 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1387 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1389 PR rtl-optimization/84123
1390 * combine.c (change_zero_ext): Check if hard register satisfies
1391 can_change_dest_mode before calling gen_lowpart_SUBREG.
1393 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1396 * ira.c (ira_init_register_move_cost): Remove assert.
1398 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1400 PR rtl-optimization/84071
1401 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1402 * doc/tm.texi: Regenerate.
1404 2018-01-31 Richard Biener <rguenther@suse.de>
1406 PR tree-optimization/84132
1407 * tree-data-ref.c (analyze_miv_subscript): Properly
1408 check whether evolution_function_is_affine_multivariate_p
1409 before calling gcd_of_steps_may_divide_p.
1411 2018-01-31 Julia Koval <julia.koval@intel.com>
1414 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1415 * config/i386/i386.md (rdpid_rex64) New.
1416 (rdpid): Make 32bit only.
1418 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1421 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1422 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1424 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1427 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1429 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1431 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1433 PR rtl-optimization/84071
1434 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1435 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1437 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1439 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1440 (arc_attribute_table): Add 'aux' attribute.
1441 (arc_in_small_data_p): Consider aux like variables.
1442 (arc_is_aux_reg_p): New function.
1443 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1444 (arc_get_aux_arg): New function.
1445 (prepare_move_operands): Handle aux-register access.
1446 (arc_handle_aux_attribute): New function.
1447 * doc/extend.texi (ARC Variable attributes): Add subsection.
1449 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1451 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1452 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1453 (arc_attribute_table): Add 'uncached' attribute.
1454 (arc_print_operand): Print '.di' flag for uncached memory
1456 (arc_in_small_data_p): Do not consider for small data the uncached
1458 (arc_is_uncached_mem_p): New function.
1459 * config/arc/predicates.md (compact_store_memory_operand): Check
1460 for uncached memory accesses.
1461 (nonvol_nonimm_operand): Likewise.
1462 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1464 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1467 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1468 falign-loops=): Add Optimization flag.
1470 2018-01-30 Jeff Law <law@redhat.com>
1473 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1474 INT_REGISTERS_SAVED. Check it prior to calling
1475 get_scratch_register_on_entry.
1476 (ix86_adjust_stack_and_probe): Similarly.
1477 (ix86_emit_probe_stack_range): Similarly.
1478 (ix86_expand_prologue): Corresponding changes.
1480 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1483 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1484 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1486 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1489 * lra-constraints.c (curr_insn_transform): Process AND in the
1492 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1494 PR rtl-optimization/83986
1495 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1496 dependence against last_pending_memory_flush in addition to
1499 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1501 PR tree-optimization/81611
1502 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1505 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1508 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1511 2018-01-30 Richard Biener <rguenther@suse.de>
1512 Jakub Jelinek <jakub@redhat.com>
1514 PR tree-optimization/84111
1515 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1516 inner loops added during recursion, as they don't have up-to-date
1519 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1522 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1523 (can_inline_edge_by_limits_p): ... here.
1524 (can_early_inline_edge_p, check_callers,
1525 update_caller_keys, update_callee_keys, recursive_inlining,
1526 add_new_edges_to_heap, speculation_useful_p,
1527 inline_small_functions,
1528 inline_small_functions, flatten_function,
1529 inline_to_all_callers_1): Update.
1531 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1533 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1536 2018-01-30 Richard Biener <rguenther@suse.de>
1538 PR tree-optimization/83008
1539 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1540 invariant and constant vector uses in stmts when they need
1543 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1546 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1547 * configure: Regenerate.
1549 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1551 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1553 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1554 Use gen_rtx_REG rather than gen_lowpart.
1556 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1558 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1559 rather than 0 when creating partial subregs.
1561 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1563 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1566 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1569 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1570 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1571 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1572 flags. This restores the settings used before the 2017-07-24.
1573 Turning off pre increment/decrement/modify allows IVOPTS to
1574 optimize DF/SF loops where the index is an int.
1576 2018-01-29 Richard Biener <rguenther@suse.de>
1577 Kelvin Nilsen <kelvin@gcc.gnu.org>
1580 * tree-vect-stmts.c (vectorizable_call): Don't call
1581 targetm.vectorize_builtin_md_vectorized_function if callee is
1584 2018-01-22 Carl Love <cel@us.ibm.com>
1586 * doc/extend.tex: Fix typo in second arg in
1587 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1589 2018-01-29 Richard Biener <rguenther@suse.de>
1591 PR tree-optimization/84086
1592 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1593 (flush_ssaname_freelist): When SSA names were released reset
1594 the SCEV hash table.
1596 2018-01-29 Richard Biener <rguenther@suse.de>
1598 PR tree-optimization/84057
1599 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1600 removed paths when removing edges.
1602 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1604 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1605 -mfunction-return=@var{choice}.
1607 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1610 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1612 (layout::print_source_line): Likewise.
1613 (test_get_line_width_without_trailing_whitespace): Add test cases.
1615 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1618 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1621 2018-01-26 Jim Wilson <jimw@sifive.com>
1623 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1625 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1628 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1630 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1631 and CMP + SUB-immediate -> SUBS.
1633 2018-01-26 Martin Sebor <msebor@redhat.com>
1635 PR tree-optimization/83896
1636 * tree-ssa-strlen.c (get_string_len): Rename...
1637 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1638 Avoid assuming length is constant.
1639 (handle_char_store): Use HOST_WIDE_INT for string length.
1641 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1644 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1645 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1647 2018-01-26 Richard Biener <rguenther@suse.de>
1649 PR rtl-optimization/84003
1650 * dse.c (record_store): Only record redundant stores when
1651 the earlier store aliases at least all accesses the later one does.
1653 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1655 PR rtl-optimization/83985
1656 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1657 REG_CFA_RESTORE insns.
1658 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1659 REG_CFA_RESTORE insns here.
1662 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1663 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1665 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1667 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1668 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1669 (arc_init): Likewise.
1670 (arc_override_options): Likewise.
1671 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1673 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1675 * config/arc/arc.h (TARGET_DBNZ): Define.
1676 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1677 properly set the tune attribute.
1678 (dbnz): Use TARGET_DBNZ guard.
1679 * config/arc/arc.opt (mtune): Add core3 option.
1681 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1683 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1684 recognize new pic like addresses.
1685 (arc_delegitimize_address): Clean up.
1687 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1689 * config/arc/arc-arches.def: Option mrf16 valid for all
1691 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1692 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1693 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1694 * config/arc/arc-tables.opt: Regenerate.
1695 * config/arc/arc.c (arc_conditional_register_usage): Handle
1696 reduced register file case.
1697 (arc_file_start): Set must have build attributes.
1698 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1700 * config/arc/arc.opt (mrf16): Add new option.
1701 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1702 * config/arc/genmultilib.awk: Handle new mrf16 option.
1703 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1704 * config/arc/t-multilib: Regenerate.
1705 * doc/invoke.texi (ARC Options): Document mrf16 option.
1707 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1709 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1710 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1711 (arc_attribute_table): Add 'secure_call' attribute.
1712 (arc_print_operand): Print secure call operand.
1713 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1715 (arc_is_secure_call_p): New function. * config/arc/arc.md
1716 (call_i): Add support for sjli instruction.
1717 (call_value_i): Likewise.
1718 * config/arc/constraints.md (Csc): New constraint.
1720 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1721 John Eric Martin <John.Martin@emmicro-us.com>
1723 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1724 * config/arc/arc.c (_arc_jli_section): New struct.
1725 (arc_jli_section): New type.
1726 (rc_jli_sections): New static variable.
1727 (arc_handle_jli_attribute): New function.
1728 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1729 (arc_file_end): New function.
1730 (TARGET_ASM_FILE_END): Define.
1731 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1732 (arc_add_jli_section): New function.
1733 (jli_call_scan): Likewise.
1734 (arc_reorg): Call jli_call_scan.
1735 (arc_output_addsi): Remove 'S' from printing asm operand.
1736 (arc_is_jli_call_p): New function.
1737 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1739 (movhi_insn): Likewise.
1740 (movsi_insn): Likewise.
1741 (movsi_set_cc_insn): Likewise.
1742 (loadqi_update): Likewise.
1743 (load_zeroextendqisi_update): Likewise.
1744 (load_signextendqisi_update): Likewise.
1745 (loadhi_update): Likewise.
1746 (load_zeroextendhisi_update): Likewise.
1747 (load_signextendhisi_update): Likewise.
1748 (loadsi_update): Likewise.
1749 (loadsf_update): Likewise.
1750 (movsicc_insn): Likewise.
1751 (bset_insn): Likewise.
1752 (bxor_insn): Likewise.
1753 (bclr_insn): Likewise.
1754 (bmsk_insn): Likewise.
1755 (bicsi3_insn): Likewise.
1756 (cmpsi_cc_c_insn): Likewise.
1757 (movsi_ne): Likewise.
1758 (movsi_cond_exec): Likewise.
1759 (clrsbsi2): Likewise.
1770 (call_i): Remove 'S' asm letter, add jli instruction.
1771 (call_value_i): Likewise.
1772 * config/arc/arc.op (mjli-always): New option.
1773 * config/arc/constraints.md (Cji): New constraint.
1774 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1776 (subsf3_fpx): Likewise.
1777 (mulsf3_fpx): Likewise.
1778 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1780 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1782 * doc/invoke.texi (ARC): Document mjli-always option.
1784 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1786 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
1787 avoid addition with 0 and use incw and decw where possible.
1789 2018-01-26 Richard Biener <rguenther@suse.de>
1791 PR tree-optimization/81082
1792 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1793 association if it requires casting to unsigned.
1794 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1795 from fold_plusminus_mult_expr to catch important cases late when
1796 range info is available.
1798 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1800 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1801 * configure.ac (hidden_linkonce): New test.
1802 * configure: Regenerate.
1803 * config.in: Regenerate.
1805 2018-01-26 Julia Koval <julia.koval@intel.com>
1807 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1808 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1809 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1810 _mm_mask_bitshuffle_epi64_mask): Fix type.
1811 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1812 USI_FTYPE_V4DI_V4DI_USI): Remove.
1813 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1814 __builtin_ia32_vpshufbitqmb256_mask,
1815 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1816 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1817 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1819 2018-01-26 Alan Modra <amodra@gmail.com>
1822 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1823 UNSPEC_VBPERMQ. Sort other unspecs.
1825 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1827 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1829 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1832 * predict.c (drop_profile): Do not push/pop cfun; update also
1834 (handle_missing_profiles): Fix logic looking for zero profiles.
1836 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1839 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1840 on functions with #pragma omp declare simd or functions with simd
1842 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1843 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1844 Remove trailing \n from warning_at calls.
1846 2018-01-25 Tom de Vries <tom@codesourcery.com>
1849 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1850 for neutered workers.
1852 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1855 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1856 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1858 2018-01-24 Jeff Law <law@redhat.com>
1861 * i386.c (get_probe_interval): Move to earlier point.
1862 (ix86_compute_frame_layout): If -fstack-clash-protection and
1863 the frame is larger than the probe interval, then use pushes
1864 to save registers rather than reg->mem moves.
1865 (ix86_expand_prologue): Remove conditional for int_registers_saved
1868 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1871 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1872 min/max for never referenced object.
1874 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1877 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1879 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1880 attributes from DECL_ATTRIBUTES (decl) without affecting
1881 DECL_ATTRIBUTES (current_function_decl).
1882 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1883 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1885 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1887 PR tree-optimization/83979
1888 * fold-const.c (fold_comparison): Use constant_boolean_node
1889 instead of boolean_{true,false}_node.
1891 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1893 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1896 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1898 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1899 Simplify the clause that sets the length attribute.
1900 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1901 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1902 clause that sets the length attribute.
1903 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1905 2018-01-24 Tom de Vries <tom@codesourcery.com>
1908 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1909 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1910 Add strict parameter.
1911 (prevent_branch_around_nothing): Insert dummy insn between branch to
1912 label and label with no ptx insn inbetween.
1913 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1915 2018-01-24 Tom de Vries <tom@codesourcery.com>
1918 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1919 for neutered threads in warp.
1920 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1922 2018-01-24 Richard Biener <rguenther@suse.de>
1924 PR tree-optimization/83176
1925 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1928 2018-01-24 Richard Biener <rguenther@suse.de>
1930 PR tree-optimization/82819
1931 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1932 code generating pluses that are no-ops in the target precision.
1934 2018-01-24 Richard Biener <rguenther@suse.de>
1937 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1939 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1941 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1942 to merge probabilities.
1943 * predict.c (probably_never_executed): Also mark as cold functions
1944 with global 0 profile and guessed local profile.
1945 * profile-count.c (profile_probability::combine_with_count): New
1947 * profile-count.h (profile_probability::operator*,
1948 profile_probability::operator*=, profile_probability::operator/,
1949 profile_probability::operator/=): Reduce precision to adjusted
1950 and set value to guessed on contradictory divisions.
1951 (profile_probability::combine_with_freq): Remove.
1952 (profile_probability::combine_wiht_count): Declare.
1953 (profile_count::force_nonzero):: Set to adjusted.
1954 (profile_count::probability_in):: Set quality to adjusted.
1955 * tree-ssa-tail-merge.c (replace_block_by): Use
1958 2018-01-23 Andrew Waterman <andrew@sifive.com>
1959 Jim Wilson <jimw@sifive.com>
1961 * config/riscv/riscv.c (riscv_stack_boundary): New.
1962 (riscv_option_override): Set riscv_stack_boundary. Handle
1963 riscv_preferred_stack_boundary_arg.
1964 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1965 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1966 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1967 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1968 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1969 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1971 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1974 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1975 of struct ix86_frame.
1976 (ix86_expand_epilogue): Likewise. Add a local variable for
1977 the reg_save_offset field in struct ix86_frame.
1979 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1981 PR tree-optimization/82604
1982 * tree-loop-distribution.c (enum partition_kind): New enum item
1983 PKIND_PARTIAL_MEMSET.
1984 (partition_builtin_p): Support above new enum item.
1985 (generate_code_for_partition): Ditto.
1986 (compute_access_range): Differentiate cases that equality can be
1987 proven at all loops, the innermost loops or no loops.
1988 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1989 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1990 (finalize_partitions, distribute_loop): Don't fuse partition of
1991 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1992 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1995 2018-01-23 Martin Liska <mliska@suse.cz>
1997 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1998 order to ignore the predictor.
1999 (PRED_POLYMORPHIC_CALL): Likewise.
2000 (PRED_RECURSIVE_CALL): Likewise.
2002 2018-01-23 Martin Liska <mliska@suse.cz>
2004 * tree-profile.c (tree_profiling): Print function header to
2005 aware reader which function we are working on.
2006 * value-prof.c (gimple_find_values_to_profile): Do not print
2007 not interesting value histograms.
2009 2018-01-23 Martin Liska <mliska@suse.cz>
2011 * profile-count.h (enum profile_quality): Add
2012 profile_uninitialized as the first value. Do not number values
2013 as they are zero based.
2014 (profile_count::verify): Update sanity check.
2015 (profile_probability::verify): Likewise.
2017 2018-01-23 Nathan Sidwell <nathan@acm.org>
2019 * doc/invoke.texi (ffor-scope): Deprecate.
2021 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2023 PR tree-optimization/83510
2024 * domwalk.c (set_all_edges_as_executable): New function.
2025 (dom_walker::dom_walker): Convert bool param
2026 "skip_unreachable_blocks" to enum reachability. Move setup of
2027 edge flags to set_all_edges_as_executable and only do it when
2028 reachability is REACHABLE_BLOCKS.
2029 * domwalk.h (enum dom_walker::reachability): New enum.
2030 (dom_walker::dom_walker): Convert bool param
2031 "skip_unreachable_blocks" to enum reachability.
2032 (set_all_edges_as_executable): New decl.
2033 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2034 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2036 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2037 but converting true to REACHABLE_BLOCKS.
2038 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2040 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2041 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2042 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2044 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2045 if check_all_array_refs will be called.
2047 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2049 * tree.c (selftest::test_location_wrappers): Add more test
2052 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2054 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2055 (selftest::test_bit_in_range): Likewise.
2057 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2060 * doc/sourcebuild.texi (vect_float): Say that the selector
2061 only describes the situation when -funsafe-math-optimizations is on.
2062 (vect_float_strict): Document.
2064 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2066 PR tree-optimization/83965
2067 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2068 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2069 instead of checking only for a reduction.
2070 (vect_recog_widen_sum_pattern): Likewise.
2072 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2074 * predict.c (probably_never_executed): Only use precise profile info.
2075 (compute_function_frequency): Skip after inlining hack since we now
2076 have quality checking.
2078 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2080 * profile-count.h (profile_probability::very_unlikely,
2081 profile_probability::unlikely, profile_probability::even): Set
2082 precision to guessed.
2084 2018-01-23 Richard Biener <rguenther@suse.de>
2086 PR tree-optimization/83963
2087 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2088 Properly terminate dominator walk when crossing the exit edge not
2089 when visiting its source block.
2091 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2094 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2095 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2097 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2099 PR tree-optimization/83957
2100 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2101 semicolon after for body surrounded by braces.
2103 PR tree-optimization/83081
2104 * profile-count.h (profile_probability::split): New method.
2105 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2106 Use profile_probability::split.
2107 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2108 when splitting a single conditional jump into 2.
2110 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2112 PR tree-optimization/69452
2113 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2116 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2118 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2119 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2120 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2122 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2124 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2125 * config/rl78/rl78.md (movdi): New define_expand.
2126 * config/rl78/rl78.c (rl78_split_movdi): New function.
2128 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2131 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2133 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2134 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2135 128-bit to produce an UNSPEC move to get the double word with the
2136 signbit and then a shift directly to do signbit.
2137 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2138 implementation with a new version that just does either a direct
2139 move or a regular move. Move memory interface to separate insns.
2140 Move insns so they are next to the expander.
2141 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2142 with signbit move. Split big and little endian case.
2143 (signbit<mode>2_dm_mem_le): Likewise.
2144 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2145 (signbit<mode>2_dm2): Likewise.
2147 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2149 * config/rl78/rl78.md (anddi3): New define_expand.
2151 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2153 * config/rl78/rl78.md (umindi3): New define_expand.
2155 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2157 * config/rl78/rl78.md (smindi3): New define_expand.
2159 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2161 * config/rl78/rl78.md (smaxdi3): New define_expand.
2163 2018-01-22 Carl Love <cel@us.ibm.com>
2165 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2166 LVX_V1TI): Add macro expansion.
2167 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2168 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2169 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2170 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2171 Change check to determine if the instruction is a byte reversing
2172 entry. Fix typo in comment.
2173 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2174 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2175 Add def_builtin calls for new builtins.
2176 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2177 Add define_insn expansion.
2179 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2181 * config/rl78/rl78.md (umaxdi3): New define_expand.
2183 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2185 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2186 for non-QImode registers.
2188 2018-01-22 Richard Biener <rguenther@suse.de>
2190 PR tree-optimization/83963
2191 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2192 including the loop exit block.
2193 (scop_detection::merge_sese): Likewise.
2194 (scop_detection::add_scop): Do it here instead.
2196 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2198 * doc/sourcebuild.texi (arm_softfloat): Document.
2200 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2203 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2204 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2205 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2207 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2208 David Edelsohn <dje.gcc@gmail.com>
2211 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2212 Change "crset eq" to "crset 2".
2213 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2214 (*call_indirect_aix<mode>_nospec): Likewise.
2215 (*call_value_indirect_aix<mode>_nospec): Likewise.
2216 (*call_indirect_elfv2<mode>_nospec): Likewise.
2217 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2218 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2219 change assembly output from . to $.
2220 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2221 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2222 (*tablejump<mode>_internal1_nospec): Likewise.
2224 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2227 * config/sh/sh_optimize_sett_clrt.cc:
2228 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2230 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2232 PR tree-optimization/83940
2233 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2234 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2235 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2236 use it to pass back the definition type.
2237 (vect_check_store_rhs): Likewise rhs_dt_out.
2238 (vect_build_gather_load_calls): Add a mask_dt argument and use
2239 it instead of a call to vect_is_simple_use.
2240 (vectorizable_store): Update calls to vect_check_load_store_mask
2241 and vect_check_store_rhs. Use the dt returned by the latter instead
2242 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2243 instead of calls to vect_is_simple_use. Pass the scalar rather
2244 than the vector operand to vect_is_simple_use when handling
2245 second and subsequent copies of an rhs value.
2246 (vectorizable_load): Update calls to vect_check_load_store_mask
2247 and vect_build_gather_load_calls. Use the cached mask_dt and
2248 gs_info.offset_dt instead of calls to vect_is_simple_use.
2250 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2253 * tree-emutls.c: Include gimplify.h.
2254 (lower_emutls_2): New function.
2255 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2256 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2257 it before further processing.
2260 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2261 UINTVAL (trueop1) instead of INTVAL (op1).
