| blob | d3c06450004abb810d8b09db1847b517989680b8 |
1 2018-02-16 Nathan Sidwell <nathan@acm.org>
3 * doc/extend.texi (Backwards Compatibility): Mention friend
4 injection. Note for-scope is deprecated.
5 * doc/invoke.texi (-ffriend-injection): Deprecate.
7 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
9 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
10 that moved to I2, also allow destinations that are a paradoxical
11 subreg (instead of a normal reg).
13 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
15 PR target/83831
16 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
17 to QImode.
19 2018-02-16 Richard Biener <rguenther@suse.de>
21 PR tree-optimization/84037
22 PR tree-optimization/84016
23 PR target/82862
24 * config/i386/i386.c (ix86_builtin_vectorization_cost):
25 Adjust vec_construct for the fact we need additional higher latency
26 128bit inserts for AVX256 and AVX512 vector builds.
27 (ix86_add_stmt_cost): Scale vector construction cost for
28 elementwise loads.
30 2018-02-16 Richard Biener <rguenther@suse.de>
32 PR tree-optimization/84417
33 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
34 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
35 (non_rewritable_lvalue_p): Likewise, use poly-ints.
37 2018-02-16 Martin Liska <mliska@suse.cz>
39 PR sanitizer/84307
40 * internal-fn.def (ASAN_CHECK): Set proper flags.
41 (ASAN_MARK): Likewise.
43 2018-02-16 Julia Koval <julia.koval@intel.com>
45 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
46 from PTA_CANNONLAKE.
48 2018-02-16 Jakub Jelinek <jakub@redhat.com>
50 PR target/84272
51 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
52 Use ++iter rather than iter++ for std::list iterators.
53 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
54 defer deleting them until all nodes in the forest are processed. Do
55 free even leaf nodes. Change to_process into auto_vec.
57 PR bootstrap/84405
58 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
59 * vec.h (vec_default_construct): Use memset instead of placement new
60 if BROKEN_VALUE_INITIALIZATION is defined.
61 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
62 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
63 is defined.
65 PR rtl-optimization/83723
66 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
67 * lra.c (lra_substitute_pseudo): Likewise. If true, use
68 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
69 recursive calls.
70 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
71 callers.
72 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
74 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
76 PR rtl-optimization/81443
77 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
78 from inner REGs to paradoxical SUBREGs.
80 2018-02-16 Richard Biener <rguenther@suse.de>
82 PR tree-optimization/84399
83 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
84 For operands we can analyze at their definition make sure we can
85 analyze them at each use as well.
87 2018-02-16 Richard Biener <rguenther@suse.de>
89 PR tree-optimization/84190
90 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
91 volatile accesses if the decl isn't volatile.
93 2018-02-15 Jason Merrill <jason@redhat.com>
95 PR c++/84314 - ICE with templates and fastcall attribute.
96 * attribs.c (build_type_attribute_qual_variant): Don't clobber
97 TYPE_CANONICAL on an existing type.
99 2018-02-15 Jakub Jelinek <jakub@redhat.com>
101 PR tree-optimization/84383
102 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
103 dstoff nor call operand_equal_p if dstbase is NULL.
105 PR tree-optimization/84334
106 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
107 also a CONSTANT_CLASS_P, punt.
109 2018-02-14 Jim Wilson <jimw@sifive.com>
111 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
112 first SMALL_OPERAND check. New local min_second_step. Move assert
113 to where locals are set. Add TARGET_RVC support.
114 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
116 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
118 * doc/invoke.texi: Correct -Wformat-overflow code sample.
120 2018-02-14 Martin Sebor <msebor@redhat.com>
122 PR tree-optimization/83698
123 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
124 arrays constrain the offset range to their bounds.
125 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
126 (builtin_access::overlap): Avoid setting the size of overlap if it's
127 already been set.
128 (maybe_diag_overlap): Also consider arrays when deciding what values
129 of offsets to include in diagnostics.