2263 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2267 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2268 INCOMING_FRAME_SP_OFFSET if not defined.
2269 (scan_trace): Add ENTRY argument. If true and
2270 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2271 emit a note to adjust the CFA offset.
2272 (create_cfi_notes): Adjust scan_trace callers.
2273 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2274 INCOMING_FRAME_SP_OFFSET in the CIE.
2275 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2276 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2278 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2279 * doc/tm.texi: Regenerated.
2281 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2283 PR rtl-optimization/83147
2284 * lra-constraints.c (remove_inheritance_pseudos): Use
2285 lra_substitute_pseudo_within_insn.
2287 2018-01-19 Tom de Vries <tom@codesourcery.com>
2288 Cesar Philippidis <cesar@codesourcery.com>
2291 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2293 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2296 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2297 spaces for function labels.
2299 2018-01-19 Martin Liska <mliska@suse.cz>
2301 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2302 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2303 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2304 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2305 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2306 (PRED_CONST_RETURN): Change from 69 to 65.
2307 (PRED_NULL_RETURN): Change from 91 to 71.
2308 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2309 (PRED_LOOP_GUARD): Change from 66 to 73.
2311 2018-01-19 Martin Liska <mliska@suse.cz>
2313 * predict.c (predict_insn_def): Add new assert.
2314 (struct branch_predictor): Change type to signed integer.
2315 (test_prediction_value_range): Amend test to cover
2317 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2318 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2319 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2320 (PRED_LOOP_IV_COMPARE): Likewise.
2321 * predict.h (PROB_UNINITIALIZED): Define new constant.
2323 2018-01-19 Martin Liska <mliska@suse.cz>
2325 * predict.c (dump_prediction): Add new format for
2326 analyze_brprob.py script which is enabled with -details
2328 * profile-count.h (precise_p): New function.
2330 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2332 PR tree-optimization/83922
2333 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2334 there are no statements that need masking.
2335 (vect_active_double_reduction_p): New function.
2336 (vect_analyze_loop_operations): Use it when handling phis that
2337 are not in the loop header.
2339 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2341 PR tree-optimization/83914
2342 * tree-vect-loop.c (vectorizable_induction): Don't convert
2343 init_expr or apply the peeling adjustment for inductions
2344 that are nested within the vectorized loop.
2346 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2348 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2351 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2355 * function.h (gimplify_parameters): Add gimple_seq * argument.
2356 * function.c: Include gimple.h and options.h.
2357 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2358 for the added local temporaries if needed.
2359 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2360 if there are any parameter cleanups, wrap whole body into a
2361 try/finally with the cleanups.
2363 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2366 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2367 Use GET_MODE_CLASS for scalar floating point.
2369 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2373 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2374 Fix call of call_cgraph_insertion_hooks.
2376 2018-01-18 Martin Sebor <msebor@redhat.com>
2378 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2380 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2383 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2386 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2389 * common.opt: (-ffile-prefix-map): New option.
2390 * opts.c (common_handle_option): Defer it.
2391 * opts-global.c (handle_common_deferred_options): Handle it.
2392 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2393 * file-prefix-map.h: New file.
2394 (remap_debug_filename, add_debug_prefix_map): ...here.
2395 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2396 * final.c (debug_prefix_map, add_debug_prefix_map
2397 remap_debug_filename): Move to...
2398 * file-prefix-map.c: New file.
2399 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2400 generalize, get rid of alloca(), use strrchr() instead of strchr().
2401 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2402 Implement in terms of add_prefix_map().
2403 (remap_macro_filename, remap_debug_filename): Implement in term of
2405 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2406 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2407 * dbxout.c: Include file-prefix-map.h.
2408 * varasm.c: Likewise.
2409 * vmsdbgout.c: Likewise.
2410 * xcoffout.c: Likewise.
2411 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2412 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2413 * doc/invoke.texi (-ffile-prefix-map): Document.
2414 (-fdebug-prefix-map): Update description.
2416 2018-01-18 Martin Liska <mliska@suse.cz>
2418 * config/i386/i386.c (indirect_thunk_name): Document that also
2420 (output_indirect_thunk): Document why both instructions
2421 (pause and lfence) are generated.
2423 2018-01-18 Richard Biener <rguenther@suse.de>
2425 PR tree-optimization/83887
2426 * graphite-scop-detection.c
2427 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2428 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2429 (scop_detection::merge_sese): Re-implement with a flood-fill
2430 algorithm that properly finds a SESE region if it exists.
2432 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2435 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2436 pointer_diff optimizations use view_convert instead of convert.
2438 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2440 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2441 Generate different code for -mno-speculate-indirect-jumps.
2442 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2443 (*call_indirect_aix<mode>): Disable for
2444 -mno-speculate-indirect-jumps.
2445 (*call_indirect_aix<mode>_nospec): New define_insn.
2446 (*call_value_indirect_aix<mode>): Disable for
2447 -mno-speculate-indirect-jumps.
2448 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2449 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2450 -mno-speculate-indirect-jumps.
2451 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2453 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2455 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2456 long double type, set the flags for noting the default long double
2457 type, even if we don't pass or return a long double type.
2459 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2462 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2465 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2467 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2468 support for merge[hl].
2469 (fold_mergehl_helper): New helper function.
2470 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2471 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2472 (altivec_vmrglw_direct): Add xxmrglw insn.
2474 2018-01-17 Andrew Waterman <andrew@sifive.com>
2476 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2477 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2479 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2482 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2483 call the lto_location_cache before reading the
2484 DECL_SOURCE_LOCATION of the types.
2486 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2487 Richard Sandiford <richard.sandiford@linaro.org>
2489 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2490 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2491 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2492 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2493 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2495 * config/aarch64/constraints.md (aarch64_movti_operand):
2497 * config/aarch64/predicates.md (Uti): Add new constraint.
2499 2018-01-17 Carl Love <cel@us.ibm.com>
2501 * config/rs6000/vsx.md (define_expand xl_len_r,
2502 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2503 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2505 (define_expand, define_insn): Move the shift left from the
2506 define_insn to the define_expand for lxvl and stxvl instructions.
2507 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2508 and XL_LEN_R definitions to PURE.
2510 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2512 * config/i386/i386.c (indirect_thunk_name): Declare regno
2513 as unsigned int. Compare regno with INVALID_REGNUM.
2514 (output_indirect_thunk): Ditto.
2515 (output_indirect_thunk_function): Ditto.
2516 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2517 in the call to output_indirect_thunk_function.
2519 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2522 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2523 rather than the size of inner_type to determine the stack slot size
2524 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2526 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2529 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2532 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2534 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2535 endian Linux systems to optionally enable multilibs for selecting
2536 the long double type if the user configured an explicit type.
2537 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2538 have no long double multilibs if not defined.
2539 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2540 warn if the user used -mabi={ieee,ibm}longdouble and we built
2541 multilibs for long double.
2542 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2543 appropriate multilib option.
2544 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2546 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2547 for building long double multilibs.
2548 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2550 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2552 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2555 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2557 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2560 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2563 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2566 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2568 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2569 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2571 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2573 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2574 different rtl trees depending on TARGET_64BIT.
2575 (rs6000_gen_lvx): Likewise.
2577 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2579 * config/visium/visium.md (nop): Tweak comment.
2580 (hazard_nop): Likewise.
2582 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2584 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2585 -mspeculate-indirect-jumps.
2586 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2587 for -mno-speculate-indirect-jumps.
2588 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2589 (*call_value_indirect_elfv2<mode>): Disable for
2590 -mno-speculate-indirect-jumps.
2591 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2592 (indirect_jump): Emit different RTL for
2593 -mno-speculate-indirect-jumps.
2594 (*indirect_jump<mode>): Disable for
2595 -mno-speculate-indirect-jumps.
2596 (*indirect_jump<mode>_nospec): New define_insn.
2597 (tablejump): Emit different RTL for
2598 -mno-speculate-indirect-jumps.
2599 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2600 (tablejumpsi_nospec): New define_expand.
2601 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2602 (tablejumpdi_nospec): New define_expand.
2603 (*tablejump<mode>_internal1): Disable for
2604 -mno-speculate-indirect-jumps.
2605 (*tablejump<mode>_internal1_nospec): New define_insn.
2606 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2609 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2611 * caller-save.c (insert_save): Drop unnecessary parameter. All
2614 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2615 Richard Biener <rguenth@suse.de>
2618 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2619 return early, inline manually is_gimple_sizepos. Make sure if we
2620 call gimplify_expr we don't end up with a gimple constant.
2621 * tree.c (variably_modified_type_p): Don't return true for
2622 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2623 * gimplify.h (is_gimple_sizepos): Remove.
2625 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2627 PR tree-optimization/83857
2628 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2629 vectorizable_live_operation for pure SLP statements.
2630 (vectorizable_live_operation): Handle PHIs.
2632 2018-01-16 Richard Biener <rguenther@suse.de>
2634 PR tree-optimization/83867
2635 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2636 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2638 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2641 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2642 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2643 If off is not INTEGER_CST, issue a may not be aligned warning
2644 rather than isn't aligned. Use isn%'t rather than isn't.
2645 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2647 <case MULT_EXPR>: Improve the case when bottom and one of the
2648 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2649 operand, in that case check if the other operand is multiple of
2650 bottom divided by the INTEGER_CST operand.
2652 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2655 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2656 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2657 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2658 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2659 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2660 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2661 (pa_function_arg_size): New function.
2663 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2665 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2666 in a separate statement.
2668 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2670 PR tree-optimization/83847
2671 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2672 group gathers and scatters.
2674 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2676 PR rtl-optimization/86620
2677 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2679 PR rtl-optimization/83213
2680 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2681 to last if both are JUMP_INSNs.
2683 PR tree-optimization/83843
2684 * gimple-ssa-store-merging.c
2685 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2686 store_immediate_info for bswap/nop orig_stores.
2688 2018-01-15 Andrew Waterman <andrew@sifive.com>
2690 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2692 <UDIV>: Increase cost if !TARGET_DIV.
2694 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2696 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2697 (define_attr "cr_logical_3op"): New.
2698 (cceq_ior_compare): Adjust.
2699 (cceq_ior_compare_complement): Adjust.
2700 (*cceq_rev_compare): Adjust.
2701 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2702 (is_cracked_insn): Adjust.
2703 (insn_must_be_first_in_group): Adjust.
2704 * config/rs6000/40x.md: Adjust.
2705 * config/rs6000/440.md: Adjust.
2706 * config/rs6000/476.md: Adjust.
2707 * config/rs6000/601.md: Adjust.
2708 * config/rs6000/603.md: Adjust.
2709 * config/rs6000/6xx.md: Adjust.
2710 * config/rs6000/7450.md: Adjust.
2711 * config/rs6000/7xx.md: Adjust.
2712 * config/rs6000/8540.md: Adjust.
2713 * config/rs6000/cell.md: Adjust.
2714 * config/rs6000/e300c2c3.md: Adjust.
2715 * config/rs6000/e500mc.md: Adjust.
2716 * config/rs6000/e500mc64.md: Adjust.
2717 * config/rs6000/e5500.md: Adjust.
2718 * config/rs6000/e6500.md: Adjust.
2719 * config/rs6000/mpc.md: Adjust.
2720 * config/rs6000/power4.md: Adjust.
2721 * config/rs6000/power5.md: Adjust.
2722 * config/rs6000/power6.md: Adjust.
2723 * config/rs6000/power7.md: Adjust.
2724 * config/rs6000/power8.md: Adjust.
2725 * config/rs6000/power9.md: Adjust.
2726 * config/rs6000/rs64.md: Adjust.
2727 * config/rs6000/titan.md: Adjust.
2729 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2731 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2732 ix86_indirect_branch_register logic.
2734 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2736 * config/i386/constraints.md (Bs): Update
2737 ix86_indirect_branch_register check. Don't check
2738 ix86_indirect_branch_register with GOT_memory_operand.
2740 * config/i386/predicates.md (GOT_memory_operand): Don't check
2741 ix86_indirect_branch_register here.
2742 (GOT32_symbol_operand): Likewise.
2744 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2746 * config/i386/predicates.md (constant_call_address_operand):
2747 Rewrite ix86_indirect_branch_register logic.
2748 (sibcall_insn_operand): Likewise.
2750 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2752 * config/i386/constraints.md (Bs): Replace
2753 ix86_indirect_branch_thunk_register with
2754 ix86_indirect_branch_register.
2756 * config/i386/i386.md (indirect_jump): Likewise.
2757 (tablejump): Likewise.
2758 (*sibcall_memory): Likewise.
2759 (*sibcall_value_memory): Likewise.
2760 Peepholes of indirect call and jump via memory: Likewise.
2761 * config/i386/i386.opt: Likewise.
2762 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2763 (GOT_memory_operand): Likewise.
2764 (call_insn_operand): Likewise.
2765 (sibcall_insn_operand): Likewise.
2766 (GOT32_symbol_operand): Likewise.
2768 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2771 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2772 type rather than type addr's type points to.
2773 (expand_omp_atomic_mutex): Likewise.
2774 (expand_omp_atomic): Likewise.
2776 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2779 * config/i386/i386.c (output_indirect_thunk_function): Use
2780 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2781 for __x86_return_thunk.
2783 2018-01-15 Richard Biener <rguenther@suse.de>
2786 * expmed.c (extract_bit_field_1): Fix typo.
2788 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2791 * config/arm/iterators.md (VF): New mode iterator.
2792 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2793 Remove integer-related logic from pattern.
2794 (neon_vabd<mode>_3): Likewise.
2796 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2799 * common.opt (fstrict-overflow): No longer an alias.
2800 (fwrapv-pointer): New option.
2801 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2802 also for pointer types based on flag_wrapv_pointer.
2803 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2804 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2805 opts->x_flag_wrapv got set.
2806 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2807 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2808 POINTER_TYPE_OVERFLOW_UNDEFINED.
2809 * match.pd: Likewise in address comparison pattern.
2810 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2812 2018-01-15 Richard Biener <rguenther@suse.de>
2815 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2816 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2817 Reset type names to their identifier if their TYPE_DECL doesn't
2818 have linkage (and thus is used for ODR and devirt).
2819 (save_debug_info_for_decl): Remove.
2820 (save_debug_info_for_type): Likewise.
2821 (add_tree_to_fld_list): Adjust.
2822 * tree-pretty-print.c (dump_generic_node): Make dumping of
2823 type names more robust.
2825 2018-01-15 Richard Biener <rguenther@suse.de>
2827 * BASE-VER: Bump to 8.0.1.
2829 2018-01-14 Martin Sebor <msebor@redhat.com>
2832 * builtins.c (check_access): Avoid warning when the no-warning bit
2835 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2837 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2838 * ira-color (allocno_hard_regs_compare): Likewise.
2840 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2843 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2844 Use .pushsection/.popsection.
2846 2018-01-14 Martin Sebor <msebor@redhat.com>
2849 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2851 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2853 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2854 entry from extra_headers.
2855 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2856 extra_headers, make the list bitwise identical to the i?86-*-* one.
2858 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2860 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2861 -mcmodel=large with -mindirect-branch=thunk,
2862 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2863 -mfunction-return=thunk-extern.
2864 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2865 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2866 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2868 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2870 * config/i386/i386.c (print_reg): Print the name of the full
2871 integer register without '%'.
2872 (ix86_print_operand): Handle 'V'.
2873 * doc/extend.texi: Document 'V' modifier.
2875 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2877 * config/i386/constraints.md (Bs): Disallow memory operand for
2878 -mindirect-branch-register.
2880 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2881 (GOT_memory_operand): Likewise.
2882 (call_insn_operand): Likewise.
2883 (sibcall_insn_operand): Likewise.
2884 (GOT32_symbol_operand): Likewise.
2885 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2886 for -mindirect-branch-register.
2887 (tablejump): Likewise.
2888 (*sibcall_memory): Likewise.
2889 (*sibcall_value_memory): Likewise.
2890 Disallow peepholes of indirect call and jump via memory for
2891 -mindirect-branch-register.
2892 (*call_pop): Replace m with Bw.
2893 (*call_value_pop): Likewise.
2894 (*sibcall_pop_memory): Replace m with Bs.
2895 * config/i386/i386.opt (mindirect-branch-register): New option.
2896 * doc/invoke.texi: Document -mindirect-branch-register option.
2898 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2900 * config/i386/i386-protos.h (ix86_output_function_return): New.
2901 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2902 set function_return_type.
2903 (indirect_thunk_name): Add ret_p to indicate thunk for function
2905 (output_indirect_thunk_function): Pass false to
2906 indirect_thunk_name.
2907 (ix86_output_indirect_branch_via_reg): Likewise.
2908 (ix86_output_indirect_branch_via_push): Likewise.
2909 (output_indirect_thunk_function): Create alias for function
2910 return thunk if regno < 0.
2911 (ix86_output_function_return): New function.
2912 (ix86_handle_fndecl_attribute): Handle function_return.
2913 (ix86_attribute_table): Add function_return.
2914 * config/i386/i386.h (machine_function): Add
2915 function_return_type.
2916 * config/i386/i386.md (simple_return_internal): Use
2917 ix86_output_function_return.
2918 (simple_return_internal_long): Likewise.
2919 * config/i386/i386.opt (mfunction-return=): New option.
2920 (indirect_branch): Mention -mfunction-return=.
2921 * doc/extend.texi: Document function_return function attribute.
2922 * doc/invoke.texi: Document -mfunction-return= option.
2924 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2926 * config/i386/i386-opts.h (indirect_branch): New.
2927 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2928 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2929 with local indirect jump when converting indirect call and jump.
2930 (ix86_set_indirect_branch_type): New.
2931 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2932 (indirectlabelno): New.
2933 (indirect_thunk_needed): Likewise.
2934 (indirect_thunk_bnd_needed): Likewise.
2935 (indirect_thunks_used): Likewise.
2936 (indirect_thunks_bnd_used): Likewise.
2937 (INDIRECT_LABEL): Likewise.
2938 (indirect_thunk_name): Likewise.
2939 (output_indirect_thunk): Likewise.
2940 (output_indirect_thunk_function): Likewise.
2941 (ix86_output_indirect_branch_via_reg): Likewise.
2942 (ix86_output_indirect_branch_via_push): Likewise.
2943 (ix86_output_indirect_branch): Likewise.
2944 (ix86_output_indirect_jmp): Likewise.
2945 (ix86_code_end): Call output_indirect_thunk_function if needed.
2946 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2948 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2949 (ix86_attribute_table): Add indirect_branch.
2950 * config/i386/i386.h (machine_function): Add indirect_branch_type
2951 and has_local_indirect_jump.
2952 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2954 (tablejump): Likewise.
2955 (*indirect_jump): Use ix86_output_indirect_jmp.
2956 (*tablejump_1): Likewise.
2957 (simple_return_indirect_internal): Likewise.
2958 * config/i386/i386.opt (mindirect-branch=): New option.
2959 (indirect_branch): New.
2962 (thunk-inline): Likewise.
2963 (thunk-extern): Likewise.
2964 * doc/extend.texi: Document indirect_branch function attribute.
2965 * doc/invoke.texi: Document -mindirect-branch= option.
2967 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2970 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2972 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2974 * ipa-inline.c (want_inline_small_function_p): Return false if
2975 inlining has already failed with CIF_FINAL_ERROR.
2976 (update_caller_keys): Call want_inline_small_function_p before
2978 (update_callee_keys): Likewise.
2980 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2982 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2984 (rs6000_quadword_masked_address_p): Likewise.
2985 (quad_aligned_load_p): Likewise.
2986 (quad_aligned_store_p): Likewise.
2987 (const_load_sequence_p): Add comment to describe the outer-most loop.
2988 (mimic_memory_attributes_and_flags): New function.
2989 (rs6000_gen_stvx): Likewise.
2990 (replace_swapped_aligned_store): Likewise.
2991 (rs6000_gen_lvx): Likewise.
2992 (replace_swapped_aligned_load): Likewise.
2993 (replace_swapped_load_constant): Capitalize argument name in
2994 comment describing this function.
2995 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2996 and stores that access quad-word aligned addresses and replace
2997 with stvx or lvx instructions when appropriate.
2998 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2999 New function prototype.
3000 (rs6000_quadword_masked_address_p): Likewise.
3001 (rs6000_gen_lvx): Likewise.
3002 (rs6000_gen_stvx): Likewise.
3003 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
3004 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3005 when memory address is aligned.
3006 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3007 this split to select lvx instruction when memory address is aligned.
3008 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3009 instruction when memory address is aligned.
3010 (*vsx_le_perm_load_v16qi): Likewise.
3011 (four unnamed splitters): Modify to select the stvx instruction
3012 when memory is aligned.
3014 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3016 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3017 which appears in the queue multiple times.
3019 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3020 Alan Hayward <alan.hayward@arm.com>
3021 David Sherwood <david.sherwood@arm.com>
3023 * tree-vectorizer.h (vec_lower_bound): New structure.
3024 (_loop_vec_info): Add check_nonzero and lower_bounds.
3025 (LOOP_VINFO_CHECK_NONZERO): New macro.
3026 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3027 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3028 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3029 fields. Make seg_len the distance travelled, not including the
3031 (dr_direction_indicator): Declare.
3032 (dr_zero_step_indicator): Likewise.
3033 (dr_known_forward_stride_p): Likewise.
3034 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3036 (runtime_alias_check_p): Allow runtime alias checks with
3038 (operator ==): Compare access_size and align.
3039 (prune_runtime_alias_test_list): Rework for new distinction between
3040 the access_size and seg_len.
3041 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3043 (get_segment_min_max): New function.
3044 (create_intersect_range_checks): Use it.
3045 (dr_step_indicator): New function.
3046 (dr_direction_indicator): Likewise.
3047 (dr_zero_step_indicator): Likewise.
3048 (dr_known_forward_stride_p): Likewise.
3049 * tree-loop-distribution.c (data_ref_segment_size): Return
3050 DR_STEP * (niters - 1).
3051 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3053 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3054 (vect_preserves_scalar_order_p): New function, split out from...
3055 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3056 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3057 (vect_vfa_access_size): New function.
3058 (vect_vfa_align): Likewise.
3059 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3060 (dump_lower_bound): New function.
3061 (vect_check_lower_bound): Likewise.
3062 (vect_small_gap_p): Likewise.
3063 (vectorizable_with_step_bound_p): Likewise.
3064 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3065 depencies if the vectorization factor is 1. Convert the checks
3066 for nonzero steps into checks on the bounds of DR_STEP. Try using
3067 a bunds check for variable steps if the minimum required step is
3068 relatively small. Update calls to the dr_with_seg_len
3069 constructor and to vect_compile_time_alias.
3070 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3072 (vect_loop_versioning): Call it.
3073 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3075 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3077 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3078 Alan Hayward <alan.hayward@arm.com>
3079 David Sherwood <david.sherwood@arm.com>
3081 * doc/sourcebuild.texi (vect_scatter_store): Document.
3082 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3084 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3086 * genopinit.c (main): Add supports_vec_scatter_store and
3087 supports_vec_scatter_store_cached to target_optabs.
3088 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3089 IFN_MASK_SCATTER_STORE.
3090 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3092 * internal-fn.h (internal_store_fn_p): Declare.
3093 (internal_fn_stored_value_index): Likewise.
3094 * internal-fn.c (scatter_store_direct): New macro.
3095 (expand_scatter_store_optab_fn): New function.
3096 (direct_scatter_store_optab_supported_p): New macro.
3097 (internal_store_fn_p): New function.
3098 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3099 IFN_MASK_SCATTER_STORE.
3100 (internal_fn_mask_index): Likewise.
3101 (internal_fn_stored_value_index): New function.
3102 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3104 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3105 * optabs-query.c (supports_vec_scatter_store_p): New function.
3106 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3107 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3108 true for scatter stores.
3109 (vect_gather_scatter_fn_p): Handle scatter stores too.
3110 (vect_check_gather_scatter): Consider using scatter stores if
3111 supports_vec_scatter_store_p.
3112 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3114 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3115 internal_fn_stored_value_index.
3116 (check_load_store_masking): Handle scatter stores too.
3117 (vect_get_store_rhs): Make public.
3118 (vectorizable_call): Use internal_store_fn_p.
3119 (vectorizable_store): Handle scatter store internal functions.
3120 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3121 when deciding whether the end of the group has been reached.
3122 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3123 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3124 (mask_scatter_store<mode>): New insns.
3126 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3127 Alan Hayward <alan.hayward@arm.com>
3128 David Sherwood <david.sherwood@arm.com>
3130 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3131 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3132 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3134 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3135 Use vect_truncate_gather_scatter_offset if we can't treat the
3136 operation as a normal gather load or scatter store.
3137 (get_group_load_store_type): Take the gather_scatter_info
3138 as argument. Try using a gather load or scatter store for
3139 single-element groups.
3140 (get_load_store_type): Update calls to get_group_load_store_type
3141 and vect_use_strided_gather_scatters_p.
3143 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3144 Alan Hayward <alan.hayward@arm.com>
3145 David Sherwood <david.sherwood@arm.com>
3147 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3148 optional tree argument.
3149 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3151 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3152 but continue to use the current value as a fallback.
3153 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3154 to compare the updates.
3155 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3156 (get_load_store_type): Use it when handling a strided access.
3157 (vect_get_strided_load_store_ops): New function.
3158 (vect_get_data_ptr_increment): Likewise.
3159 (vectorizable_load): Handle strided gather loads. Always pass
3160 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3162 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3163 Alan Hayward <alan.hayward@arm.com>
3164 David Sherwood <david.sherwood@arm.com>
3166 * doc/md.texi (gather_load@var{m}): Document.
3167 (mask_gather_load@var{m}): Likewise.
3168 * genopinit.c (main): Add supports_vec_gather_load and
3169 supports_vec_gather_load_cached to target_optabs.
3170 * optabs-tree.c (init_tree_optimization_optabs): Use
3171 ggc_cleared_alloc to allocate target_optabs.
3172 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3173 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3175 * internal-fn.h (internal_load_fn_p): Declare.
3176 (internal_gather_scatter_fn_p): Likewise.
3177 (internal_fn_mask_index): Likewise.
3178 (internal_gather_scatter_fn_supported_p): Likewise.
3179 * internal-fn.c (gather_load_direct): New macro.
3180 (expand_gather_load_optab_fn): New function.
3181 (direct_gather_load_optab_supported_p): New macro.
3182 (direct_internal_fn_optab): New function.
3183 (internal_load_fn_p): Likewise.
3184 (internal_gather_scatter_fn_p): Likewise.
3185 (internal_fn_mask_index): Likewise.
3186 (internal_gather_scatter_fn_supported_p): Likewise.
3187 * optabs-query.c (supports_at_least_one_mode_p): New function.
3188 (supports_vec_gather_load_p): Likewise.
3189 * optabs-query.h (supports_vec_gather_load_p): Declare.
3190 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3191 and memory_type field.
3192 (NUM_PATTERNS): Bump to 15.
3193 * tree-vect-data-refs.c: Include internal-fn.h.
3194 (vect_gather_scatter_fn_p): New function.
3195 (vect_describe_gather_scatter_call): Likewise.
3196 (vect_check_gather_scatter): Try using internal functions for
3197 gather loads. Recognize existing calls to a gather load function.
3198 (vect_analyze_data_refs): Consider using gather loads if
3199 supports_vec_gather_load_p.
3200 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3201 (vect_get_gather_scatter_offset_type): Likewise.
3202 (vect_convert_mask_for_vectype): Likewise.
3203 (vect_add_conversion_to_patterm): Likewise.
3204 (vect_try_gather_scatter_pattern): Likewise.
3205 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3206 (vect_vect_recog_func_ptrs): Add it.
3207 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3208 internal_fn_mask_index and internal_gather_scatter_fn_p.
3209 (check_load_store_masking): Take the gather_scatter_info as an
3210 argument and handle gather loads.
3211 (vect_get_gather_scatter_ops): New function.
3212 (vectorizable_call): Check internal_load_fn_p.
3213 (vectorizable_load): Likewise. Handle gather load internal
3215 (vectorizable_store): Update call to check_load_store_masking.
3216 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3217 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3218 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3219 (aarch64_gather_scale_operand_d): New predicates.
3220 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3221 (mask_gather_load<mode>): New insns.
3223 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3224 Alan Hayward <alan.hayward@arm.com>
3225 David Sherwood <david.sherwood@arm.com>
3227 * optabs.def (fold_left_plus_optab): New optab.
3228 * doc/md.texi (fold_left_plus_@var{m}): Document.
3229 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3230 * internal-fn.c (fold_left_direct): Define.
3231 (expand_fold_left_optab_fn): Likewise.
3232 (direct_fold_left_optab_supported_p): Likewise.
3233 * fold-const-call.c (fold_const_fold_left): New function.
3234 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3235 * tree-parloops.c (valid_reduction_p): New function.
3236 (gather_scalar_reductions): Use it.
3237 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3238 (vect_finish_replace_stmt): Declare.
3239 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3240 (needs_fold_left_reduction_p): New function, split out from...
3241 (vect_is_simple_reduction): ...here. Accept reductions that
3242 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3243 (vect_force_simple_reduction): Also store the reduction type in
3244 the assignment's STMT_VINFO_REDUC_TYPE.
3245 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3246 (merge_with_identity): New function.
3247 (vect_expand_fold_left): Likewise.
3248 (vectorize_fold_left_reduction): Likewise.
3249 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3250 scalar phi in place for it. Check for target support and reject
3251 cases that would reassociate the operation. Defer the transform
3252 phase to vectorize_fold_left_reduction.
3253 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3254 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3255 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3257 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3259 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3260 call to ifc_temp_var.
3262 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3263 Alan Hayward <alan.hayward@arm.com>
3264 David Sherwood <david.sherwood@arm.com>
3266 * target.def (legitimize_address_displacement): Take the original
3267 offset as a poly_int.
3268 * targhooks.h (default_legitimize_address_displacement): Update
3270 * targhooks.c (default_legitimize_address_displacement): Likewise.
3271 * doc/tm.texi: Regenerate.
3272 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3273 as an argument, moving assert of ad->disp == ad->disp_term to...
3274 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3275 Try calling targetm.legitimize_address_displacement before expanding
3276 the address rather than afterwards, and adjust for the new interface.
3277 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3278 Match the new hook interface. Handle SVE addresses.
3279 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3282 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3284 * Makefile.in (OBJS): Add early-remat.o.
3285 * target.def (select_early_remat_modes): New hook.
3286 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3287 * doc/tm.texi: Regenerate.
3288 * targhooks.h (default_select_early_remat_modes): Declare.
3289 * targhooks.c (default_select_early_remat_modes): New function.
3290 * timevar.def (TV_EARLY_REMAT): New timevar.
3291 * passes.def (pass_early_remat): New pass.
3292 * tree-pass.h (make_pass_early_remat): Declare.
3293 * early-remat.c: New file.
3294 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3296 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3298 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3299 Alan Hayward <alan.hayward@arm.com>
3300 David Sherwood <david.sherwood@arm.com>
3302 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3303 vfm1 with a bound_epilog parameter.
3304 (vect_do_peeling): Update calls accordingly, and move the prologue
3305 call earlier in the function. Treat the base bound_epilog as 0 for
3306 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3307 this base when peeling for gaps.
3308 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3309 with fully-masked loops.
3310 (vect_estimate_min_profitable_iters): Handle the single peeled
3311 iteration in that case.
3313 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3314 Alan Hayward <alan.hayward@arm.com>
3315 David Sherwood <david.sherwood@arm.com>
3317 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3318 single-element interleaving even if the size is not a power of 2.
3319 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3320 accesses for single-element interleaving if the group size is
3323 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3324 Alan Hayward <alan.hayward@arm.com>
3325 David Sherwood <david.sherwood@arm.com>
3327 * doc/md.texi (fold_extract_last_@var{m}): Document.
3328 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3329 * optabs.def (fold_extract_last_optab): New optab.
3330 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3331 * internal-fn.c (fold_extract_direct): New macro.
3332 (expand_fold_extract_optab_fn): Likewise.
3333 (direct_fold_extract_optab_supported_p): Likewise.
3334 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3335 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3336 EXTRACT_LAST_REDUCTION.
3337 (get_initial_def_for_reduction): Do not create an initial vector
3338 for EXTRACT_LAST_REDUCTION reductions.
3339 (vectorizable_reduction): Leave the scalar phi in place for
3340 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3341 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3342 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3343 transform phase to vectorizable_condition.
3344 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3346 (vect_finish_stmt_generation): ...here.
3347 (vect_finish_replace_stmt): New function.
3348 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3349 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3351 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3353 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3354 Alan Hayward <alan.hayward@arm.com>
3355 David Sherwood <david.sherwood@arm.com>
3357 * doc/md.texi (extract_last_@var{m}): Document.
3358 * optabs.def (extract_last_optab): New optab.
3359 * internal-fn.def (EXTRACT_LAST): New internal function.
3360 * internal-fn.c (cond_unary_direct): New macro.
3361 (expand_cond_unary_optab_fn): Likewise.
3362 (direct_cond_unary_optab_supported_p): Likewise.
3363 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3364 loops using EXTRACT_LAST.
3365 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3366 (extract_last_<mode>): ...this optab.
3367 (vec_extract<mode><Vel>): Update accordingly.
3369 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3370 Alan Hayward <alan.hayward@arm.com>
3371 David Sherwood <david.sherwood@arm.com>
3373 * target.def (empty_mask_is_expensive): New hook.
3374 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3375 * doc/tm.texi: Regenerate.
3376 * targhooks.h (default_empty_mask_is_expensive): Declare.
3377 * targhooks.c (default_empty_mask_is_expensive): New function.
3378 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3379 if the target says that empty masks are expensive.
3380 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3382 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3384 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3385 Alan Hayward <alan.hayward@arm.com>
3386 David Sherwood <david.sherwood@arm.com>
3388 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3389 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3390 (vect_use_loop_mask_for_alignment_p): New function.
3391 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3392 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3393 niters_skip argument. Make sure that the first niters_skip elements
3394 of the first iteration are inactive.
3395 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3396 Update call to vect_set_loop_masks_directly.
3397 (get_misalign_in_elems): New function, split out from...
3398 (vect_gen_prolog_loop_niters): ...here.
3399 (vect_update_init_of_dr): Take a code argument that specifies whether
3400 the adjustment should be added or subtracted.
3401 (vect_update_init_of_drs): Likewise.
3402 (vect_prepare_for_masked_peels): New function.
3403 (vect_do_peeling): Skip prologue peeling if we're using a mask
3404 instead. Update call to vect_update_inits_of_drs.
3405 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3407 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3408 alignment. Do not include the number of peeled iterations in
3409 the minimum threshold in that case.
3410 (vectorizable_induction): Adjust the start value down by
3411 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3412 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3413 Take the number of skipped iterations into account when calculating
3415 * tree-vect-stmts.c (vect_gen_while_not): New function.
3417 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3418 Alan Hayward <alan.hayward@arm.com>
3419 David Sherwood <david.sherwood@arm.com>
3421 * doc/sourcebuild.texi (vect_fully_masked): Document.
3422 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3424 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3426 (vect_analyze_loop_2): ...here. Don't check the vectorization
3427 factor against the number of loop iterations if the loop is
3430 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3431 Alan Hayward <alan.hayward@arm.com>
3432 David Sherwood <david.sherwood@arm.com>
3434 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3435 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3436 (dump_groups): Update accordingly.
3437 (iv_use::mem_type): New member variable.
3438 (address_p): New function.
3439 (record_use): Add a mem_type argument and initialize the new
3441 (record_group_use): Add a mem_type argument. Use address_p.
3442 Remove obsolete null checks of base_object. Update call to record_use.
3443 (find_interesting_uses_op): Update call to record_group_use.
3444 (find_interesting_uses_cond): Likewise.
3445 (find_interesting_uses_address): Likewise.
3446 (get_mem_type_for_internal_fn): New function.
3447 (find_address_like_use): Likewise.
3448 (find_interesting_uses_stmt): Try find_address_like_use before
3449 calling find_interesting_uses_op.
3450 (addr_offset_valid_p): Use the iv mem_type field as the type
3451 of the addressed memory.
3452 (add_autoinc_candidates): Likewise.
3453 (get_address_cost): Likewise.
3454 (split_small_address_groups_p): Use address_p.
3455 (split_address_groups): Likewise.
3456 (add_iv_candidate_for_use): Likewise.
3457 (autoinc_possible_for_pair): Likewise.
3458 (rewrite_groups): Likewise.
3459 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3460 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3461 (get_alias_ptr_type_for_ptr_address): New function.
3462 (rewrite_use_address): Rewrite address uses in calls that were
3463 identified by find_address_like_use.
3465 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3466 Alan Hayward <alan.hayward@arm.com>
3467 David Sherwood <david.sherwood@arm.com>
3469 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3471 * gimple-expr.h (is_gimple_addressable: Likewise.
3472 * gimple-expr.c (is_gimple_address): Likewise.
3473 * internal-fn.c (expand_call_mem_ref): New function.
3474 (expand_mask_load_optab_fn): Use it.
3475 (expand_mask_store_optab_fn): Likewise.
3477 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3478 Alan Hayward <alan.hayward@arm.com>
3479 David Sherwood <david.sherwood@arm.com>
3481 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3482 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3483 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3484 (cond_umax@var{mode}): Document.
3485 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3486 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3487 (cond_umin_optab, cond_umax_optab): New optabs.
3488 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3489 (COND_IOR, COND_XOR): New internal functions.
3490 * internal-fn.h (get_conditional_internal_fn): Declare.
3491 * internal-fn.c (cond_binary_direct): New macro.
3492 (expand_cond_binary_optab_fn): Likewise.
3493 (direct_cond_binary_optab_supported_p): Likewise.
3494 (get_conditional_internal_fn): New function.
3495 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3496 Cope with reduction statements that are vectorized as calls rather
3498 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3499 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3500 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3501 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3502 (UNSPEC_COND_EOR): New unspecs.
3503 (optab): Add mappings for them.
3504 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3505 (sve_int_op, sve_fp_op): New int attributes.
3507 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3508 Alan Hayward <alan.hayward@arm.com>
3509 David Sherwood <david.sherwood@arm.com>
3511 * optabs.def (while_ult_optab): New optab.
3512 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3513 * internal-fn.def (WHILE_ULT): New internal function.
3514 * internal-fn.h (direct_internal_fn_supported_p): New override
3515 that takes two types as argument.
3516 * internal-fn.c (while_direct): New macro.
3517 (expand_while_optab_fn): New function.
3518 (convert_optab_supported_p): Likewise.
3519 (direct_while_optab_supported_p): New macro.
3520 * wide-int.h (wi::udiv_ceil): New function.
3521 * tree-vectorizer.h (rgroup_masks): New structure.
3522 (vec_loop_masks): New typedef.
3523 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3525 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3526 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3527 (vect_max_vf): New function.
3528 (slpeel_make_loop_iterate_ntimes): Delete.
3529 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3530 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3531 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3532 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3533 internal-fn.h, stor-layout.h and optabs-query.h.
3534 (vect_set_loop_mask): New function.
3535 (add_preheader_seq): Likewise.
3536 (add_header_seq): Likewise.
3537 (interleave_supported_p): Likewise.
3538 (vect_maybe_permute_loop_masks): Likewise.
3539 (vect_set_loop_masks_directly): Likewise.
3540 (vect_set_loop_condition_masked): Likewise.
3541 (vect_set_loop_condition_unmasked): New function, split out from
3542 slpeel_make_loop_iterate_ntimes.
3543 (slpeel_make_loop_iterate_ntimes): Rename to..
3544 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3545 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3546 (vect_do_peeling): Update call accordingly.
3547 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3549 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3550 mask_compare_type, can_fully_mask_p and fully_masked_p.
3551 (release_vec_loop_masks): New function.
3552 (_loop_vec_info): Use it to free the loop masks.
3553 (can_produce_all_loop_masks_p): New function.
3554 (vect_get_max_nscalars_per_iter): Likewise.
3555 (vect_verify_full_masking): Likewise.
3556 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3557 retries, and free the mask rgroups before retrying. Check loop-wide
3558 reasons for disallowing fully-masked loops. Make the final decision
3559 about whether use a fully-masked loop or not.
3560 (vect_estimate_min_profitable_iters): Do not assume that peeling
3561 for the number of iterations will be needed for fully-masked loops.
3562 (vectorizable_reduction): Disable fully-masked loops.
3563 (vectorizable_live_operation): Likewise.
3564 (vect_halve_mask_nunits): New function.
3565 (vect_double_mask_nunits): Likewise.
3566 (vect_record_loop_mask): Likewise.
3567 (vect_get_loop_mask): Likewise.
3568 (vect_transform_loop): Handle the case in which the final loop
3569 iteration might handle a partial vector. Call vect_set_loop_condition
3570 instead of slpeel_make_loop_iterate_ntimes.
3571 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3572 (check_load_store_masking): New function.
3573 (prepare_load_store_mask): Likewise.
3574 (vectorizable_store): Handle fully-masked loops.
3575 (vectorizable_load): Likewise.
3576 (supportable_widening_operation): Use vect_halve_mask_nunits for
3578 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3579 (vect_gen_while): New function.
3580 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3581 (aarch64_uqdec<mode>): New insn.
3583 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3584 Alan Hayward <alan.hayward@arm.com>
3585 David Sherwood <david.sherwood@arm.com>
3587 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3588 (reduc_xor_scal_optab): New optabs.
3589 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3590 (reduc_xor_scal_@var{m}): Document.
3591 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3592 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3594 * fold-const-call.c (fold_const_call): Handle them.