131 2018-02-14 Martin Sebor <msebor@redhat.com>
133 PR c/84108
134 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
135 that correspond to the kind of a declaration.
137 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
139 PR target/83984
140 * config/pa/pa.md: Load address of PIC label using the linkage table
141 if the label is nonlocal.
143 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
145 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
146 warning message if user requests -maltivec=be.
147 * doc/invoke.texi: Document deprecation of -maltivec=be.
149 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
151 PR target/84220
152 * config/rs6000/rs6000-c.c: Update definitions for
153 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
154 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
156 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
158 PR target/84239
159 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
160 add _get_ssp intrinsics. Remove argument from
161 __builtin_ia32_rdssp[d|q].
162 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
163 * config/i386/i386-builtin.def: Remove argument from
164 __builtin_ia32_rdssp[d|q].
165 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
166 ix86_expand_special_args_builtin for _rdssp[d|q].
167 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
168 Clear register before usage.
169 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
170 Add documentation for new _get_ssp and _inc_ssp intrinsics.
172 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
174 PR tree-optimization/84357
175 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
176 operand 1 of an ARRAY_REF too.
178 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
180 PR target/83831
181 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
182 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
183 declarations.
184 (set_of_reg): New struct.
185 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
186 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
187 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
188 functions.
189 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
190 Split into bitclr, bitset, bitinvert patterns if appropriate.
191 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
192 use rx_fuse_in_memory_bitop.
193 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
194 to named insn, correct maximum insn length.
196 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
198 PR target/79242
199 * machmode.def: Define a complex mode for PARTIAL_INT.
200 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
201 MODE_PARTIAL_INT.
202 * doc/rtl.texi: Document CSPImode.
203 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
204 handling.
205 (msp430_hard_regno_nregs_with_padding): Likewise.
207 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
209 PR target/84279
210 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
212 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
214 PR rtl-optimization/84169
215 * combine.c (try_combine): New variable split_i2i3. Set it to true if
216 we generated a parallel as new i3 and we split that to new i2 and i3
217 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
218 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
219 those to i2, not i1. Partially rewrite this scan code.
221 2018-02-13 Jakub Jelinek <jakub@redhat.com>
223 PR c/82210
224 * stor-layout.c (place_field): For variable length fields, adjust
225 offset_align afterwards not just based on the field's alignment,
226 but also on the size.
228 PR middle-end/84309
229 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
230 of exps and logs in the use_exp2 case.
232 2018-02-13 Jeff Law <law@redhat.com>
234 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
235 entry for "vector".
237 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
238 ARGS as unused.
240 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
242 PR debug/84342
243 PR debug/84319
244 * common.opt (gas-loc-support, gas-locview-support): New.
245 (ginline-points, ginternal-reset-location-views): New.
246 * doc/invoke.texi: Document them. Use @itemx where intended.
247 (gvariable-location-views): Adjust.
248 * target.def (reset_location_view): New.
249 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
250 (TARGET_RESET_LOCATION_VIEW): New.
251 * doc/tm.texi: Rebuilt.
252 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
253 (dwarf2out_default_as_locview_support): New.
254 (output_asm_line_debug_info): Use option variables.
255 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
256 (output_loc_list): Likewise.
257 (add_high_low_attributes): Check option variables.
258 Don't output entry view attribute in strict mode.
259 (gen_inlined_subroutine_die): Check option variables.
260 (dwarf2out_inline_entry): Likewise.
261 (init_sections_and_labels): Likewise.
262 (dwarf2out_early_finish): Likewise.
263 (maybe_reset_location_view): New, from...
264 (dwarf2out_var_location): ... here. Call it.
265 * debug.h (dwarf2out_default_as_loc_support): Declare.
266 (dwarf2out_default_as_locview_support): Declare.
267 * hooks.c (hook_int_rtx_insn_0): New.
268 * hooks.h (hook_int_rtx_insn_0): Declare.