3595 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3596 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3597 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3598 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3599 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3600 (UNSPEC_XORV): New unspecs.
3601 (optab): Add entries for them.
3602 (BITWISEV): New int iterator.
3603 (bit_reduc_op): New int attributes.
3605 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3606 Alan Hayward <alan.hayward@arm.com>
3607 David Sherwood <david.sherwood@arm.com>
3609 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3610 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3611 * optabs.def (vec_shl_insert_optab): New optab.
3612 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3613 (duplicate_and_interleave): Likewise.
3614 * tree-vect-loop.c: Include internal-fn.h.
3615 (neutral_op_for_slp_reduction): New function, split out from
3616 get_initial_defs_for_reduction.
3617 (get_initial_def_for_reduction): Handle option 2 for variable-length
3618 vectors by loading the neutral value into a vector and then shifting
3619 the initial value into element 0.
3620 (get_initial_defs_for_reduction): Replace the code argument with
3621 the neutral value calculated by neutral_op_for_slp_reduction.
3622 Use gimple_build_vector for constant-length vectors.
3623 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3624 but the first group_size elements have a neutral value.
3625 Use duplicate_and_interleave otherwise.
3626 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3627 Update call to get_initial_defs_for_reduction. Handle SLP
3628 reductions for variable-length vectors by creating one vector
3629 result for each scalar result, with the elements associated
3630 with other scalar results stubbed out with the neutral value.
3631 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3632 Require IFN_VEC_SHL_INSERT for double reductions on
3633 variable-length vectors, or SLP reductions that have
3634 a neutral value. Require can_duplicate_and_interleave_p
3635 support for variable-length unchained SLP reductions if there
3636 is no neutral value, such as for MIN/MAX reductions. Also require
3637 the number of vector elements to be a multiple of the number of
3638 SLP statements when doing variable-length unchained SLP reductions.
3639 Update call to vect_create_epilog_for_reduction.
3640 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3641 and remove initial values.
3642 (duplicate_and_interleave): Make public.
3643 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3644 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3646 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3647 Alan Hayward <alan.hayward@arm.com>
3648 David Sherwood <david.sherwood@arm.com>
3650 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3651 (can_duplicate_and_interleave_p): New function.
3652 (vect_get_and_check_slp_defs): Take the vector of statements
3653 rather than just the current one. Remove excess parentheses.
3654 Restriction rejectinon of vect_constant_def and vect_external_def
3655 for variable-length vectors to boolean types, or types for which
3656 can_duplicate_and_interleave_p is false.
3657 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3658 (duplicate_and_interleave): New function.
3659 (vect_get_constant_vectors): Use gimple_build_vector for
3660 constant-length vectors and suitable variable-length constant
3661 vectors. Use duplicate_and_interleave for other variable-length
3662 vectors. Don't defer the update when inserting new statements.
3664 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3665 Alan Hayward <alan.hayward@arm.com>
3666 David Sherwood <david.sherwood@arm.com>
3668 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3669 min_profitable_iters doesn't go negative.
3671 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3672 Alan Hayward <alan.hayward@arm.com>
3673 David Sherwood <david.sherwood@arm.com>
3675 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3676 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3677 * optabs.def (vec_mask_load_lanes_optab): New optab.
3678 (vec_mask_store_lanes_optab): Likewise.
3679 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3680 (MASK_STORE_LANES): Likewise.
3681 * internal-fn.c (mask_load_lanes_direct): New macro.
3682 (mask_store_lanes_direct): Likewise.
3683 (expand_mask_load_optab_fn): Handle masked operations.
3684 (expand_mask_load_lanes_optab_fn): New macro.
3685 (expand_mask_store_optab_fn): Handle masked operations.
3686 (expand_mask_store_lanes_optab_fn): New macro.
3687 (direct_mask_load_lanes_optab_supported_p): Likewise.
3688 (direct_mask_store_lanes_optab_supported_p): Likewise.
3689 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3691 (vect_load_lanes_supported): Likewise.
3692 * tree-vect-data-refs.c (strip_conversion): New function.
3693 (can_group_stmts_p): Likewise.
3694 (vect_analyze_data_ref_accesses): Use it instead of checking
3695 for a pair of assignments.
3696 (vect_store_lanes_supported): Take a masked_p parameter.
3697 (vect_load_lanes_supported): Likewise.
3698 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3699 vect_store_lanes_supported and vect_load_lanes_supported.
3700 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3701 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3702 parameter. Don't allow gaps for masked accesses.
3703 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3704 and vect_load_lanes_supported.
3705 (get_load_store_type): Take a masked_p parameter and update
3706 call to get_group_load_store_type.
3707 (vectorizable_store): Update call to get_load_store_type.
3708 Handle IFN_MASK_STORE_LANES.
3709 (vectorizable_load): Update call to get_load_store_type.
3710 Handle IFN_MASK_LOAD_LANES.
3712 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3713 Alan Hayward <alan.hayward@arm.com>
3714 David Sherwood <david.sherwood@arm.com>
3716 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3718 * config/aarch64/aarch64-protos.h
3719 (aarch64_sve_struct_memory_operand_p): Declare.
3720 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3721 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3722 (VPRED, vpred): Handle SVE structure modes.
3723 * config/aarch64/constraints.md (Utx): New constraint.
3724 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3725 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3726 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3727 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3728 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3729 structure modes. Split into pieces after RA.
3730 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3731 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3733 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3734 SVE structure modes.
3735 (aarch64_classify_address): Likewise.
3736 (sizetochar): Move earlier in file.
3737 (aarch64_print_operand): Handle SVE register lists.
3738 (aarch64_array_mode): New function.
3739 (aarch64_sve_struct_memory_operand_p): Likewise.
3740 (TARGET_ARRAY_MODE): Redefine.
3742 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3743 Alan Hayward <alan.hayward@arm.com>
3744 David Sherwood <david.sherwood@arm.com>
3746 * target.def (array_mode): New target hook.
3747 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3748 * doc/tm.texi: Regenerate.
3749 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3750 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3751 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3753 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3756 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3757 Alan Hayward <alan.hayward@arm.com>
3758 David Sherwood <david.sherwood@arm.com>
3760 * fold-const.c (fold_binary_loc): Check the argument types
3761 rather than the result type when testing for a vector operation.
3763 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3765 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3766 * doc/tm.texi: Regenerate.
3768 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3769 Alan Hayward <alan.hayward@arm.com>
3770 David Sherwood <david.sherwood@arm.com>
3772 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3773 (sve): Document new AArch64 extension.
3774 * doc/md.texi (w): Extend the description of the AArch64
3775 constraint to include SVE vectors.
3776 (Upl, Upa): Document new AArch64 predicate constraints.
3777 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3779 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3780 (msve-vector-bits=): New option.
3781 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3782 SVE when these are disabled.
3783 (sve): New extension.
3784 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3785 modes. Adjust their number of units based on aarch64_sve_vg.
3786 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3787 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3788 aarch64_addr_query_type.
3789 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3790 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3791 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3792 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3793 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3794 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3795 (aarch64_simd_imm_zero_p): Delete.
3796 (aarch64_check_zero_based_sve_index_immediate): Declare.
3797 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3798 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3799 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3800 (aarch64_sve_float_mul_immediate_p): Likewise.
3801 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3803 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3804 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3805 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3806 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3807 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3808 (aarch64_regmode_natural_size): Likewise.
3809 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3810 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3812 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3813 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3814 for VG and the SVE predicate registers.
3815 (V_ALIASES): Add a "z"-prefixed alias.
3816 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3817 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3818 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3819 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3820 (REG_CLASS_NAMES): Add entries for them.
3821 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3822 and the predicate registers.
3823 (aarch64_sve_vg): Declare.
3824 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3825 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3826 (REGMODE_NATURAL_SIZE): Define.
3827 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3829 * config/aarch64/aarch64.c: Include cfgrtl.h.
3830 (simd_immediate_info): Add a constructor for series vectors,
3831 and an associated step field.
3832 (aarch64_sve_vg): New variable.
3833 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3834 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3835 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3836 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3837 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3838 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3839 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3840 (aarch64_get_mask_mode): New functions.
3841 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3842 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3843 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3844 predicate modes and predicate registers. Explicitly restrict
3845 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3846 to store a vector mode if it is recognized by
3847 aarch64_classify_vector_mode.
3848 (aarch64_regmode_natural_size): New function.
3849 (aarch64_hard_regno_caller_save_mode): Return the original mode
3851 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3852 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3853 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3854 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3856 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3857 does not overlap dest if the function is frame-related. Handle
3859 (aarch64_split_add_offset): New function.
3860 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3861 them aarch64_add_offset.
3862 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3863 and update call to aarch64_sub_sp.
3864 (aarch64_add_cfa_expression): New function.
3865 (aarch64_expand_prologue): Pass extra temporary registers to the
3866 functions above. Handle the case in which we need to emit new
3867 DW_CFA_expressions for registers that were originally saved
3868 relative to the stack pointer, but now have to be expressed
3869 relative to the frame pointer.
3870 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3872 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3873 IP0 and IP1 values for SVE frames.
3874 (aarch64_expand_vec_series): New function.
3875 (aarch64_expand_sve_widened_duplicate): Likewise.
3876 (aarch64_expand_sve_const_vector): Likewise.
3877 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3878 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3879 into the register, rather than emitting a SET directly.
3880 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3881 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3882 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3883 (offset_9bit_signed_scaled_p): New functions.
3884 (aarch64_replicate_bitmask_imm): New function.
3885 (aarch64_bitmask_imm): Use it.
3886 (aarch64_cannot_force_const_mem): Reject expressions involving
3887 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3888 (aarch64_classify_index): Handle SVE indices, by requiring
3889 a plain register index with a scale that matches the element size.
3890 (aarch64_classify_address): Handle SVE addresses. Assert that
3891 the mode of the address is VOIDmode or an integer mode.
3892 Update call to aarch64_classify_symbol.
3893 (aarch64_classify_symbolic_expression): Update call to
3894 aarch64_classify_symbol.
3895 (aarch64_const_vec_all_in_range_p): New function.
3896 (aarch64_print_vector_float_operand): Likewise.
3897 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3898 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3899 and the FP immediates 1.0 and 0.5.
3900 (aarch64_print_address_internal): Handle SVE addresses.
3901 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3902 (aarch64_regno_regclass): Handle predicate registers.
3903 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3905 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3906 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3907 (aarch64_convert_sve_vector_bits): New function.
3908 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3909 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3911 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3912 Handle SVE vector and predicate modes. Accept VL-based constants
3913 that need only one temporary register, and VL offsets that require
3914 no temporary registers.
3915 (aarch64_conditional_register_usage): Mark the predicate registers
3916 as fixed if SVE isn't available.
3917 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3918 Return true for SVE vector and predicate modes.
3919 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3920 rather than an unsigned int. Handle SVE modes.
3921 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3923 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3925 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3926 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3927 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3928 (aarch64_sve_float_mul_immediate_p): New functions.
3929 (aarch64_sve_valid_immediate): New function.
3930 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3931 Explicitly reject structure modes. Check for INDEX constants.
3932 Handle PTRUE and PFALSE constants.
3933 (aarch64_check_zero_based_sve_index_immediate): New function.
3934 (aarch64_simd_imm_zero_p): Delete.
3935 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3936 vector modes. Accept constants in the range of CNT[BHWD].
3937 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3938 ask for an Advanced SIMD mode.
3939 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3940 (aarch64_simd_vector_alignment): Handle SVE predicates.
3941 (aarch64_vectorize_preferred_vector_alignment): New function.
3942 (aarch64_simd_vector_alignment_reachable): Use it instead of
3944 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3945 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3947 (MAX_VECT_LEN): Delete.
3948 (expand_vec_perm_d): Add a vec_flags field.
3949 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3950 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3951 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3953 (aarch64_evpc_rev): Rename to...
3954 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3955 (aarch64_evpc_rev_global): New function.
3956 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3957 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3959 (aarch64_evpc_sve_tbl): New function.
3960 (aarch64_expand_vec_perm_const_1): Update after rename of
3961 aarch64_evpc_rev. Handle SVE permutes too, trying
3962 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3963 than aarch64_evpc_tbl.
3964 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3965 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3966 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3967 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3968 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3969 (aarch64_expand_sve_vcond): New functions.
3970 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3971 of aarch64_vector_mode_p.
3972 (aarch64_dwarf_poly_indeterminate_value): New function.
3973 (aarch64_compute_pressure_classes): Likewise.
3974 (aarch64_can_change_mode_class): Likewise.
3975 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3976 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3977 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3978 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3979 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3980 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3981 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3982 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3984 (Dn, Dl, Dr): Accept const as well as const_vector.
3985 (Dz): Likewise. Compare against CONST0_RTX.
3986 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3987 of "vector" where appropriate.
3988 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3989 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3990 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3991 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3992 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3993 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3994 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3995 (v_int_equiv): Extend to SVE modes.
3996 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3998 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3999 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
4000 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
4001 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
4002 (SVE_COND_FP_CMP): New int iterators.
4003 (perm_hilo): Handle the new unpack unspecs.
4004 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4006 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4007 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4008 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4009 (aarch64_equality_operator, aarch64_constant_vector_operand)
4010 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4011 (aarch64_sve_nonimmediate_operand): Likewise.
4012 (aarch64_sve_general_operand): Likewise.
4013 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4014 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4015 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4016 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4017 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4018 (aarch64_sve_float_arith_immediate): Likewise.
4019 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4020 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4021 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4022 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4023 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4024 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4025 (aarch64_sve_float_arith_operand): Likewise.
4026 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4027 (aarch64_sve_float_mul_operand): Likewise.
4028 (aarch64_sve_vec_perm_operand): Likewise.
4029 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4030 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4031 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4032 as well as const_vector.
4033 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4034 in file. Use CONST0_RTX and CONSTM1_RTX.
4035 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4036 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4037 Use aarch64_simd_imm_zero.
4038 * config/aarch64/aarch64-sve.md: New file.
4039 * config/aarch64/aarch64.md: Include it.
4040 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4041 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4042 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4043 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4044 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4045 (sve): New attribute.
4046 (enabled): Disable instructions with the sve attribute unless
4048 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4049 aarch64_expand_mov_immediate.
4050 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4051 CNT[BHSD] immediates.
4052 (movti): Split CONST_POLY_INT moves into two halves.
4053 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4054 Split additions that need a temporary here if the destination
4055 is the stack pointer.
4056 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4057 (*add<mode>3_poly_1): New instruction.
4058 (set_clobber_cc): New expander.
4060 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4062 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4063 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4064 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4065 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4066 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4067 Change innermode from fixed_mode_size to machine_mode.
4068 (simplify_subreg): Update call accordingly. Handle a constant-sized
4069 subreg of a variable-length CONST_VECTOR.
4071 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4072 Alan Hayward <alan.hayward@arm.com>
4073 David Sherwood <david.sherwood@arm.com>
4075 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4076 (add_offset_to_base): New function, split out from...
4077 (create_mem_ref): ...here. When handling a scale other than 1,
4078 check first whether the address is valid without the offset.
4079 Add it into the base if so, leaving the index and scale as-is.
4081 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4084 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4085 fold_for_warn before checking if arg2 is INTEGER_CST.
4087 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4089 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4090 (store_multiple_operation): Delete.
4091 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4092 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4093 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4094 guarded by TARGET_STRING.
4095 (rs6000_output_load_multiple): Delete.
4096 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4097 OPTION_MASK_STRING / TARGET_STRING handling.
4098 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4099 (const rs6000_opt_masks) <"string">: Change mask to 0.
4100 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4101 (MASK_STRING): Delete.
4102 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4104 (load_multiple): Delete.
4111 (store_multiple): Delete.
4118 (movmemsi_8reg): Delete.
4119 (corresponding unnamed define_insn): Delete.
4120 (movmemsi_6reg): Delete.
4121 (corresponding unnamed define_insn): Delete.
4122 (movmemsi_4reg): Delete.
4123 (corresponding unnamed define_insn): Delete.
4124 (movmemsi_2reg): Delete.
4125 (corresponding unnamed define_insn): Delete.
4126 (movmemsi_1reg): Delete.
4127 (corresponding unnamed define_insn): Delete.
4128 * config/rs6000/rs6000.opt (mno-string): New.
4129 (mstring): Replace by deprecation warning stub.
4130 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4132 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4134 * regrename.c (regrename_do_replace): If replacing the same
4135 reg multiple times, try to reuse last created gen_raw_REG.
4138 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4139 main to workaround a bug in GDB.
4141 2018-01-12 Tom de Vries <tom@codesourcery.com>
4144 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4146 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4148 PR rtl-optimization/80481
4149 * ira-color.c (get_cap_member): New function.
4150 (allocnos_conflict_by_live_ranges_p): Use it.
4151 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4152 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4154 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4157 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4158 (*saddl_se_1): Ditto.
4160 (*ssubl_se_1): Ditto.
4162 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4164 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4165 rather than wi::to_widest for DR_INITs.
4166 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4167 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4168 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4170 (vect_analyze_group_access_1): Note that here.
4172 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4174 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4175 polynomial type sizes.
4177 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4179 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4180 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4181 (gimple_add_tmp_var): Likewise.
4183 2018-01-12 Martin Liska <mliska@suse.cz>
4185 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4186 (gimple_alloc_sizes): Likewise.
4187 (dump_gimple_statistics): Use PRIu64 in printf format.
4188 * gimple.h: Change uint64_t to int.
4190 2018-01-12 Martin Liska <mliska@suse.cz>
4192 * tree-core.h: Use uint64_t instead of int.
4193 * tree.c (tree_node_counts): Likewise.
4194 (tree_node_sizes): Likewise.
4195 (dump_tree_statistics): Use PRIu64 in printf format.
4197 2018-01-12 Martin Liska <mliska@suse.cz>
4199 * Makefile.in: As qsort_chk is implemented in vec.c, add
4200 vec.o to linkage of gencfn-macros.
4201 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4202 passing the info to record_node_allocation_statistics.
4203 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4205 * ggc-common.c (struct ggc_usage): Add operator== and use
4206 it in operator< and compare function.
4207 * mem-stats.h (struct mem_usage): Likewise.
4208 * vec.c (struct vec_usage): Remove operator< and compare
4209 function. Can be simply inherited.
4211 2018-01-12 Martin Jambor <mjambor@suse.cz>
4214 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4215 * tree-ssa-math-opts.c: Include domwalk.h.
4216 (convert_mult_to_fma_1): New function.
4217 (fma_transformation_info): New type.
4218 (fma_deferring_state): Likewise.
4219 (cancel_fma_deferring): New function.
4220 (result_of_phi): Likewise.
4221 (last_fma_candidate_feeds_initial_phi): Likewise.
4222 (convert_mult_to_fma): Added deferring logic, split actual
4223 transformation to convert_mult_to_fma_1.
4224 (math_opts_dom_walker): New type.
4225 (math_opts_dom_walker::after_dom_children): New method, body moved
4226 here from pass_optimize_widening_mul::execute, added deferring logic
4228 (pass_optimize_widening_mul::execute): Moved most of code to
4229 math_opts_dom_walker::after_dom_children.
4230 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4231 * config/i386/i386.c (ix86_option_override_internal): Added
4232 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4234 2018-01-12 Richard Biener <rguenther@suse.de>
4237 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4238 inline instance vars.
4240 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4243 * config/rx/rx.c (rx_is_restricted_memory_address):
4246 2018-01-12 Richard Biener <rguenther@suse.de>
4248 PR tree-optimization/80846
4249 * target.def (split_reduction): New target hook.
4250 * targhooks.c (default_split_reduction): New function.
4251 * targhooks.h (default_split_reduction): Declare.
4252 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4253 target requests first reduce vectors by combining low and high
4255 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4256 (get_vectype_for_scalar_type_and_size): Export.
4257 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4258 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4259 * doc/tm.texi: Regenerate.
4260 * config/i386/i386.c (ix86_split_reduction): Implement
4261 TARGET_VECTORIZE_SPLIT_REDUCTION.
4263 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4266 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4267 in PIC mode except for TARGET_VXWORKS_RTP.
4268 * config/sparc/sparc.c: Include cfgrtl.h.
4269 (TARGET_INIT_PIC_REG): Define.
4270 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4271 (sparc_pic_register_p): New predicate.
4272 (sparc_legitimate_address_p): Use it.
4273 (sparc_legitimize_pic_address): Likewise.
4274 (sparc_delegitimize_address): Likewise.
4275 (sparc_mode_dependent_address_p): Likewise.
4276 (gen_load_pcrel_sym): Remove 4th parameter.
4277 (load_got_register): Adjust call to above. Remove obsolete stuff.
4278 (sparc_expand_prologue): Do not call load_got_register here.
4279 (sparc_flat_expand_prologue): Likewise.
4280 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4281 (sparc_use_pseudo_pic_reg): New function.
4282 (sparc_init_pic_reg): Likewise.