269 * toplev.c (process_options): Take -gas-loc-support and
270 -gas-locview-support from dwarf2out. Enable
271 -gvariable-location-views by default only with locview
272 assembler support. Enable -ginternal-reset-location-views by
273 default only if the target defines the corresponding hook.
274 Enable -ginline-points by default if location views are
275 enabled; force it disabled if statement frontiers are
276 disabled.
277 * tree-inline.c (expand_call_inline): Check option variables.
278 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
280 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
282 PR tree-optimization/84321
283 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
284 handling. Also check whether the anti-range contains any values
285 that satisfy the mask; switch to a VR_RANGE if not.
287 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
289 PR sanitizer/84340
290 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
292 2018-02-13 Martin Jambor <mjambor@suse.cz>
294 PR c++/83990
295 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
296 of call statements, also set location of a load to a temporary.
298 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
300 * config/rl78/rl78.c (add_vector_labels): New function.
301 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
302 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
303 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
304 which checks that no arguments are passed.
305 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
306 * doc/extend.texi: Documentation for the new attribute.
308 2018-02-13 Andreas Schwab <schwab@suse.de>
310 * config/riscv/linux.h (CPP_SPEC): Define.
312 2018-02-13 Jakub Jelinek <jakub@redhat.com>
314 PR target/84335
315 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
316 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
317 OPTION_MASK_ISA_AES as first argument to def_builtin_const
318 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
319 instead of OPTION_MASK_ISA_PCLMUL as first argument to
320 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
321 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
322 temporarily for AES and PCLMUL builtins.
324 PR tree-optimization/84339
325 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
326 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
327 Formatting fixes.
329 PR middle-end/84309
330 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
331 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
332 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
333 inline function.
334 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
335 inline function.
336 * omp-simd-clone.h: New file.
337 * omp-simd-clone.c: Include omp-simd-clone.h.
338 (expand_simd_clones): No longer static.
339 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
340 cgraph.h and omp-simd-clone.h.
341 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
342 (vect_recog_widen_shift_pattern): Formatting fix.
343 (vect_pattern_recog_1): Don't check optab for calls.
345 PR target/84336
346 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
347 operands[2] into a REG before using gen_lowpart on it.
349 2018-02-12 Jeff Law <law@redhat.com>
351 PR target/83760
352 * config/sh/sh.c (find_barrier): Consider a sibling call
353 a barrier as well.
355 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
356 successfully back substituting a reg.
358 2018-02-12 Richard Biener <rguenther@suse.de>
360 PR tree-optimization/84037
361 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
362 parameter, move visited init to caller.
363 (vect_slp_analyze_operations): Separate cost from validity
364 check, initialize visited once for all instances.
365 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
366 for all instances.
367 * tree-vect-stmts.c (vect_model_simple_cost): Make early
368 out an assert.
369 (vect_model_promotion_demotion_cost): Likewise.
370 (vectorizable_bswap): Guard cost modeling with !slp_node
371 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
372 SLP stmts.
373 (vectorizable_call): Likewise.
374 (vectorizable_conversion): Likewise.
375 (vectorizable_assignment): Likewise.
376 (vectorizable_shift): Likewise.
377 (vectorizable_operation): Likewise.
378 (vectorizable_store): Likewise.
379 (vectorizable_load): Likewise.
380 (vectorizable_condition): Likewise.
381 (vectorizable_comparison): Likewise.
383 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
385 PR sanitizer/84307
386 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
387 (ASAN_MARK): Fix fnspec to account for return value, change pointer
388 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
390 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
392 PR middle-end/83665
393 * params.def (inline-min-speedup): Increase from 8 to 15.
394 (max-inline-insns-auto): Decrease from 40 to 30.
395 * ipa-split.c (consider_split): Add some buffer for function to
396 be considered inlining candidate.
397 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
398 default values.
400 2018-02-12 Richard Biener <rguenther@suse.de>
402 PR tree-optimization/84037
403 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
404 matched stmts if we cannot swap the non-matched ones.