4283 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4284 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4286 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4288 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4289 Add item for branch_cost.
4291 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4293 PR rtl-optimization/83565
4294 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4295 not extend the result to a larger mode for rotate operations.
4296 (num_sign_bit_copies1): Likewise.
4298 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4301 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4303 Use values-Xc.o for -pedantic.
4304 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4306 2018-01-12 Martin Liska <mliska@suse.cz>
4309 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4311 (possible_polymorphic_call_targets): Use it.
4312 (ipa_devirt): Likewise.
4314 2018-01-12 Martin Liska <mliska@suse.cz>
4316 * profile-count.h (enum profile_quality): Use 0 as invalid
4317 enum value of profile_quality.
4319 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4321 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4322 -mext-string options.
4324 2018-01-12 Richard Biener <rguenther@suse.de>
4326 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4327 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4328 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4330 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4332 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4334 * configure.ac (--with-long-double-format): Add support for the
4335 configuration option to change the default long double format on
4337 * config.gcc (powerpc*-linux*-*): Likewise.
4338 * configure: Regenerate.
4339 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4340 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4341 used without modification.
4343 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4345 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4346 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4347 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4348 MISC_BUILTIN_SPEC_BARRIER.
4349 (rs6000_init_builtins): Likewise.
4350 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4352 (speculation_barrier): New define_insn.
4353 * doc/extend.texi: Document __builtin_speculation_barrier.
4355 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4358 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4359 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4360 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4362 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4363 integral modes instead of "ss" and "sd".
4364 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4365 vectors with 32-bit and 64-bit elements.
4366 (vecdupssescalarmodesuffix): New mode attribute.
4367 (vec_dup<mode>): Use it.
4369 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4372 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4373 frame if argument is passed on stack.
4375 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4378 * ree.c (combine_reaching_defs): Optimize also
4379 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4380 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4382 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4385 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4387 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4390 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4391 after they are computed.
4393 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4395 PR tree-optimization/83695
4396 * gimple-loop-linterchange.cc
4397 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4398 reset cached scev information after interchange.
4399 (pass_linterchange::execute): Remove call to scev_reset_htab.
4401 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4403 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4404 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4405 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4406 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4407 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4408 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4409 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4410 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4411 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4412 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4413 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4414 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4415 (V_lane_reg): Likewise.
4416 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4418 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4419 (vfmal_lane_low<mode>_intrinsic,
4420 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4421 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4422 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4423 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4424 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4425 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4427 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4429 * config/arm/arm-cpus.in (fp16fml): New feature.
4430 (ALL_SIMD): Add fp16fml.
4431 (armv8.2-a): Add fp16fml as an option.
4432 (armv8.3-a): Likewise.
4433 (armv8.4-a): Add fp16fml as part of fp16.
4434 * config/arm/arm.h (TARGET_FP16FML): Define.
4435 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4437 * config/arm/arm-modes.def (V2HF): Define.
4438 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4439 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4440 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4441 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4442 vfmsl_low, vfmsl_high): New set of builtins.
4443 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4444 (vfml_op): New code attribute.
4445 (VFMLHALVES): New int iterator.
4446 (VFML, VFMLSEL): New mode attributes.
4447 (V_reg): Define mapping for V2HF.
4448 (V_hi, V_lo): New mode attributes.
4449 (VF_constraint): Likewise.
4450 (vfml_half, vfml_half_selector): New int attributes.
4451 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4453 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4454 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4456 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4457 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4458 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4459 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4461 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4462 Document new effective target and option set.
4464 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4466 * config/arm/arm-cpus.in (armv8_4): New feature.
4467 (ARMv8_4a): New fgroup.
4468 (armv8.4-a): New arch.
4469 * config/arm/arm-tables.opt: Regenerate.
4470 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4471 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4472 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4473 Add matching rules for -march=armv8.4-a and extensions.
4474 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4476 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4479 * config/rx/rx.md (BW): New mode attribute.
4480 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4482 2018-01-11 Richard Biener <rguenther@suse.de>
4484 PR tree-optimization/83435
4485 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4486 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4487 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4489 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4490 Alan Hayward <alan.hayward@arm.com>
4491 David Sherwood <david.sherwood@arm.com>
4493 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4495 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4496 (aarch64_print_address_internal): Use it to check for a zero offset.
4498 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4499 Alan Hayward <alan.hayward@arm.com>
4500 David Sherwood <david.sherwood@arm.com>
4502 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4503 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4504 Return a poly_int64 rather than a HOST_WIDE_INT.
4505 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4506 rather than a HOST_WIDE_INT.
4507 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4508 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4509 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4510 final_offset from HOST_WIDE_INT to poly_int64.
4511 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4512 to_constant when getting the number of units in an Advanced SIMD
4514 (aarch64_builtin_vectorized_function): Check for a constant number
4516 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4518 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4519 attribute instead of GET_MODE_NUNITS.
4520 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4521 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4522 GET_MODE_SIZE for fixed-size registers.
4523 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4524 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4525 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4526 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4527 (aarch64_print_operand, aarch64_print_address_internal)
4528 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4529 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4530 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4531 Handle polynomial GET_MODE_SIZE.
4532 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4533 wider than SImode without modification.
4534 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4535 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4536 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4537 passing and returning SVE modes.
4538 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4539 rather than GEN_INT.
4540 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4541 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4542 (aarch64_allocate_and_probe_stack_space): Likewise.
4543 (aarch64_layout_frame): Cope with polynomial offsets.
4544 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4545 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4547 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4548 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4549 poly_int64 rather than a HOST_WIDE_INT.
4550 (aarch64_get_separate_components, aarch64_process_components)
4551 (aarch64_expand_prologue, aarch64_expand_epilogue)
4552 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4553 (aarch64_anchor_offset): New function, split out from...
4554 (aarch64_legitimize_address): ...here.
4555 (aarch64_builtin_vectorization_cost): Handle polynomial
4556 TYPE_VECTOR_SUBPARTS.
4557 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4559 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4560 number of elements from the PARALLEL rather than the mode.
4561 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4562 rather than GET_MODE_BITSIZE.
4563 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4564 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4565 (aarch64_expand_vec_perm_const_1): Handle polynomial
4566 d->perm.length () and d->perm elements.
4567 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4568 Apply to_constant to d->perm elements.
4569 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4570 polynomial CONST_VECTOR_NUNITS.
4571 (aarch64_move_pointer): Take amount as a poly_int64 rather
4573 (aarch64_progress_pointer): Avoid temporary variable.
4574 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4575 the mode attribute instead of GET_MODE.
4577 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4578 Alan Hayward <alan.hayward@arm.com>
4579 David Sherwood <david.sherwood@arm.com>
4581 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4582 x exists before using it.
4583 (aarch64_add_constant_internal): Rename to...
4584 (aarch64_add_offset_1): ...this. Replace regnum with separate
4585 src and dest rtxes. Handle the case in which they're different,
4586 including when the offset is zero. Replace scratchreg with an rtx.
4587 Use 2 additions if there is no spare register into which we can
4588 move a 16-bit constant.
4589 (aarch64_add_constant): Delete.
4590 (aarch64_add_offset): Replace reg with separate src and dest
4591 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4592 Use aarch64_add_offset_1.
4593 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4594 an rtx rather than an int. Take the delta as a poly_int64
4595 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4596 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4597 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4598 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4599 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4601 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4602 aarch64_add_constant.
4604 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4606 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4607 Use scalar_float_mode.
4609 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4611 * config/aarch64/aarch64-simd.md
4612 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4613 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4614 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4615 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4616 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4617 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4618 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4619 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4620 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4621 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4623 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4626 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4627 targ_options->x_arm_arch_string is non NULL.
4629 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4631 * config/aarch64/aarch64.h
4632 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4634 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4637 * expmed.c (emit_store_flag_force): Swap if const op0
4638 and change VOIDmode to mode of op0.
4640 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4642 PR rtl-optimization/83761
4643 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4644 than bytes to mode_for_size.
4646 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4649 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4650 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4653 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4656 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4657 when in layout mode.
4658 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4659 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4662 2018-01-10 Michael Collison <michael.collison@arm.com>
4664 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4665 * config/aarch64/aarch64-option-extension.def: Add
4666 AARCH64_OPT_EXTENSION of 'fp16fml'.
4667 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4668 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4669 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4670 * config/aarch64/constraints.md (Ui7): New constraint.
4671 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4672 (VFMLA_SEL_W): Ditto.
4675 (VFMLA16_LOW): New int iterator.
4676 (VFMLA16_HIGH): Ditto.
4677 (UNSPEC_FMLAL): New unspec.
4678 (UNSPEC_FMLSL): Ditto.
4679 (UNSPEC_FMLAL2): Ditto.
4680 (UNSPEC_FMLSL2): Ditto.
4681 (f16mac): New code attribute.
4682 * config/aarch64/aarch64-simd-builtins.def
4683 (aarch64_fmlal_lowv2sf): Ditto.
4684 (aarch64_fmlsl_lowv2sf): Ditto.
4685 (aarch64_fmlalq_lowv4sf): Ditto.
4686 (aarch64_fmlslq_lowv4sf): Ditto.
4687 (aarch64_fmlal_highv2sf): Ditto.
4688 (aarch64_fmlsl_highv2sf): Ditto.
4689 (aarch64_fmlalq_highv4sf): Ditto.
4690 (aarch64_fmlslq_highv4sf): Ditto.
4691 (aarch64_fmlal_lane_lowv2sf): Ditto.
4692 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4693 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4694 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4695 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4696 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4697 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4698 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4699 (aarch64_fmlal_lane_highv2sf): Ditto.
4700 (aarch64_fmlsl_lane_highv2sf): Ditto.
4701 (aarch64_fmlal_laneq_highv2sf): Ditto.
4702 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4703 (aarch64_fmlalq_lane_highv4sf): Ditto.
4704 (aarch64_fmlsl_lane_highv4sf): Ditto.
4705 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4706 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4707 * config/aarch64/aarch64-simd.md:
4708 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4709 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4710 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4711 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4712 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4713 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4714 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4715 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4716 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4717 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4718 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4719 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4720 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4721 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4722 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4723 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4724 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4725 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4726 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4727 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4728 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4729 (vfmlsl_low_u32): Ditto.
4730 (vfmlalq_low_u32): Ditto.
4731 (vfmlslq_low_u32): Ditto.
4732 (vfmlal_high_u32): Ditto.
4733 (vfmlsl_high_u32): Ditto.
4734 (vfmlalq_high_u32): Ditto.
4735 (vfmlslq_high_u32): Ditto.
4736 (vfmlal_lane_low_u32): Ditto.
4737 (vfmlsl_lane_low_u32): Ditto.
4738 (vfmlal_laneq_low_u32): Ditto.
4739 (vfmlsl_laneq_low_u32): Ditto.
4740 (vfmlalq_lane_low_u32): Ditto.
4741 (vfmlslq_lane_low_u32): Ditto.
4742 (vfmlalq_laneq_low_u32): Ditto.
4743 (vfmlslq_laneq_low_u32): Ditto.
4744 (vfmlal_lane_high_u32): Ditto.
4745 (vfmlsl_lane_high_u32): Ditto.
4746 (vfmlal_laneq_high_u32): Ditto.
4747 (vfmlsl_laneq_high_u32): Ditto.
4748 (vfmlalq_lane_high_u32): Ditto.
4749 (vfmlslq_lane_high_u32): Ditto.
4750 (vfmlalq_laneq_high_u32): Ditto.
4751 (vfmlslq_laneq_high_u32): Ditto.
4752 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4753 (AARCH64_FL_FOR_ARCH8_4): New.
4754 (AARCH64_ISA_F16FML): New ISA flag.
4755 (TARGET_F16FML): New feature flag for fp16fml.
4756 (doc/invoke.texi): Document new fp16fml option.
4758 2018-01-10 Michael Collison <michael.collison@arm.com>
4760 * config/aarch64/aarch64-builtins.c:
4761 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4762 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4763 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4764 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4765 (AARCH64_ISA_SHA3): New ISA flag.
4766 (TARGET_SHA3): New feature flag for sha3.
4767 * config/aarch64/iterators.md (sha512_op): New int attribute.
4768 (CRYPTO_SHA512): New int iterator.
4769 (UNSPEC_SHA512H): New unspec.
4770 (UNSPEC_SHA512H2): Ditto.
4771 (UNSPEC_SHA512SU0): Ditto.
4772 (UNSPEC_SHA512SU1): Ditto.
4773 * config/aarch64/aarch64-simd-builtins.def
4774 (aarch64_crypto_sha512hqv2di): New builtin.
4775 (aarch64_crypto_sha512h2qv2di): Ditto.
4776 (aarch64_crypto_sha512su0qv2di): Ditto.
4777 (aarch64_crypto_sha512su1qv2di): Ditto.
4778 (aarch64_eor3qv8hi): Ditto.
4779 (aarch64_rax1qv2di): Ditto.
4780 (aarch64_xarqv2di): Ditto.
4781 (aarch64_bcaxqv8hi): Ditto.
4782 * config/aarch64/aarch64-simd.md:
4783 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4784 (aarch64_crypto_sha512su0qv2di): Ditto.
4785 (aarch64_crypto_sha512su1qv2di): Ditto.
4786 (aarch64_eor3qv8hi): Ditto.
4787 (aarch64_rax1qv2di): Ditto.
4788 (aarch64_xarqv2di): Ditto.
4789 (aarch64_bcaxqv8hi): Ditto.
4790 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4791 (vsha512h2q_u64): Ditto.
4792 (vsha512su0q_u64): Ditto.
4793 (vsha512su1q_u64): Ditto.
4794 (veor3q_u16): Ditto.
4795 (vrax1q_u64): Ditto.
4797 (vbcaxq_u16): Ditto.
4798 * config/arm/types.md (crypto_sha512): New type attribute.
4799 (crypto_sha3): Ditto.
4800 (doc/invoke.texi): Document new sha3 option.
4802 2018-01-10 Michael Collison <michael.collison@arm.com>
4804 * config/aarch64/aarch64-builtins.c:
4805 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4806 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4807 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4808 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4809 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4810 (AARCH64_ISA_SM4): New ISA flag.
4811 (TARGET_SM4): New feature flag for sm4.
4812 * config/aarch64/aarch64-simd-builtins.def
4813 (aarch64_sm3ss1qv4si): Ditto.
4814 (aarch64_sm3tt1aq4si): Ditto.
4815 (aarch64_sm3tt1bq4si): Ditto.
4816 (aarch64_sm3tt2aq4si): Ditto.
4817 (aarch64_sm3tt2bq4si): Ditto.
4818 (aarch64_sm3partw1qv4si): Ditto.
4819 (aarch64_sm3partw2qv4si): Ditto.
4820 (aarch64_sm4eqv4si): Ditto.
4821 (aarch64_sm4ekeyqv4si): Ditto.
4822 * config/aarch64/aarch64-simd.md:
4823 (aarch64_sm3ss1qv4si): Ditto.
4824 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4825 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4826 (aarch64_sm4eqv4si): Ditto.
4827 (aarch64_sm4ekeyqv4si): Ditto.
4828 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4829 (sm3part_op): Ditto.
4830 (CRYPTO_SM3TT): Ditto.
4831 (CRYPTO_SM3PART): Ditto.
4832 (UNSPEC_SM3SS1): New unspec.
4833 (UNSPEC_SM3TT1A): Ditto.
4834 (UNSPEC_SM3TT1B): Ditto.
4835 (UNSPEC_SM3TT2A): Ditto.
4836 (UNSPEC_SM3TT2B): Ditto.
4837 (UNSPEC_SM3PARTW1): Ditto.
4838 (UNSPEC_SM3PARTW2): Ditto.
4839 (UNSPEC_SM4E): Ditto.
4840 (UNSPEC_SM4EKEY): Ditto.
4841 * config/aarch64/constraints.md (Ui2): New constraint.
4842 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4843 * config/arm/types.md (crypto_sm3): New type attribute.
4844 (crypto_sm4): Ditto.
4845 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4846 (vsm3tt1aq_u32): Ditto.
4847 (vsm3tt1bq_u32): Ditto.
4848 (vsm3tt2aq_u32): Ditto.
4849 (vsm3tt2bq_u32): Ditto.
4850 (vsm3partw1q_u32): Ditto.
4851 (vsm3partw2q_u32): Ditto.
4852 (vsm4eq_u32): Ditto.
4853 (vsm4ekeyq_u32): Ditto.
4854 (doc/invoke.texi): Document new sm4 option.
4856 2018-01-10 Michael Collison <michael.collison@arm.com>
4858 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4859 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4860 (AARCH64_FL_FOR_ARCH8_4): New.
4861 (AARCH64_FL_V8_4): New flag.
4862 (doc/invoke.texi): Document new armv8.4-a option.
4864 2018-01-10 Michael Collison <michael.collison@arm.com>
4866 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4867 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4868 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4869 * config/aarch64/aarch64-option-extension.def: Add
4870 AARCH64_OPT_EXTENSION of 'sha2'.
4871 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4872 (crypto): Disable sha2 and aes if crypto disabled.
4873 (crypto): Enable aes and sha2 if enabled.
4874 (simd): Disable sha2 and aes if simd disabled.
4875 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4877 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4878 (TARGET_SHA2): New feature flag for sha2.
4879 (TARGET_AES): New feature flag for aes.
4880 * config/aarch64/aarch64-simd.md:
4881 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4882 conditional on TARGET_AES.
4883 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4884 (aarch64_crypto_sha1hsi): Make pattern conditional
4886 (aarch64_crypto_sha1hv4si): Ditto.
4887 (aarch64_be_crypto_sha1hv4si): Ditto.
4888 (aarch64_crypto_sha1su1v4si): Ditto.
4889 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4890 (aarch64_crypto_sha1su0v4si): Ditto.
4891 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4892 (aarch64_crypto_sha256su0v4si): Ditto.
4893 (aarch64_crypto_sha256su1v4si): Ditto.
4894 (doc/invoke.texi): Document new aes and sha2 options.
4896 2018-01-10 Martin Sebor <msebor@redhat.com>
4898 PR tree-optimization/83781
4899 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4902 2018-01-11 Martin Sebor <msebor@gmail.com>
4903 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4905 PR tree-optimization/83501
4906 PR tree-optimization/81703
4908 * tree-ssa-strlen.c (get_string_cst): Rename...
4909 (get_string_len): ...to this. Handle global constants.
4910 (handle_char_store): Adjust.
4912 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4913 Jim Wilson <jimw@sifive.com>
4915 * config/riscv/riscv-protos.h (riscv_output_return): New.
4916 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4917 (riscv_attribute_table, riscv_output_return),
4918 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4919 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4920 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4922 (riscv_expand_prologue): Add early return for naked function.
4923 (riscv_expand_epilogue): Likewise.
4924 (riscv_function_ok_for_sibcall): Return false for naked function.
4925 (riscv_set_current_function): New.
4926 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4927 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4928 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4929 * doc/extend.texi (RISC-V Function Attributes): New.
4931 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4933 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4934 check for 128-bit long double before checking TCmode.
4935 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4936 128-bit long doubles before checking TFmode or TCmode.
4937 (FLOAT128_IBM_P): Likewise.
4939 2018-01-10 Martin Sebor <msebor@redhat.com>
4941 PR tree-optimization/83671
4942 * builtins.c (c_strlen): Unconditionally return zero for the empty
4944 Use -Warray-bounds for warnings.
4945 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4946 for non-constant array indices with COMPONENT_REF, arrays of
4947 arrays, and pointers to arrays.
4948 (gimple_fold_builtin_strlen): Determine and set length range for
4949 non-constant character arrays.
4951 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4954 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4957 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4959 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4961 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4964 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4965 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4966 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4967 indexed_or_indirect_operand predicate.
4968 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4969 (*vsx_le_perm_load_v8hi): Likewise.
4970 (*vsx_le_perm_load_v16qi): Likewise.
4971 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4972 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4973 (*vsx_le_perm_store_v8hi): Likewise.
4974 (*vsx_le_perm_store_v16qi): Likewise.
4975 (eight unnamed splitters): Likewise.
4977 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4979 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4980 * config/rs6000/emmintrin.h: Likewise.
4981 * config/rs6000/mmintrin.h: Likewise.
4982 * config/rs6000/xmmintrin.h: Likewise.
4984 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4987 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4989 * tree.c (tree_nop_conversion): Return true for location wrapper
4991 (maybe_wrap_with_location): New function.
4992 (selftest::check_strip_nops): New function.
4993 (selftest::test_location_wrappers): New function.
4994 (selftest::tree_c_tests): Call it.
4995 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4996 (maybe_wrap_with_location): New decl.
4997 (EXPR_LOCATION_WRAPPER_P): New macro.
4998 (location_wrapper_p): New inline function.
4999 (tree_strip_any_location_wrapper): New inline function.
5001 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
5004 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5005 stack_realign_offset for the largest alignment of stack slot
5007 (ix86_find_max_used_stack_alignment): New function.