406 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
408 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
409 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
410 _mm_maskz_scalef_round_ss): New intrinsics.
411 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
412 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
413 __builtin_ia32_scalefss_round): Remove.
414 (__builtin_ia32_scalefsd_mask_round,
415 __builtin_ia32_scalefss_mask_round): New intrinsics.
416 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
417 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
418 ((match_operand:VF_128 2 "<round_nimm_predicate>"
419 "<round_constraint>")): Changed to ...
420 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
421 "<round_scalar_constraint>")): ... this.
422 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
423 %0, %1, %2<round_op3>}"): Changed to ...
424 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
425 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
426 %2<round_scalar_mask_op3>}"): ... this.
427 * config/i386/subst.md (round_scalar_nimm_predicate): New.
429 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
431 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
432 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
433 (_mm_maskz_sqrt_round_ss): New intrinsics.
434 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
435 (__builtin_ia32_sqrtsd_mask_round)
436 (__builtin_ia32_sqrtss_mask_round): New builtins.
437 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
438 (__builtin_ia32_sqrtss_round): Remove.
439 (__builtin_ia32_sqrtsd_mask_round)
440 (__builtin_ia32_sqrtss_mask_round): New builtins.
441 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
442 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
443 ((match_operand:VF_128 1 "vector_operand"
444 "xBm,<round_constraint>")): Changed to ...
445 ((match_operand:VF_128 1 "vector_operand"
446 "xBm,<round_scalar_constraint>")): ... this.
447 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
448 %0, %2, %<iptr>1<round_op3>}): Changed to ...
449 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
450 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
451 %<iptr>1<round_scalar_mask_op3>}): ... this.
452 ((set_attr "prefix" "<round_prefix>")): Changed to ...
453 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
455 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
457 PR target/84266
458 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
459 Cast vec_cmpeq result to correct type.
460 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
461 Cast vec_cmpgt result to correct type.
463 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
465 * final.c (final_scan_insn_1): Renamed from...
466 (final_scan_insn): ... this. New wrapper, to recover
467 seen from the outermost call in recursive ones.
468 * config/sparc/sparc.c (output_return): Drop seen from call.
469 (output_sibcall): Likewise.
470 * config/visium/visium.c (output_branch): Likewise.
472 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
474 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
475 function label.
477 2018-02-10 Alan Modra <amodra@gmail.com>
479 PR target/84300
480 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
481 Specify LR as an input.
483 2018-02-10 Jakub Jelinek <jakub@redhat.com>
485 PR sanitizer/83987
486 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
487 remove_member_access_dummy_vars): New functions.
488 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
489 lower_omp_1, execute_lower_omp): Use them.
491 PR rtl-optimization/84308
492 * shrink-wrap.c (spread_components): Release todo vector.
494 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
496 PR rtl-optimization/57193
497 * ira-color.c (struct allocno_color_data): Add member
498 conflict_allocno_hard_prefs.
499 (update_conflict_allocno_hard_prefs): New.
500 (bucket_allocno_compare_func): Add a preference based on
501 conflict_allocno_hard_prefs.
502 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
503 (color_allocnos): Remove a dead code. Initiate
504 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
506 2018-02-09 Jakub Jelinek <jakub@redhat.com>
508 PR target/84226
509 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
510 constraint from =wa to wa. Avoid a subreg on the output operand,
511 instead use a pseudo and subreg it in a move.
512 (p9_xxbrd_<mode>): Changed to ...
513 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
514 (p9_xxbrd_v2df): New expander.
515 (p9_xxbrw_<mode>): Changed to ...
516 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
517 (p9_xxbrw_v4sf): New expander.
519 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
521 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
523 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
525 PR target/83926
526 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
527 multiply in 32-bit mode.
528 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
529 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
530 mode.
532 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
534 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
535 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
536 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
537 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
539 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
541 PR lto/84213
542 * dwarf2out.c (is_trivial_indirect_ref): New function.