5008 (ix86_finalize_stack_frame_flags): Use it. Set
5009 max_used_stack_alignment if we don't realign stack.
5010 * config/i386/i386.h (machine_function): Add
5011 max_used_stack_alignment.
5013 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5015 * config/arm/arm.opt (-mbranch-cost): New option.
5016 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5019 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5022 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5023 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5025 2018-01-10 Richard Biener <rguenther@suse.de>
5028 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5029 early out so it also covers the case where we have a non-NULL
5032 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5034 PR tree-optimization/83753
5035 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5036 for non-strided grouped accesses if the number of elements is 1.
5038 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5041 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5042 * i386.h (TARGET_USE_GATHER): Define.
5043 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5045 2018-01-10 Martin Liska <mliska@suse.cz>
5048 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5049 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5051 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5052 CLEANUP_NO_PARTITIONING is not set.
5054 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5056 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5057 for vectors, as a partial revert of r254296.
5058 * rtl.h (const_vec_p): Delete.
5059 (const_vec_duplicate_p): Don't test for vector CONSTs.
5060 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5061 * expmed.c (make_tree): Likewise.
5064 * common.md (E, F): Use CONSTANT_P instead of checking for
5066 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5067 checking for CONST_VECTOR.
5069 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5072 * predict.c (force_edge_cold): Handle in more sane way edges
5075 2018-01-09 Carl Love <cel@us.ibm.com>
5077 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5079 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5080 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5081 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5082 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5083 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5084 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5085 * config/rs6000/rs6000-protos.h: Add extern defition for
5086 rs6000_generate_float2_double_code.
5087 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5089 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5090 (float2_v2df): Add define_expand.
5092 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5095 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5096 op_mode in the force_to_mode call.
5098 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5100 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5101 instead of checking each element individually.
5102 (aarch64_evpc_uzp): Likewise.
5103 (aarch64_evpc_zip): Likewise.
5104 (aarch64_evpc_ext): Likewise.
5105 (aarch64_evpc_rev): Likewise.
5106 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5107 instead of checking each element individually. Return true without
5109 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5110 whether all selected elements come from the same input, instead of
5111 checking each element individually. Remove calls to gen_rtx_REG,
5112 start_sequence and end_sequence and instead assert that no rtl is
5115 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5117 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5118 order of HIGH and CONST checks.
5120 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5122 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5123 if the destination isn't an SSA_NAME.
5125 2018-01-09 Richard Biener <rguenther@suse.de>
5127 PR tree-optimization/83668
5128 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5130 (canonicalize_loop_form): ... here, renamed from ...
5131 (canonicalize_loop_closed_ssa_form): ... this and amended to
5132 swap successor edges for loop exit blocks to make us use
5133 the RPO order we need for initial schedule generation.
5135 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5137 PR tree-optimization/64811
5138 * match.pd: When optimizing comparisons with Inf, avoid
5139 introducing or losing exceptions from comparisons with NaN.
5141 2018-01-09 Martin Liska <mliska@suse.cz>
5144 * asan.c (shadow_mem_size): Add gcc_assert.
5146 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5148 Don't save registers in main().
5151 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5152 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5153 * config/avr/avr.c (avr_set_current_function): Don't error if
5154 naked, OS_task or OS_main are specified at the same time.
5155 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5157 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5159 * common/config/avr/avr-common.c (avr_option_optimization_table):
5160 Switch on -mmain-is-OS_task for optimizing compilations.
5162 2018-01-09 Richard Biener <rguenther@suse.de>
5164 PR tree-optimization/83572
5165 * graphite.c: Include cfganal.h.
5166 (graphite_transform_loops): Connect infinite loops to exit
5167 and remove fake edges at the end.
5169 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5171 * ipa-inline.c (edge_badness): Revert accidental checkin.
5173 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5176 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5177 symbols; not inline clones.
5179 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5182 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5183 hard registers. Formatting fixes.
5185 PR preprocessor/83722
5186 * gcc.c (try_generate_repro): Pass
5187 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5188 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5191 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5192 Kito Cheng <kito.cheng@gmail.com>
5194 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5195 (riscv_leaf_function_p): Delete.
5196 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5198 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5200 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5202 (do_ifelse): New function.
5203 (do_isel): New function.
5204 (do_sub3): New function.
5205 (do_add3): New function.
5206 (do_load_mask_compare): New function.
5207 (do_overlap_load_compare): New function.
5208 (expand_compare_loop): New function.
5209 (expand_block_compare): Call expand_compare_loop() when appropriate.
5210 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5212 (-mblock-compare-inline-loop-limit): New option.
5214 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5217 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5218 Reverse order of second and third operands in first alternative.
5219 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5220 of first and second elements in UNSPEC_VPERMR vector.
5221 (altivec_expand_vec_perm_le): Likewise.
5223 2018-01-08 Jeff Law <law@redhat.com>
5225 PR rtl-optimizatin/81308
5226 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5227 (process_switch): If group_case_labels makes a change, then set
5229 (pass_convert_switch::execute): If a switch is converted, then
5230 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5232 PR rtl-optimization/81308
5233 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5236 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5238 PR target/83663 - Revert r255946
5239 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5240 generation for cases where splatting a value is not useful.
5241 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5242 across a vec_duplicate and a paradoxical subreg forming a vector
5243 mode to a vec_concat.
5245 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5247 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5248 -march=armv8.3-a variants.
5249 * config/arm/t-multilib: Likewise.
5250 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5252 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5254 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5256 (cceq_ior_compare_complement): Give it a name so I can use it, and
5257 change boolean_or_operator predicate to boolean_operator so it can
5258 be used to generate a crand.
5259 (eqne): New code iterator.
5260 (bd/bd_neg): New code_attrs.
5261 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5262 a single define_insn.
5263 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5264 decrement (bdnzt/bdnzf/bdzt/bdzf).
5265 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5266 with the new names of the branch decrement patterns, and added the
5267 names of the branch decrement conditional patterns.
5269 2018-01-08 Richard Biener <rguenther@suse.de>
5271 PR tree-optimization/83563
5272 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5275 2018-01-08 Richard Biener <rguenther@suse.de>
5278 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5280 2018-01-08 Richard Biener <rguenther@suse.de>
5282 PR tree-optimization/83685
5283 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5284 references to abnormals.
5286 2018-01-08 Richard Biener <rguenther@suse.de>
5289 * dwarf2out.c (output_indirect_strings): Handle empty
5290 skeleton_debug_str_hash.
5291 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5293 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5295 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5296 (emit_store_direct): Likewise.
5297 (arc_trampoline_adjust_address): Likewise.
5298 (arc_asm_trampoline_template): New function.
5299 (arc_initialize_trampoline): Use asm_trampoline_template.
5300 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5301 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5302 * config/arc/arc.md (flush_icache): Delete pattern.
5304 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5306 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5307 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5310 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5313 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5314 by not USED_FOR_TARGET.
5315 (make_pass_resolve_sw_modes): Likewise.
5317 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5319 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5322 2018-01-08 Richard Biener <rguenther@suse.de>
5325 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5327 2018-01-08 Richard Biener <rguenther@suse.de>
5330 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5332 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5335 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5336 basic blocks with a small number of successors.
5337 (convert_control_dep_chain_into_preds): Improve handling of
5339 (dump_predicates): Split apart into...
5340 (dump_pred_chain): ...here...
5341 (dump_pred_info): ...and here.
5342 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5343 (can_chain_union_be_invalidated_p): Improve check for invalidation
5345 (uninit_uses_cannot_happen): Avoid unnecessary if
5346 convert_control_dep_chain_into_preds yielded nothing.
5348 2018-01-06 Martin Sebor <msebor@redhat.com>
5350 PR tree-optimization/83640
5351 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5352 subtracting negative offset from size.
5353 (builtin_access::overlap): Adjust offset bounds of the access to fall
5354 within the size of the object if possible.
5356 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5358 PR rtl-optimization/83699
5359 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5360 extract_bit_field_as_subreg to cases in which the extracted
5361 value is also a vector.
5363 * lra-constraints.c (process_alt_operands): Test for the equivalence
5364 substitutions when detecting a possible reload cycle.
5366 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5369 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5370 by default if flag_selective_schedling{,2}. Formatting fixes.
5372 PR rtl-optimization/83682
5373 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5374 if it has non-VECTOR_MODE element mode.
5375 (vec_duplicate_p): Likewise.
5378 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5379 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5381 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5384 * config/i386/i386-builtin.def
5385 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5386 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5387 Require also OPTION_MASK_ISA_AVX512F in addition to
5388 OPTION_MASK_ISA_GFNI.
5389 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5390 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5391 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5392 to OPTION_MASK_ISA_GFNI.
5393 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5394 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5395 OPTION_MASK_ISA_AVX512BW.
5396 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5397 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5398 addition to OPTION_MASK_ISA_GFNI.
5399 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5400 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5401 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5402 to OPTION_MASK_ISA_GFNI.
5403 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5404 a requirement for all ISAs rather than any of them with a few
5406 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5408 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5409 bitmasks to be enabled with 3 exceptions, instead of requiring any
5410 enabled ISA with lots of exceptions.
5411 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5412 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5413 Change avx512bw in isa attribute to avx512f.
5414 * config/i386/sgxintrin.h: Add license boilerplate.
5415 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5416 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5417 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5418 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5420 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5421 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5422 temporarily sse2 rather than sse if not enabled already.
5425 * config/i386/sse.md (VI248_VLBW): Rename to ...
5426 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5427 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5428 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5429 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5430 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5431 mode iterator instead of VI248_VLBW.
5433 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5435 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5436 (record_modified): Skip clobbers; add debug output.
5437 (param_change_prob): Use sreal frequencies.
5439 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5441 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5442 punt for user-aligned variables.
5444 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5446 * tree-chrec.c (chrec_contains_symbols): Return true for
5449 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5452 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5453 of (x|y) == x for BICS pattern.
5455 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5457 PR tree-optimization/83605
5458 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5459 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5462 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5464 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5465 * config/epiphany/rtems.h: New file.
5467 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5468 Uros Bizjak <ubizjak@gmail.com>
5471 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5472 QIreg_operand instead of register_operand predicate.
5473 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5474 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5475 comments instead of -fmitigate[-_]rop.
5477 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5480 * cgraphunit.c (symbol_table::compile): Switch to text_section
5481 before calling assembly_start debug hook.
5482 * run-rtl-passes.c (run_rtl_passes): Likewise.
5485 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5487 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5488 range_int_cst_p rather than !symbolic_range_p before calling
5489 extract_range_from_multiplicative_op_1.
5491 2018-01-04 Jeff Law <law@redhat.com>
5493 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5494 redundant test in assertion.
5496 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5498 * doc/rtl.texi: Document machine_mode wrapper classes.
5500 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5502 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5505 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5507 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5508 the VEC_PERM_EXPR fold to fail.
5510 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5513 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5514 to switched_sections.
5516 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5519 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5522 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5525 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5526 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5528 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5531 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5532 is BLKmode and bitpos not zero or mode change is needed.
5534 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5537 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5540 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5543 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5544 instead of MULT rtx. Update all corresponding splitters.
5546 (*ssub<modesuffix>): Ditto.
5548 (*cmp_sadd_di): Update split patterns.
5549 (*cmp_sadd_si): Ditto.
5550 (*cmp_sadd_sidi): Ditto.
5551 (*cmp_ssub_di): Ditto.
5552 (*cmp_ssub_si): Ditto.
5553 (*cmp_ssub_sidi): Ditto.
5554 * config/alpha/predicates.md (const23_operand): New predicate.
5555 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5556 Look for ASHIFT, not MULT inner operand.
5557 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5559 2018-01-04 Martin Liska <mliska@suse.cz>
5561 PR gcov-profile/83669
5562 * gcov.c (output_intermediate_file): Add version to intermediate
5564 * doc/gcov.texi: Document new field 'version' in intermediate
5565 file format. Fix location of '-k' option of gcov command.
5567 2018-01-04 Martin Liska <mliska@suse.cz>
5570 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5572 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5574 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5576 2018-01-03 Martin Sebor <msebor@redhat.com>
5578 PR tree-optimization/83655
5579 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5580 checking calls with invalid arguments.
5582 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5584 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5585 (vectorizable_mask_load_store): Delete.
5586 (vectorizable_call): Return false for masked loads and stores.
5587 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5588 instead of gimple_assign_rhs1.
5589 (vectorizable_load): Handle IFN_MASK_LOAD.
5590 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5592 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5594 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5596 (vectorizable_mask_load_store): ...here.
5597 (vectorizable_load): ...and here.
5599 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5601 * tree-vect-stmts.c (vect_build_all_ones_mask)
5602 (vect_build_zero_merge_argument): New functions, split out from...
5603 (vectorizable_load): ...here.
5605 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5607 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5609 (vectorizable_mask_load_store): ...here.
5610 (vectorizable_store): ...and here.
5612 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5614 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5616 (vectorizable_mask_load_store): ...here.
5618 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5620 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5621 (vect_model_store_cost): Take a vec_load_store_type instead of a
5623 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5624 (vect_model_store_cost): Take a vec_load_store_type instead of a
5626 (vectorizable_mask_load_store): Update accordingly.
5627 (vectorizable_store): Likewise.
5628 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5630 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5632 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5633 IFN_MASK_LOAD calls here rather than...
5634 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5636 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5637 Alan Hayward <alan.hayward@arm.com>
5638 David Sherwood <david.sherwood@arm.com>
5640 * expmed.c (extract_bit_field_1): For vector extracts,
5641 fall back to extract_bit_field_as_subreg if vec_extract
5644 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5645 Alan Hayward <alan.hayward@arm.com>
5646 David Sherwood <david.sherwood@arm.com>
5648 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5649 they are variable or constant sized.
5650 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5651 slots for constant-sized data.
5653 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5654 Alan Hayward <alan.hayward@arm.com>
5655 David Sherwood <david.sherwood@arm.com>
5657 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5658 handling COND_EXPRs with boolean comparisons, try to find a better
5659 basis for the mask type than the boolean itself.
5661 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5663 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5664 is calculated and how it can be overridden.
5665 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5666 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5668 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5671 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5672 Alan Hayward <alan.hayward@arm.com>
5673 David Sherwood <david.sherwood@arm.com>
5675 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5676 Remove the mode argument.
5677 (aarch64_simd_valid_immediate): Remove the mode and inverse
5679 * config/aarch64/iterators.md (bitsize): New iterator.
5680 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5681 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5682 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5683 aarch64_simd_valid_immediate.
5684 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5685 (aarch64_reg_or_bic_imm): Likewise.
5686 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5687 with an insn_type enum and msl with a modifier_type enum.
5688 Replace element_width with a scalar_mode. Change the shift
5689 to unsigned int. Add constructors for scalar_float_mode and
5690 scalar_int_mode elements.
5691 (aarch64_vect_float_const_representable_p): Delete.
5692 (aarch64_can_const_movi_rtx_p)
5693 (aarch64_simd_scalar_immediate_valid_for_move)
5694 (aarch64_simd_make_constant): Update call to
5695 aarch64_simd_valid_immediate.
5696 (aarch64_advsimd_valid_immediate_hs): New function.
5697 (aarch64_advsimd_valid_immediate): Likewise.
5698 (aarch64_simd_valid_immediate): Remove mode and inverse
5699 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5700 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5701 and aarch64_float_const_representable_p on the result.
5702 (aarch64_output_simd_mov_immediate): Remove mode argument.
5703 Update call to aarch64_simd_valid_immediate and use of
5704 simd_immediate_info.
5705 (aarch64_output_scalar_simd_mov_immediate): Update call
5708 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5709 Alan Hayward <alan.hayward@arm.com>
5710 David Sherwood <david.sherwood@arm.com>
5712 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5713 (mode_nunits): Likewise CONST_MODE_NUNITS.
5714 * machmode.def (ADJUST_NUNITS): Document.
5715 * genmodes.c (mode_data::need_nunits_adj): New field.
5716 (blank_mode): Update accordingly.
5717 (adj_nunits): New variable.
5718 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5720 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5721 listed in adj_nunits.
5722 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5723 listed in adj_nunits. Don't emit case statements for such modes.
5724 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5725 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5726 nothing if adj_nunits is nonnull.
5727 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5728 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5729 (emit_mode_fbit): Update use of print_maybe_const_decl.
5730 (emit_move_size): Likewise. Treat the array as non-const
5732 (emit_mode_adjustments): Handle adj_nunits.
5734 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5736 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5737 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5738 (VECTOR_MODES): Use it.
5739 (make_vector_modes): Take the prefix as an argument.
5741 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5742 Alan Hayward <alan.hayward@arm.com>
5743 David Sherwood <david.sherwood@arm.com>
5745 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5746 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5747 for MODE_VECTOR_BOOL.
5748 * machmode.def (VECTOR_BOOL_MODE): Document.
5749 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5750 (make_vector_bool_mode): New function.
5751 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5753 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5754 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5756 * stor-layout.c (int_mode_for_mode): Likewise.
5757 * tree.c (build_vector_type_for_mode): Likewise.
5758 * varasm.c (output_constant_pool_2): Likewise.
5759 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5760 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5761 for MODE_VECTOR_BOOL.
5762 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5763 of mode class checks.
5764 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5765 instead of a list of mode class checks.
5766 (expand_vector_scalar_condition): Likewise.
5767 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5769 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5770 Alan Hayward <alan.hayward@arm.com>
5771 David Sherwood <david.sherwood@arm.com>
5773 * machmode.h (mode_size): Change from unsigned short to
5775 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5776 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5777 or if measurement_type is not polynomial.
5778 (fixed_size_mode::includes_p): Check for constant-sized modes.
5779 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5780 return a poly_uint16 rather than an unsigned short.
5781 (emit_mode_size): Change the type of mode_size from unsigned short
5782 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5783 (emit_mode_adjustments): Cope with polynomial vector sizes.
5784 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5786 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5788 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5789 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5790 * caller-save.c (setup_save_areas): Likewise.
5791 (replace_reg_with_saved_mem): Likewise.
5792 * calls.c (emit_library_call_value_1): Likewise.
5793 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5794 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5795 (gen_lowpart_for_combine): Likewise.
5796 * convert.c (convert_to_integer_1): Likewise.
5797 * cse.c (equiv_constant, cse_insn): Likewise.
5798 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5799 (cselib_subst_to_values): Likewise.
5800 * dce.c (word_dce_process_block): Likewise.
5801 * df-problems.c (df_word_lr_mark_ref): Likewise.
5802 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5803 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5804 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5805 (rtl_for_decl_location): Likewise.
5806 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5807 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5808 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5809 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5810 (expand_expr_real_1): Likewise.
5811 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5812 (pad_below): Likewise.
5813 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5814 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5815 * ira.c (get_subreg_tracking_sizes): Likewise.
5816 * ira-build.c (ira_create_allocno_objects): Likewise.
5817 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5818 (ira_sort_regnos_for_alter_reg): Likewise.
5819 * ira-costs.c (record_operand_costs): Likewise.
5820 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5821 (resolve_simple_move): Likewise.
5822 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5823 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5824 (lra_constraints): Likewise.
5825 (CONST_POOL_OK_P): Reject variable-sized modes.
5826 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5827 (add_pseudo_to_slot, lra_spill): Likewise.
5828 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5829 * optabs-query.c (get_best_extraction_insn): Likewise.
5830 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5831 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5832 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5833 * recog.c (offsettable_address_addr_space_p): Likewise.
5834 * regcprop.c (maybe_mode_change): Likewise.
5835 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5836 * regrename.c (build_def_use): Likewise.
5837 * regstat.c (dump_reg_info): Likewise.
5838 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5839 (find_reloads, find_reloads_subreg_address): Likewise.
5840 * reload1.c (eliminate_regs_1): Likewise.
5841 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5842 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5843 (simplify_binary_operation_1, simplify_subreg): Likewise.
5844 * targhooks.c (default_function_arg_padding): Likewise.
5845 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5846 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5847 (verify_gimple_assign_ternary): Likewise.
5848 * tree-inline.c (estimate_move_cost): Likewise.
5849 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5850 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5851 (get_address_cost_ainc): Likewise.
5852 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5853 (vect_supportable_dr_alignment): Likewise.
5854 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5855 (vectorizable_reduction): Likewise.
5856 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5857 (vectorizable_operation, vectorizable_load): Likewise.
5858 * tree.c (build_same_sized_truth_vector_type): Likewise.
5859 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5860 * var-tracking.c (emit_note_insn_var_location): Likewise.
5861 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5862 (ADDR_VEC_ALIGN): Likewise.
5864 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5865 Alan Hayward <alan.hayward@arm.com>
5866 David Sherwood <david.sherwood@arm.com>
5868 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5870 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5871 or if measurement_type is polynomial.
5872 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5873 * combine.c (make_extraction): Likewise.
5874 * dse.c (find_shift_sequence): Likewise.
5875 * dwarf2out.c (mem_loc_descriptor): Likewise.
5876 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5877 (extract_bit_field, extract_low_bits): Likewise.
5878 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5879 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5880 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5881 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5882 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5883 * reload.c (find_reloads): Likewise.
5884 * reload1.c (alter_reg): Likewise.
5885 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5886 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5887 * tree-if-conv.c (predicate_mem_writes): Likewise.
5888 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5889 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5890 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5891 * valtrack.c (dead_debug_insert_temp): Likewise.
5892 * varasm.c (mergeable_constant_section): Likewise.
5893 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5895 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5896 Alan Hayward <alan.hayward@arm.com>
5897 David Sherwood <david.sherwood@arm.com>
5899 * expr.c (expand_assignment): Cope with polynomial mode sizes
5900 when assigning to a CONCAT.
5902 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5903 Alan Hayward <alan.hayward@arm.com>
5904 David Sherwood <david.sherwood@arm.com>
5906 * machmode.h (mode_precision): Change from unsigned short to
5908 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5910 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5911 or if measurement_type is not polynomial.
5912 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5913 in which the mode is already known to be a scalar_int_mode.
5914 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5915 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5917 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5918 for GET_MODE_PRECISION.
5919 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5920 for GET_MODE_PRECISION.
5921 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5923 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5924 (expand_field_assignment, make_extraction): Likewise.
5925 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5926 (get_last_value): Likewise.
5927 * convert.c (convert_to_integer_1): Likewise.
5928 * cse.c (cse_insn): Likewise.
5929 * expr.c (expand_expr_real_1): Likewise.
5930 * lra-constraints.c (simplify_operand_subreg): Likewise.
5931 * optabs-query.c (can_atomic_load_p): Likewise.
5932 * optabs.c (expand_atomic_load): Likewise.
5933 (expand_atomic_store): Likewise.
5934 * ree.c (combine_reaching_defs): Likewise.
5935 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5936 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5937 * tree.h (type_has_mode_precision_p): Likewise.
5938 * ubsan.c (instrument_si_overflow): Likewise.
5940 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5941 Alan Hayward <alan.hayward@arm.com>
5942 David Sherwood <david.sherwood@arm.com>
5944 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5945 polynomial numbers of units.
5946 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5947 (valid_vector_subparts_p): New function.
5948 (build_vector_type): Remove temporary shim and take the number
5949 of units as a poly_uint64 rather than an int.
5950 (build_opaque_vector_type): Take the number of units as a
5951 poly_uint64 rather than an int.
5952 * tree.c (build_vector_from_ctor): Handle polynomial
5953 TYPE_VECTOR_SUBPARTS.
5954 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5955 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5956 (build_vector_from_val): If the number of units is variable,
5957 use build_vec_duplicate_cst for constant operands and
5958 VEC_DUPLICATE_EXPR otherwise.
5959 (make_vector_type): Remove temporary is_constant ().
5960 (build_vector_type, build_opaque_vector_type): Take the number of
5961 units as a poly_uint64 rather than an int.
5962 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5964 * cfgexpand.c (expand_debug_expr): Likewise.
5965 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5966 (store_constructor, expand_expr_real_1): Likewise.
5967 (const_scalar_mask_from_tree): Likewise.
5968 * fold-const-call.c (fold_const_reduction): Likewise.
5969 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5970 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5971 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5972 (fold_relational_const): Likewise.
5973 (native_interpret_vector): Likewise. Change the size from an
5974 int to an unsigned int.
5975 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5976 TYPE_VECTOR_SUBPARTS.
5977 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5978 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5979 duplicating a non-constant operand into a variable-length vector.
5980 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5981 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5982 * ipa-icf.c (sem_variable::equals): Likewise.
5983 * match.pd: Likewise.
5984 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5985 * print-tree.c (print_node): Likewise.
5986 * stor-layout.c (layout_type): Likewise.
5987 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5988 * tree-cfg.c (verify_gimple_comparison): Likewise.
5989 (verify_gimple_assign_binary): Likewise.
5990 (verify_gimple_assign_ternary): Likewise.
5991 (verify_gimple_assign_single): Likewise.
5992 * tree-pretty-print.c (dump_generic_node): Likewise.
5993 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5994 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5995 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5996 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5997 (vect_shift_permute_load_chain): Likewise.
5998 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5999 (expand_vector_condition, optimize_vector_constructor): Likewise.
6000 (lower_vec_perm, get_compute_type): Likewise.
6001 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6002 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
6003 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
6004 (vect_recog_mask_conversion_pattern): Likewise.
6005 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6006 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6007 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6008 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6009 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6010 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6011 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6012 (supportable_widening_operation): Likewise.
6013 (supportable_narrowing_operation): Likewise.
6014 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6016 * varasm.c (output_constant): Likewise.
6018 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6019 Alan Hayward <alan.hayward@arm.com>
6020 David Sherwood <david.sherwood@arm.com>
6022 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6023 so that both the length == 3 and length != 3 cases set up their
6024 own permute vectors. Add comments explaining why we know the
6025 number of elements is constant.
6026 (vect_permute_load_chain): Likewise.
6028 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6029 Alan Hayward <alan.hayward@arm.com>
6030 David Sherwood <david.sherwood@arm.com>
6032 * machmode.h (mode_nunits): Change from unsigned char to
6034 (ONLY_FIXED_SIZE_MODES): New macro.
6035 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6036 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6037 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6039 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6040 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6041 or if measurement_type is not polynomial.
6042 * genmodes.c (ZERO_COEFFS): New macro.
6043 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6045 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6046 Use ZERO_COEFFS when emitting initializers.
6047 * data-streamer.h (bp_pack_poly_value): New function.
6048 (bp_unpack_poly_value): Likewise.
6049 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6050 for GET_MODE_NUNITS.
6051 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6052 for GET_MODE_NUNITS.
6053 * tree.c (make_vector_type): Remove temporary shim and make
6054 the real function take the number of units as a poly_uint64
6056 (build_vector_type_for_mode): Handle polynomial nunits.
6057 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6058 * emit-rtl.c (const_vec_series_p_1): Likewise.
6059 (gen_rtx_CONST_VECTOR): Likewise.
6060 * fold-const.c (test_vec_duplicate_folding): Likewise.
6061 * genrecog.c (validate_pattern): Likewise.
6062 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6063 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6064 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6065 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6066 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6067 * rtlanal.c (subreg_get_info): Likewise.
6068 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6069 (vect_grouped_load_supported): Likewise.
6070 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6071 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6072 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6073 (simplify_const_unary_operation, simplify_binary_operation_1)
6074 (simplify_const_binary_operation, simplify_ternary_operation)
6075 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6076 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6077 instead of CONST_VECTOR_NUNITS.
6078 * varasm.c (output_constant_pool_2): Likewise.
6079 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6080 explicit-encoded elements in the XVEC for variable-length vectors.
6082 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6084 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6086 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6087 Alan Hayward <alan.hayward@arm.com>
6088 David Sherwood <david.sherwood@arm.com>
6090 * coretypes.h (fixed_size_mode): Declare.
6091 (fixed_size_mode_pod): New typedef.
6092 * builtins.h (target_builtins::x_apply_args_mode)
6093 (target_builtins::x_apply_result_mode): Change type to
6094 fixed_size_mode_pod.
6095 * builtins.c (apply_args_size, apply_result_size, result_vector)
6096 (expand_builtin_apply_args_1, expand_builtin_apply)
6097 (expand_builtin_return): Update accordingly.
6099 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6101 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6102 * cselib.c (cselib_hash_rtx): Likewise.
6103 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6104 CONST_VECTOR encoding.
6106 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6107 Jeff Law <law@redhat.com>
6110 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6111 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6112 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6113 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6116 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6117 explicitly probe *sp in a noreturn function if there were any callee
6118 register saves or frame pointer is needed.
6120 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6123 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6124 BLKmode for ternary, binary or unary expressions.
6127 * var-tracking.c (delete_vta_debug_insn): New inline function.
6128 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6129 insns from get_insns () to NULL instead of each bb separately.
6130 Use delete_vta_debug_insn. No longer static.
6131 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6132 delete_vta_debug_insns callers.
6133 * rtl.h (delete_vta_debug_insns): Declare.
6134 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6135 instead of variable_tracking_main.
6137 2018-01-03 Martin Sebor <msebor@redhat.com>
6139 PR tree-optimization/83603
6140 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6141 arguments past the endof the argument list in functions declared
6142 without a prototype.
6143 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6144 Avoid checking when arguments are null.
6146 2018-01-03 Martin Sebor <msebor@redhat.com>
6149 * doc/extend.texi (attribute const): Fix a typo.
6150 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6151 issuing -Wsuggest-attribute for void functions.
6153 2018-01-03 Martin Sebor <msebor@redhat.com>
6155 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6156 offset_int::from instead of wide_int::to_shwi.
6157 (maybe_diag_overlap): Remove assertion.
6158 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6159 * gimple-ssa-sprintf.c (format_directive): Same.
6160 (parse_directive): Same.
6161 (sprintf_dom_walker::compute_format_length): Same.
6162 (try_substitute_return_value): Same.
6164 2018-01-03 Jeff Law <law@redhat.com>
6167 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6168 non-constant residual for zero at runtime and avoid probing in
6169 that case. Reorganize code for trailing problem to mirror handling
6172 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6174 PR tree-optimization/83501
6175 * tree-ssa-strlen.c (get_string_cst): New.
6176 (handle_char_store): Call get_string_cst.
6178 2018-01-03 Martin Liska <mliska@suse.cz>
6180 PR tree-optimization/83593
6181 * tree-ssa-strlen.c: Include tree-cfg.h.
6182 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6183 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6184 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6186 (strlen_dom_walker::before_dom_children): Call
6187 gimple_purge_dead_eh_edges. Dump tranformation with details
6189 (strlen_dom_walker::before_dom_children): Update call by adding
6190 new argument cleanup_eh.
6191 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6193 2018-01-03 Martin Liska <mliska@suse.cz>
6196 * cif-code.def (VARIADIC_THUNK): New enum value.
6197 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6200 2018-01-03 Jan Beulich <jbeulich@suse.com>
6202 * sse.md (mov<mode>_internal): Tighten condition for when to use
6203 vmovdqu<ssescalarsize> for TI and OI modes.
6205 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6207 Update copyright years.
6209 2018-01-03 Martin Liska <mliska@suse.cz>
6212 * ipa-visibility.c (function_and_variable_visibility): Skip
6213 functions with noipa attribure.
6215 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6217 * gcc.c (process_command): Update copyright notice dates.
6218 * gcov-dump.c (print_version): Ditto.
6219 * gcov.c (print_version): Ditto.
6220 * gcov-tool.c (print_version): Ditto.
6221 * gengtype.c (create_file): Ditto.
6222 * doc/cpp.texi: Bump @copying's copyright year.
6223 * doc/cppinternals.texi: Ditto.
6224 * doc/gcc.texi: Ditto.
6225 * doc/gccint.texi: Ditto.
6226 * doc/gcov.texi: Ditto.
6227 * doc/install.texi: Ditto.
6228 * doc/invoke.texi: Ditto.
6230 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6232 * vector-builder.h (vector_builder::m_full_nelts): Change from
6233 unsigned int to poly_uint64.
6234 (vector_builder::full_nelts): Update prototype accordingly.
6235 (vector_builder::new_vector): Likewise.
6236 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6237 (vector_builder::operator ==): Likewise.
6238 (vector_builder::finalize): Likewise.
6239 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6240 Take the number of elements as a poly_uint64 rather than an
6242 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6243 from unsigned int to poly_uint64.
6244 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6245 (vec_perm_indices::new_vector): Likewise.
6246 (vec_perm_indices::length): Likewise.
6247 (vec_perm_indices::nelts_per_input): Likewise.
6248 (vec_perm_indices::input_nelts): Likewise.
6249 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6250 number of elements per input as a poly_uint64 rather than an
6251 unsigned int. Use the original encoding for variable-length
6252 vectors, rather than clamping each individual element.
6253 For the second and subsequent elements in each pattern,
6254 clamp the step and base before clamping their sum.
6255 (vec_perm_indices::series_p): Handle polynomial element counts.
6256 (vec_perm_indices::all_in_range_p): Likewise.
6257 (vec_perm_indices_to_tree): Likewise.
6258 (vec_perm_indices_to_rtx): Likewise.
6259 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6260 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6261 (tree_vector_builder::new_binary_operation): Handle polynomial
6262 element counts. Return false if we need to know the number
6263 of elements at compile time.
6264 * fold-const.c (fold_vec_perm): Punt if the number of elements
6265 isn't known at compile time.
6267 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6269 * vec-perm-indices.h (vec_perm_builder): Change element type
6270 from HOST_WIDE_INT to poly_int64.
6271 (vec_perm_indices::element_type): Update accordingly.
6272 (vec_perm_indices::clamp): Handle polynomial element_types.
6273 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6274 (vec_perm_indices::all_in_range_p): Likewise.
6275 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6277 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6278 polynomial vec_perm_indices element types.
6279 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6280 * fold-const.c (fold_vec_perm): Likewise.
6281 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6282 * tree-vect-generic.c (lower_vec_perm): Likewise.
6283 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6284 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6285 element type to HOST_WIDE_INT.
6287 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6288 Alan Hayward <alan.hayward@arm.com>
6289 David Sherwood <david.sherwood@arm.com>
6291 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6292 rather than an int. Use plus_constant.
6293 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6294 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6296 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6297 Alan Hayward <alan.hayward@arm.com>
6298 David Sherwood <david.sherwood@arm.com>
6300 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6301 a HOST_WIDE_INT to a poly_int64.
6303 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6304 Alan Hayward <alan.hayward@arm.com>
6305 David Sherwood <david.sherwood@arm.com>
6307 * calls.c (load_register_parameters): Cope with polynomial
6308 mode sizes. Require a constant size for BLKmode parameters
6309 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6310 forces a parameter to be padded at the lsb end in order to
6311 fill a complete number of words, require the parameter size
6312 to be ordered wrt UNITS_PER_WORD.
6314 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6315 Alan Hayward <alan.hayward@arm.com>
6316 David Sherwood <david.sherwood@arm.com>
6318 * reload1.c (spill_stack_slot_width): Change element type
6319 from unsigned int to poly_uint64_pod.
6320 (alter_reg): Treat mode sizes as polynomial.
6322 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6323 Alan Hayward <alan.hayward@arm.com>
6324 David Sherwood <david.sherwood@arm.com>
6326 * reload.c (complex_word_subreg_p): New function.
6327 (reload_inner_reg_of_subreg, push_reload): Use it.
6329 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6330 Alan Hayward <alan.hayward@arm.com>
6331 David Sherwood <david.sherwood@arm.com>
6333 * lra-constraints.c (process_alt_operands): Reject matched
6334 operands whose sizes aren't ordered.
6335 (match_reload): Refer to this check here.
6337 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6338 Alan Hayward <alan.hayward@arm.com>
6339 David Sherwood <david.sherwood@arm.com>
6341 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6342 that the mode size is in the set {1, 2, 4, 8, 16}.
6344 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6345 Alan Hayward <alan.hayward@arm.com>
6346 David Sherwood <david.sherwood@arm.com>
6348 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6349 Use plus_constant instead of gen_rtx_PLUS.
6351 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6352 Alan Hayward <alan.hayward@arm.com>
6353 David Sherwood <david.sherwood@arm.com>
6355 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6356 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6357 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6358 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6359 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6360 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6361 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6362 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6363 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6364 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6366 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6367 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6368 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6369 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6370 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6371 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6372 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6373 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6374 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6375 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6377 * expr.c (emit_move_resolve_push): Treat the input and result
6378 of PUSH_ROUNDING as a poly_int64.
6379 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6380 (emit_push_insn): Likewise.
6381 * lra-eliminations.c (mark_not_eliminable): Likewise.
6382 * recog.c (push_operand): Likewise.
6383 * reload1.c (elimination_effects): Likewise.
6384 * rtlanal.c (nonzero_bits1): Likewise.
6385 * calls.c (store_one_arg): Likewise. Require the padding to be
6386 known at compile time.
6388 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6389 Alan Hayward <alan.hayward@arm.com>
6390 David Sherwood <david.sherwood@arm.com>
6392 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6393 Use plus_constant instead of gen_rtx_PLUS.
6395 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6396 Alan Hayward <alan.hayward@arm.com>
6397 David Sherwood <david.sherwood@arm.com>
6399 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6402 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6403 Alan Hayward <alan.hayward@arm.com>
6404 David Sherwood <david.sherwood@arm.com>
6406 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6407 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6408 via stack temporaries. Treat the mode size as polynomial too.
6410 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6411 Alan Hayward <alan.hayward@arm.com>
6412 David Sherwood <david.sherwood@arm.com>
6414 * expr.c (expand_expr_real_2): When handling conversions involving
6415 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6416 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6417 as a poly_uint64 too.
6419 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6420 Alan Hayward <alan.hayward@arm.com>
6421 David Sherwood <david.sherwood@arm.com>
6423 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6425 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6426 Alan Hayward <alan.hayward@arm.com>
6427 David Sherwood <david.sherwood@arm.com>
6429 * combine.c (can_change_dest_mode): Handle polynomial
6430 REGMODE_NATURAL_SIZE.
6431 * expmed.c (store_bit_field_1): Likewise.
6432 * expr.c (store_constructor): Likewise.
6433 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6434 and polynomial REGMODE_NATURAL_SIZE.
6435 (gen_lowpart_common): Likewise.
6436 * reginfo.c (record_subregs_of_mode): Likewise.
6437 * rtlanal.c (read_modify_subreg_p): Likewise.
6439 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6440 Alan Hayward <alan.hayward@arm.com>
6441 David Sherwood <david.sherwood@arm.com>
6443 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6444 numbers of elements.
6446 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6447 Alan Hayward <alan.hayward@arm.com>
6448 David Sherwood <david.sherwood@arm.com>
6450 * match.pd: Cope with polynomial numbers of vector elements.
6452 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6453 Alan Hayward <alan.hayward@arm.com>
6454 David Sherwood <david.sherwood@arm.com>
6456 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6457 in a POINTER_PLUS_EXPR.
6459 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6460 Alan Hayward <alan.hayward@arm.com>
6461 David Sherwood <david.sherwood@arm.com>
6463 * omp-simd-clone.c (simd_clone_subparts): New function.
6464 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6465 (ipa_simd_modify_function_body): Likewise.
6467 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6468 Alan Hayward <alan.hayward@arm.com>
6469 David Sherwood <david.sherwood@arm.com>
6471 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6472 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6473 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6474 (expand_vector_condition, vector_element): Likewise.
6475 (subparts_gt): New function.
6476 (get_compute_type): Use subparts_gt.
6477 (count_type_subparts): Delete.
6478 (expand_vector_operations_1): Use subparts_gt instead of
6479 count_type_subparts.
6481 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6482 Alan Hayward <alan.hayward@arm.com>
6483 David Sherwood <david.sherwood@arm.com>
6485 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6486 (vect_compile_time_alias): ...this new function. Do the calculation
6487 on poly_ints rather than trees.
6488 (vect_prune_runtime_alias_test_list): Update call accordingly.
6490 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6491 Alan Hayward <alan.hayward@arm.com>
6492 David Sherwood <david.sherwood@arm.com>
6494 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6496 (vect_schedule_slp_instance): Likewise.
6498 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6499 Alan Hayward <alan.hayward@arm.com>
6500 David Sherwood <david.sherwood@arm.com>
6502 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6503 constant and extern definitions for variable-length vectors.
6504 (vect_get_constant_vectors): Note that the number of units
6505 is known to be constant.
6507 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6508 Alan Hayward <alan.hayward@arm.com>
6509 David Sherwood <david.sherwood@arm.com>
6511 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6512 of units as polynomial. Choose between WIDE and NARROW based
6515 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6516 Alan Hayward <alan.hayward@arm.com>
6517 David Sherwood <david.sherwood@arm.com>
6519 * tree-vect-stmts.c (simd_clone_subparts): New function.
6520 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6522 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6523 Alan Hayward <alan.hayward@arm.com>
6524 David Sherwood <david.sherwood@arm.com>
6526 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6527 vectors as polynomial. Use build_index_vector for
6530 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6531 Alan Hayward <alan.hayward@arm.com>
6532 David Sherwood <david.sherwood@arm.com>
6534 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6535 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6536 for variable-length vectors.
6537 (vectorizable_mask_load_store): Treat the number of units as
6538 polynomial, asserting that it is constant if the condition has
6539 already been enforced.
6540 (vectorizable_store, vectorizable_load): Likewise.
6542 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6543 Alan Hayward <alan.hayward@arm.com>
6544 David Sherwood <david.sherwood@arm.com>
6546 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6547 of units as polynomial. Punt if we can't tell at compile time
6548 which vector contains the final result.