543 (dwarf2out_late_global_decl): Do not generate a location
544 attribute for variables that have a non-trivial DECL_VALUE_EXPR
545 and that are not defined in the current unit.
547 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
549 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
550 instead of a libcall for UNORDERED.
552 2018-02-09 Tamar Christina <tamar.christina@arm.com>
554 PR target/82641
555 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
556 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
558 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
560 PR target/PR84295
561 * config/s390/s390.c (s390_set_current_function): Invoke
562 s390_indirect_branch_settings also if fndecl didn't change.
564 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
566 * config/rs6000/rs6000.md (blockage): Set length to zero.
568 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
570 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
572 2018-02-09 Jakub Jelinek <jakub@redhat.com>
574 PR sanitizer/84285
575 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
576 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
577 -static-lib*san.
579 PR debug/84252
580 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
581 PARALLEL incoming that failed vt_get_decl_and_offset check.
583 PR middle-end/84237
584 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
585 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
586 TREE_READONLY bit.
587 (get_variable_section): For decls in named .bss* sections pass true as
588 second argument to bss_initializer_p.
590 2018-02-09 Marek Polacek <polacek@redhat.com>
591 Jakub Jelinek <jakub@redhat.com>
593 PR c++/83659
594 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
595 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
596 Sync some changes from cxx_fold_indirect_ref.
598 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
600 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
601 markers.
602 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
603 (BLOCK_INLINE_ENTRY_LABEL): New.
604 (dwarf2out_var_location): Disregard inline entry markers.
605 (inline_entry_data): New struct.
606 (inline_entry_data_hasher): New hashtable type.
607 (inline_entry_data_hasher::hash): New.
608 (inline_entry_data_hasher::equal): New.
609 (inline_entry_data_table): New variable.
610 (add_high_low_attributes): Add DW_AT_entry_pc and
611 DW_AT_GNU_entry_view attributes if a pending entry is found
612 in inline_entry_data_table. Add old entry_pc attribute only
613 if debug nonbinding markers are disabled.
614 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
615 markers are enabled.
616 (block_within_block_p, dwarf2out_inline_entry): New.
617 (dwarf2out_finish): Check that no entries remained in
618 inline_entry_data_table.
619 * final.c (reemit_insn_block_notes): Handle inline entry notes.
620 (final_scan_insn, notice_source_line): Likewise.
621 (rest_of_clean_state): Skip inline entry markers.
622 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
623 markers.
624 * gimple.c (gimple_build_debug_inline_entry): New.
625 * gimple.h (enum gimple_debug_subcode): Add
626 GIMPLE_DEBUG_INLINE_ENTRY.
627 (gimple_build_debug_inline_entry): Declare.
628 (gimple_debug_inline_entry_p): New.
629 (gimple_debug_nonbind_marker_p): Adjust.
630 * insn-notes.def (INLINE_ENTRY): New.
631 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
632 inline entry marker notes.
633 (print_insn): Likewise.
634 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
635 (INSN_DEBUG_MARKER_KIND): Likewise.
636 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
637 * tree-inline.c (expand_call_inline): Build and insert
638 debug_inline_entry stmt.
639 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
640 inline entry blocks early, if nonbind markers are enabled.
641 (dump_scope_block): Dump fragment info.
642 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
643 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
644 (gimple_build_debug_inline_entry): New.
645 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
646 Enable/disable inline entry points too.
647 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
648 (DEBUG_INSN): Describe inline entry markers.
650 * common.opt (gvariable-location-views): New.
651 (gvariable-location-views=incompat5): New.
652 * config.in: Rebuilt.
653 * configure: Rebuilt.
654 * configure.ac: Test assembler for view support.
655 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
656 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
657 * dwarf2out.c (var_loc_view): New typedef.
658 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
659 (dwarf2out_locviews_in_attribute): New.
660 (dwarf2out_locviews_in_loclist): New.
661 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
662 (enum dw_line_info_opcode): Add LI_adv_address.
663 (struct dw_line_info_table): Add view.
664 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
665 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
666 (zero_view_p): New variable.
667 (ZERO_VIEW_P): New macro.
668 (output_asm_line_debug_info): New.
669 (struct var_loc_node): Add view.
670 (add_AT_view_list, AT_loc_list): New.
671 (add_var_loc_to_decl): Add view param. Test it against last.
672 (new_loc_list): Add view params. Record them.
673 (AT_loc_list_ptr): Handle loc and view lists.
674 (view_list_to_loc_list_val_node): New.
675 (print_dw_val): Handle dw_val_class_view_list.
676 (size_of_die): Likewise.
677 (value_format): Likewise.
678 (loc_list_has_views): New.
679 (gen_llsym): Set vl_symbol too.
680 (maybe_gen_llsym, skip_loc_list_entry): New.
681 (dwarf2out_maybe_output_loclist_view_pair): New.
682 (output_loc_list): Output view list or entries too.
683 (output_view_list_offset): New.
684 (output_die): Handle dw_val_class_view_list.
685 (output_dwarf_version): New.
686 (output_compilation_unit_header): Use it.
687 (output_skeleton_debug_sections): Likewise.
688 (output_rnglists, output_line_info): Likewise.
689 (output_pubnames, output_aranges): Update version comments.
690 (output_one_line_info_table): Output view numbers in asm comments.
691 (dw_loc_list): Determine current endview, pass it to new_loc_list.
692 Call maybe_gen_llsym.
693 (loc_list_from_tree_1): Adjust.
694 (add_AT_location_description): Create view list attribute if
695 needed, check it's absent otherwise.
696 (convert_cfa_to_fb_loc_list): Adjust.
697 (maybe_emit_file): Call output_asm_line_debug_info for test.
698 (dwarf2out_var_location): Reset views as needed. Precompute
699 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
700 attribute. Set view.
701 (new_line_info_table): Reset next view.
702 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
703 (dwarf2out_source_line): Likewise. Output view resets and labels to
704 the assembler, or select appropriate line info opcodes.
705 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
706 (optimize_string_length): Catch it. Adjust.
707 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
708 dw_val_class_view_list, and remove it if no longer needed.
709 (hash_loc_list): Hash view numbers.
710 (loc_list_hasher::equal): Compare them.
711 (optimize_location_lists): Check whether a view list symbol is
712 needed, and whether the locview attribute is present, and
713 whether they match. Remove the locview attribute if no longer
714 needed.
715 (index_location_lists): Call skip_loc_list_entry for test.
716 (dwarf2out_finish): Call output_asm_line_debug_info for test.
717 Use output_dwarf_version.
718 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
719 (struct dw_val_node): Add val_view_list.
720 * final.c (SEEN_NEXT_VIEW): New.
721 (set_next_view_needed): New.
722 (clear_next_view_needed): New.
723 (maybe_output_next_view): New.
724 (final_start_function): Rename to...
725 (final_start_function_1): ... this. Take pointer to FIRST,
726 add SEEN parameter. Emit param bindings in the initial view.
727 (final_start_function): Reintroduce SEEN-less interface.
728 (final): Rename to...
729 (final_1): ... this. Take SEEN parameter. Output final pending
730 next view at the end.
731 (final): Reintroduce seen-less interface.
732 (final_scan_insn): Output pending next view before switching
733 sections or ending a block. Mark the next view as needed when
734 outputting variable locations. Notify debug backend of section
735 changes, and of location view changes.
736 (rest_of_handle_final): Adjust.
737 * toplev.c (process_options): Autodetect value for debug variable
738 location views option. Warn on incompat5 without -gdwarf-5.
739 * doc/invoke.texi (gvariable-location-views): New.
740 (gvariable-location-views=incompat5): New.
741 (gno-variable-location-views): New.
743 2018-02-08 David Malcolm <dmalcolm@redhat.com>
745 PR tree-optimization/84136
746 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
747 that the result of find_edge is non-NULL.