6550 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6551 Alan Hayward <alan.hayward@arm.com>
6552 David Sherwood <david.sherwood@arm.com>
6554 * tree-vect-loop.c (vectorizable_induction): Treat the number
6555 of units as polynomial. Punt on SLP inductions. Use an integer
6556 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6557 cast of such a series for variable-length floating-point
6560 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6561 Alan Hayward <alan.hayward@arm.com>
6562 David Sherwood <david.sherwood@arm.com>
6564 * tree.h (build_index_vector): Declare.
6565 * tree.c (build_index_vector): New function.
6566 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6567 of units as polynomial, forcibly converting it to a constant if
6568 vectorizable_reduction has already enforced the condition.
6569 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6570 to create a {1,2,3,...} vector.
6571 (vectorizable_reduction): Treat the number of units as polynomial.
6572 Choose vectype_in based on the largest scalar element size rather
6573 than the smallest number of units. Enforce the restrictions
6576 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6577 Alan Hayward <alan.hayward@arm.com>
6578 David Sherwood <david.sherwood@arm.com>
6580 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6581 number of units as polynomial.
6583 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6584 Alan Hayward <alan.hayward@arm.com>
6585 David Sherwood <david.sherwood@arm.com>
6587 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6588 * target.def (autovectorize_vector_sizes): Return the vector sizes
6589 by pointer, using vector_sizes rather than a bitmask.
6590 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6591 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6592 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6594 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6595 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6596 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6597 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6598 * omp-general.c (omp_max_vf): Likewise.
6599 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6600 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6601 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6602 * tree-vect-slp.c (vect_slp_bb): Likewise.
6603 * doc/tm.texi: Regenerate.
6604 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6606 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6607 the vector size as a poly_uint64 rather than an unsigned int.
6608 (current_vector_size): Change from an unsigned int to a poly_uint64.
6609 (get_vectype_for_scalar_type): Update accordingly.
6610 * tree.h (build_truth_vector_type): Take the size and number of
6611 units as a poly_uint64 rather than an unsigned int.
6612 (build_vector_type): Add a temporary overload that takes
6613 the number of units as a poly_uint64 rather than an unsigned int.
6614 * tree.c (make_vector_type): Likewise.
6615 (build_truth_vector_type): Take the number of units as a poly_uint64
6616 rather than an unsigned int.
6618 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6619 Alan Hayward <alan.hayward@arm.com>
6620 David Sherwood <david.sherwood@arm.com>
6622 * target.def (get_mask_mode): Take the number of units and length
6623 as poly_uint64s rather than unsigned ints.
6624 * targhooks.h (default_get_mask_mode): Update accordingly.
6625 * targhooks.c (default_get_mask_mode): Likewise.
6626 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6627 * doc/tm.texi: Regenerate.
6629 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6630 Alan Hayward <alan.hayward@arm.com>
6631 David Sherwood <david.sherwood@arm.com>
6633 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6634 * omp-general.c (omp_max_vf): Likewise.
6635 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6636 (expand_omp_simd): Handle polynomial safelen.
6637 * omp-low.c (omplow_simd_context): Add a default constructor.
6638 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6639 (lower_rec_simd_input_clauses): Update accordingly.
6640 (lower_rec_input_clauses): Likewise.
6642 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6643 Alan Hayward <alan.hayward@arm.com>
6644 David Sherwood <david.sherwood@arm.com>
6646 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6647 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6648 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6649 (vect_analyze_slp_cost): Likewise.
6650 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6651 (vect_model_load_cost): Likewise.
6653 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6654 Alan Hayward <alan.hayward@arm.com>
6655 David Sherwood <david.sherwood@arm.com>
6657 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6658 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6659 from an unsigned int * to a poly_uint64_pod *.
6660 (calculate_unrolling_factor): New function.
6661 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6663 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6664 Alan Hayward <alan.hayward@arm.com>
6665 David Sherwood <david.sherwood@arm.com>
6667 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6668 from an unsigned int to a poly_uint64.
6669 (_loop_vec_info::slp_unrolling_factor): Likewise.
6670 (_loop_vec_info::vectorization_factor): Change from an int
6672 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6673 (vect_get_num_vectors): New function.
6674 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6675 (vect_get_num_copies): Use vect_get_num_vectors.
6676 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6677 to an unsigned int *.
6678 (vect_analyze_data_refs): Change min_vf from an int * to a
6680 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6681 than an unsigned HOST_WIDE_INT.
6682 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6683 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6684 to an unsigned int *.
6685 (vect_analyze_data_ref_dependences): Likewise.
6686 (vect_compute_data_ref_alignment): Handle polynomial vf.
6687 (vect_enhance_data_refs_alignment): Likewise.
6688 (vect_prune_runtime_alias_test_list): Likewise.
6689 (vect_shift_permute_load_chain): Likewise.
6690 (vect_supportable_dr_alignment): Likewise.
6691 (dependence_distance_ge_vf): Take the vectorization factor as a
6692 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6693 (vect_analyze_data_refs): Change min_vf from an int * to a
6695 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6696 vfm1 as a poly_uint64 rather than an int. Make the same change
6697 for the returned bound_scalar.
6698 (vect_gen_vector_loop_niters): Handle polynomial vf.
6699 (vect_do_peeling): Likewise. Update call to
6700 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6701 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6703 * tree-vect-loop.c (vect_determine_vectorization_factor)
6704 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6705 (vect_get_known_peeling_cost): Likewise.
6706 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6707 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6708 (vect_transform_loop): Likewise. Use the lowest possible VF when
6709 updating the upper bounds of the loop.
6710 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6712 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6713 polynomial unroll factors.
6714 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6715 (vect_make_slp_decision): Likewise.
6716 (vect_supported_load_permutation_p): Likewise, and polynomial
6718 (vect_analyze_slp_cost): Handle polynomial vf.
6719 (vect_slp_analyze_node_operations): Likewise.
6720 (vect_slp_analyze_bb_1): Likewise.
6721 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6722 than an unsigned HOST_WIDE_INT.
6723 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6724 (vectorizable_load): Handle polynomial vf.
6725 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6727 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6729 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6730 Alan Hayward <alan.hayward@arm.com>
6731 David Sherwood <david.sherwood@arm.com>
6733 * match.pd: Handle bit operations involving three constants
6734 and try to fold one pair.
6736 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6738 * tree-vect-loop-manip.c: Include gimple-fold.h.
6739 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6740 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6741 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6742 Add a path that uses a step of VF instead of 1, but disable it
6744 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6745 and niters_no_overflow parameters. Update calls to
6746 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6747 Create a new SSA name if the latter choses to use a ste other
6748 than zero, and return it via niters_vector_mult_vf_var.
6749 * tree-vect-loop.c (vect_transform_loop): Update calls to
6750 vect_do_peeling, vect_gen_vector_loop_niters and
6751 slpeel_make_loop_iterate_ntimes.
6752 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6753 (vect_gen_vector_loop_niters): Update declarations after above changes.
6755 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6757 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6758 128-bit round to integer instructions.
6759 (ceil<mode>2): Likewise.
6760 (btrunc<mode>2): Likewise.
6761 (round<mode>2): Likewise.
6763 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6765 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6766 unaligned VSX load/store on P8/P9.
6767 (expand_block_clear): Allow the use of unaligned VSX
6768 load/store on P8/P9.
6770 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6772 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6774 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6775 swap associated with both a load and a store.
6777 2018-01-02 Andrew Waterman <andrew@sifive.com>
6779 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6780 * config/riscv/riscv.md (clear_cache): Use it.
6782 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6784 * web.c: Remove out-of-date comment.
6786 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6788 * expr.c (fixup_args_size_notes): Check that any existing
6789 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6790 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6791 (emit_single_push_insn): ...here.
6793 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6795 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6796 (const_vector_encoded_nelts): New function.
6797 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6798 (const_vector_int_elt, const_vector_elt): Declare.
6799 * emit-rtl.c (const_vector_int_elt_1): New function.
6800 (const_vector_elt): Likewise.
6801 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6802 of CONST_VECTOR_ELT.
6804 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6806 * expr.c: Include rtx-vector-builder.h.
6807 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6808 directly on the tree encoding.
6809 (const_vector_from_tree): Likewise.
6810 * optabs.c: Include rtx-vector-builder.h.
6811 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6812 sequence of "u" values.
6813 * vec-perm-indices.c: Include rtx-vector-builder.h.
6814 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6815 directly on the vec_perm_indices encoding.
6817 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6819 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6820 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6821 * rtx-vector-builder.h: New file.
6822 * rtx-vector-builder.c: Likewise.
6823 * rtl.h (rtx_def::u2): Add a const_vector field.
6824 (CONST_VECTOR_NPATTERNS): New macro.
6825 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6826 (CONST_VECTOR_DUPLICATE_P): Likewise.
6827 (CONST_VECTOR_STEPPED_P): Likewise.
6828 (CONST_VECTOR_ENCODED_ELT): Likewise.
6829 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6830 (unwrap_const_vec_duplicate): Likewise.
6831 (const_vec_series_p): Check for a non-duplicated vector encoding.
6832 Say that the function only returns true for integer vectors.
6833 * emit-rtl.c: Include rtx-vector-builder.h.
6834 (gen_const_vec_duplicate_1): Delete.
6835 (gen_const_vector): Call gen_const_vec_duplicate instead of
6836 gen_const_vec_duplicate_1.
6837 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6838 (gen_const_vec_duplicate): Use rtx_vector_builder.
6839 (gen_const_vec_series): Likewise.
6840 (gen_rtx_CONST_VECTOR): Likewise.
6841 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6842 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6843 Build a new vector rather than modifying a CONST_VECTOR in-place.
6844 (handle_special_swappables): Update call accordingly.
6845 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6846 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6847 Build a new vector rather than modifying a CONST_VECTOR in-place.
6848 (handle_special_swappables): Update call accordingly.
6850 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6852 * simplify-rtx.c (simplify_const_binary_operation): Use
6853 CONST_VECTOR_ELT instead of XVECEXP.
6855 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6857 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6858 the selector elements to be different from the data elements
6859 if the selector is a VECTOR_CST.
6860 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6861 ssizetype for the selector.
6863 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6865 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6866 before testing each element individually.
6867 * tree-vect-generic.c (lower_vec_perm): Likewise.
6869 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6871 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6872 * selftest-run-tests.c (selftest::run_tests): Call it.
6873 * vector-builder.h (vector_builder::operator ==): New function.
6874 (vector_builder::operator !=): Likewise.
6875 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6876 (vec_perm_indices::all_from_input_p): New function.
6877 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6878 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6879 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6880 instead of reading the VECTOR_CST directly. Detect whether both
6881 vector inputs are the same before constructing the vec_perm_indices,
6882 and update the number of inputs argument accordingly. Use the
6883 utility functions added above. Only construct sel2 if we need to.
6885 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6887 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6888 the broadcast of the low byte.
6889 (expand_mult_highpart): Use an explicit encoding for the permutes.
6890 * optabs-query.c (can_mult_highpart_p): Likewise.
6891 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6892 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6893 (vectorizable_bswap): Likewise.
6894 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6895 explicit encoding for the power-of-2 permutes.
6896 (vect_permute_store_chain): Likewise.
6897 (vect_grouped_load_supported): Likewise.
6898 (vect_permute_load_chain): Likewise.
6900 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6902 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6903 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6904 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6905 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6906 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6907 (vect_gen_perm_mask_any): Likewise.
6909 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6911 * int-vector-builder.h: New file.
6912 * vec-perm-indices.h: Include int-vector-builder.h.
6913 (vec_perm_indices): Redefine as an int_vector_builder.
6914 (auto_vec_perm_indices): Delete.
6915 (vec_perm_builder): Redefine as a stand-alone class.
6916 (vec_perm_indices::vec_perm_indices): New function.
6917 (vec_perm_indices::clamp): Likewise.
6918 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6919 (vec_perm_indices::new_vector): New function.
6920 (vec_perm_indices::new_expanded_vector): Update for new
6921 vec_perm_indices class.
6922 (vec_perm_indices::rotate_inputs): New function.
6923 (vec_perm_indices::all_in_range_p): Operate directly on the
6924 encoded form, without computing elided elements.
6925 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6926 encoding. Update for new vec_perm_indices class.
6927 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6928 the given vec_perm_builder.
6929 (expand_vec_perm_var): Update vec_perm_builder constructor.
6930 (expand_mult_highpart): Use vec_perm_builder instead of
6931 auto_vec_perm_indices.
6932 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6933 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6934 or double series encoding as appropriate.
6935 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6936 vec_perm_indices instead of auto_vec_perm_indices.
6937 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6938 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6939 (vect_permute_store_chain): Likewise.
6940 (vect_grouped_load_supported): Likewise.
6941 (vect_permute_load_chain): Likewise.
6942 (vect_shift_permute_load_chain): Likewise.
6943 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6944 (vect_transform_slp_perm_load): Likewise.
6945 (vect_schedule_slp_instance): Likewise.
6946 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6947 (vectorizable_mask_load_store): Likewise.
6948 (vectorizable_bswap): Likewise.
6949 (vectorizable_store): Likewise.
6950 (vectorizable_load): Likewise.
6951 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6952 vec_perm_indices instead of auto_vec_perm_indices. Use
6953 tree_to_vec_perm_builder to read the vector from a tree.
6954 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6955 vec_perm_builder instead of a vec_perm_indices.
6956 (have_whole_vector_shift): Use vec_perm_builder and
6957 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6958 truncation to calc_vec_perm_mask_for_shift.
6959 (vect_create_epilog_for_reduction): Likewise.
6960 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6961 from auto_vec_perm_indices to vec_perm_indices.
6962 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6963 instead of changing individual elements.
6964 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6965 the vector in d.perm.
6966 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6967 from auto_vec_perm_indices to vec_perm_indices.
6968 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6969 instead of changing individual elements.
6970 (arm_vectorize_vec_perm_const): Use new_vector to install
6971 the vector in d.perm.
6972 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6973 Update vec_perm_builder constructor.
6974 (rs6000_expand_interleave): Likewise.
6975 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6976 (rs6000_expand_interleave): Likewise.
6978 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6980 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6981 to qimode could truncate the indices.
6982 * optabs.c (expand_vec_perm_var): Likewise.
6984 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6986 * Makefile.in (OBJS): Add vec-perm-indices.o.
6987 * vec-perm-indices.h: New file.
6988 * vec-perm-indices.c: Likewise.
6989 * target.h (vec_perm_indices): Replace with a forward class
6991 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6992 * optabs.h: Include vec-perm-indices.h.
6993 (expand_vec_perm): Delete.
6994 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6995 (expand_vec_perm_const): Declare.
6996 * target.def (vec_perm_const_ok): Replace with...
6997 (vec_perm_const): ...this new hook.
6998 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6999 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
7000 * doc/tm.texi: Regenerate.
7001 * optabs.def (vec_perm_const): Delete.
7002 * doc/md.texi (vec_perm_const): Likewise.
7003 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
7004 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7005 expand_vec_perm for constant permutation vectors. Assert that
7006 the mode of variable permutation vectors is the integer equivalent
7007 of the mode that is being permuted.
7008 * optabs-query.h (selector_fits_mode_p): Declare.
7009 * optabs-query.c: Include vec-perm-indices.h.
7010 (selector_fits_mode_p): New function.
7011 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7012 is defined, instead of checking whether the vec_perm_const_optab
7013 exists. Use targetm.vectorize.vec_perm_const instead of
7014 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7015 fit in the vector mode before using a variable permute.
7016 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7017 vec_perm_indices instead of an rtx.
7018 (expand_vec_perm): Replace with...
7019 (expand_vec_perm_const): ...this new function. Take the selector
7020 as a vec_perm_indices rather than an rtx. Also take the mode of
7021 the selector. Update call to shift_amt_for_vec_perm_mask.
7022 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7023 Use vec_perm_indices::new_expanded_vector to expand the original
7024 selector into bytes. Check whether the indices fit in the vector
7025 mode before using a variable permute.
7026 (expand_vec_perm_var): Make global.
7027 (expand_mult_highpart): Use expand_vec_perm_const.
7028 * fold-const.c: Includes vec-perm-indices.h.
7029 * tree-ssa-forwprop.c: Likewise.
7030 * tree-vect-data-refs.c: Likewise.
7031 * tree-vect-generic.c: Likewise.
7032 * tree-vect-loop.c: Likewise.
7033 * tree-vect-slp.c: Likewise.
7034 * tree-vect-stmts.c: Likewise.
7035 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7037 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7038 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7039 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7040 (aarch64_vectorize_vec_perm_const): ...this new function.
7041 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7042 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7043 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7044 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7045 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7046 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7047 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7049 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7050 check for NEON modes.
7051 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7052 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7053 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7054 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7056 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7057 the old VEC_PERM_CONST conditions.
7058 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7059 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7060 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7061 (ia64_vectorize_vec_perm_const_ok): Merge into...
7062 (ia64_vectorize_vec_perm_const): ...this new function.
7063 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7064 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7065 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7066 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7067 * config/mips/mips.c (mips_expand_vec_perm_const)
7068 (mips_vectorize_vec_perm_const_ok): Merge into...
7069 (mips_vectorize_vec_perm_const): ...this new function.
7070 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7071 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7072 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7073 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7074 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7075 (rs6000_expand_vec_perm_const): Delete.
7076 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7078 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7079 (altivec_expand_vec_perm_const_le): Take each operand individually.
7080 Operate on constant selectors rather than rtxes.
7081 (altivec_expand_vec_perm_const): Likewise. Update call to
7082 altivec_expand_vec_perm_const_le.
7083 (rs6000_expand_vec_perm_const): Delete.
7084 (rs6000_vectorize_vec_perm_const_ok): Delete.
7085 (rs6000_vectorize_vec_perm_const): New function.
7086 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7087 an element count and rtx array.
7088 (rs6000_expand_extract_even): Update call accordingly.
7089 (rs6000_expand_interleave): Likewise.
7090 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7091 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7092 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7093 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7094 (rs6000_expand_vec_perm_const): Delete.
7095 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7096 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7097 (altivec_expand_vec_perm_const_le): Take each operand individually.
7098 Operate on constant selectors rather than rtxes.
7099 (altivec_expand_vec_perm_const): Likewise. Update call to
7100 altivec_expand_vec_perm_const_le.
7101 (rs6000_expand_vec_perm_const): Delete.
7102 (rs6000_vectorize_vec_perm_const_ok): Delete.
7103 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7104 reference to the SPE evmerge intructions.
7105 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7106 an element count and rtx array.
7107 (rs6000_expand_extract_even): Update call accordingly.
7108 (rs6000_expand_interleave): Likewise.
7109 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7110 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7112 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7114 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7116 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7117 vector mode and that that mode matches the mode of the data
7119 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7120 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7121 directly using expand_vec_perm_1 when forcing selectors into
7123 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7125 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7127 * optabs-query.h (can_vec_perm_p): Delete.
7128 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7129 * optabs-query.c (can_vec_perm_p): Split into...
7130 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7131 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7132 particular selector is valid.
7133 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7134 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7135 (vect_grouped_load_supported): Likewise.
7136 (vect_shift_permute_load_chain): Likewise.
7137 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7138 (vect_transform_slp_perm_load): Likewise.
7139 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7140 (vectorizable_bswap): Likewise.
7141 (vect_gen_perm_mask_checked): Likewise.
7142 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7143 implementations of variable permutation vectors into account
7144 when deciding which selector to use.
7145 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7146 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7147 with a false third argument.
7148 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7149 to test whether the constant selector is valid and can_vec_perm_var_p
7150 to test whether a variable selector is valid.
7152 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7154 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7155 * optabs-query.c (can_vec_perm_p): Likewise.
7156 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7157 instead of vec_perm_indices.
7158 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7159 (vect_gen_perm_mask_checked): Likewise,
7160 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7161 (vect_gen_perm_mask_checked): Likewise,
7163 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7165 * optabs-query.h (qimode_for_vec_perm): Declare.
7166 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7167 (qimode_for_vec_perm): ...this new function.
7168 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7170 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7172 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7173 does not have a conditional at the top.
7175 2018-01-02 Richard Biener <rguenther@suse.de>
7177 * ipa-inline.c (big_speedup_p): Fix expression.
7179 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7182 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7185 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7189 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7190 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7191 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7192 cond_taken_branch_cost 3->4.
7194 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7196 PR tree-optimization/83581
7197 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7198 TODO_cleanup_cfg if any changes have been made.
7201 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7202 convert_modes if target mode has the right side, but different mode
7206 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7207 last argument when extracting from CONCAT. If either from_real or
7208 from_imag is NULL, use expansion through memory. If result is not
7209 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7210 the parts directly to inner mode, if even that fails, use expansion
7214 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7215 check for bswap in mode rather than HImode and use that in expand_unop
7218 Copyright (C) 2018 Free Software Foundation, Inc.
7220 Copying and distribution of this file, with or without modification,
7221 are permitted in any medium without royalty provided the copyright
7222 notice and this notice are preserved.