749 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
751 PR target/83008
752 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
753 storing integer register in SImode. Fix cost of 256 and 512
754 byte aligned SSE register store.
756 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
758 * config/i386/i386.c (ix86_multiplication_cost): Fix
759 multiplication cost for TARGET_AVX512DQ.
761 2018-02-08 Marek Polacek <polacek@redhat.com>
763 PR tree-optimization/84238
764 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
765 get_range_strlen.
767 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
769 PR tree-optimization/84265
770 * tree-vect-stmts.c (vectorizable_store): Don't treat
771 VMAT_CONTIGUOUS accesses as grouped.
772 (vectorizable_load): Likewise.
774 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
776 PR tree-optimization/81635
777 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
778 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
779 (test_round_for_mask): New functions.
780 (wide_int_cc_tests): Call test_round_for_mask.
781 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
782 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
783 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
784 range returned by get_range_info.
786 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
788 PR ipa/81360
789 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
790 * symtab.c: Include builtins.h
791 (symtab_node::output_to_lto_symbol_table_p): Move here
792 from lto-streamer-out.c:output_symbol_p.
793 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
794 (output_symbol_p): Move all logic to symtab.c
795 (produce_symtab): Update.
797 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
799 * config/s390/s390-opts.h (enum indirect_branch): Define.
800 * config/s390/s390-protos.h (s390_return_addr_from_memory)
801 (s390_indirect_branch_via_thunk)
802 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
803 (enum s390_indirect_branch_type): Define.
804 * config/s390/s390.c (struct s390_frame_layout, struct
805 machine_function): Remove.
806 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
807 (indirect_branch_table_label_no, indirect_branch_table_name):
808 Define variables.
809 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
810 (enum s390_indirect_branch_option): Define.
811 (s390_return_addr_from_memory): New function.
812 (s390_handle_string_attribute): New function.
813 (s390_attribute_table): Add new attribute handler.
814 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
815 (s390_indirect_branch_via_thunk): New function.
816 (s390_indirect_branch_via_inline_thunk): New function.
817 (s390_function_ok_for_sibcall): When jumping via thunk disallow
818 sibling call optimization for non z10 compiles.
819 (s390_emit_call): Force indirect branch target to be a single
820 register. Add r1 clobber for non-z10 compiles.
821 (s390_emit_epilogue): Emit return jump via return_use expander.
822 (s390_reorg): Handle JUMP_INSNs as execute targets.
823 (s390_option_override_internal): Perform validity checks for the
824 new command line options.
825 (s390_indirect_branch_attrvalue): New function.
826 (s390_indirect_branch_settings): New function.
827 (s390_set_current_function): Invoke s390_indirect_branch_settings.
828 (s390_output_indirect_thunk_function): New function.
829 (s390_code_end): Implement target hook.
830 (s390_case_values_threshold): Implement target hook.
831 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
832 macros.
833 * config/s390/s390.h (struct s390_frame_layout)
834 (struct machine_function): Move here from s390.c.
835 (TARGET_INDIRECT_BRANCH_NOBP_RET)
836 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
837 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
838 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
839 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
840 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
841 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
842 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
843 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
844 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
845 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
846 (mnemonic attribute): Add values which aren't recognized
847 automatically.
848 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
849 pattern for branch conversion. Fix mnemonic attribute.
850 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
851 indirect branch via thunk if requested.
852 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
853 ("*indirect_jump"): Disable for branch conversion using out of
854 line thunks.
855 ("indirect_jump_via_thunk<mode>_z10")
856 ("indirect_jump_via_thunk<mode>")
857 ("indirect_jump_via_inlinethunk<mode>_z10")
858 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
859 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
860 ("casesi_jump_via_inlinethunk<mode>_z10")
861 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
862 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
863 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
864 ("*indirect2_jump"): Disable for branch conversion.
865 ("casesi_jump"): Turn into expander and expand patterns for branch
866 conversion.
867 ("return_use"): New expander.
868 ("*return"): Emit return via thunk and rename it to ...
869 ("*return<mode>"): ... this one.
870 * config/s390/s390.opt: Add new options and and enum for the
871 option values.
873 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
875 * lra-constraints.c (match_reload): Unconditionally use
876 gen_lowpart_SUBREG, rather than selecting between that
877 and equivalent gen_rtx_SUBREG code.
879 2018-02-08 Richard Biener <rguenther@suse.de>
881 PR tree-optimization/84233
882 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
883 changed flag instead of boguously re-using phi_inserted.
885 2018-02-08 Martin Jambor <mjambor@suse.cz>
887 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
888 static local variables.
890 2018-02-08 Richard Biener <rguenther@suse.de>
892 PR tree-optimization/84278
893 * tree-vect-stmts.c (vectorizable_store): When looking for
894 smaller vector types to perform grouped strided loads/stores
895 make sure the mode is supported by the target.
896 (vectorizable_load): Likewise.
898 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
900 * config/aarch64/aarch64.c (aarch64_components_for_bb):
901 Increase LDP/STP opportunities by adding adjacent callee-saves.
903 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
905 PR rtl-optimization/84068
906 PR rtl-optimization/83459
907 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
909 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
911 PR tree-optimization/84224
912 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
913 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
914 non-zero arguments.
916 2018-02-07 Iain Sandoe <iain@codesourcery.com>
918 PR target/84113
919 * config/rs6000/altivec.md (*restore_world): Remove LR use.
920 * config/rs6000/predicates.md (restore_world_operation): Adjust op
921 count, remove one USE.
923 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
925 * doc/install.texi (Configuration): Document the
926 --with-long-double-format={ibm,ieee} PowerPC configuration
927 options.
929 PR target/84154
930 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
931 Convert from define_expand to be define_insn_and_split. Rework
932 float/double/_Float128 conversions to QI/HI/SImode to work with
933 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
934 conversions to QI/HImode types did a store and then a load to
935 truncate the value. For conversions to VSX registers, don't split
936 the insn, instead emit the code directly. Use the code iterator
937 any_fix to combine signed and unsigned conversions.
938 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
939 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
940 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
941 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
942 (fix_<mode>di2_hw): Likewise.
943 (fixuns_<mode>di2_hw): Likewise.
944 (fix_<mode>si2_hw): Likewise.
945 (fixuns_<mode>si2_hw): Likewise.
946 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
947 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
948 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
949 fix<uns>_trunc<SFDF:mode>si2_p8.
950 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
951 used.
952 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
953 (fix<uns>_<mode>_mem): Likewise.
954 (fctiw<u>z_<mode>_mem): Likewise.
955 (fix<uns>_<mode>_mem): Likewise.
956 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
957 the register allocator from doing a direct move to the GPRs to do
958 a store, and instead use the ISA 3.0 store byte/half-word from
959 vector register instruction. For IEEE 128-bit floating point,
960 also optimize stores of 32-bit ints.
961 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
963 2018-02-07 Alan Hayward <alan.hayward@arm.com>
965 * genextract.c (push_pathstr_operand): New function to support
966 [a-zA-Z].
967 (walk_rtx): Call push_pathstr_operand.
968 (print_path): Support [a-zA-Z].
970 2018-02-07 Richard Biener <rguenther@suse.de>
972 PR tree-optimization/84037
973 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
974 (cse_and_gimplify_to_preheader): Declare.
975 (vect_get_place_in_interleaving_chain): Likewise.
976 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
977 ivexpr_map.
978 (_loop_vec_info::~_loop_vec_info): Delete it.
979 (cse_and_gimplify_to_preheader): New function.
980 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
981 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
982 (vectorizable_load): Likewise. For grouped stores always base
983 the IV on the first element.
984 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
985 condition before gimplifying.
987 2018-02-07 Jakub Jelinek <jakub@redhat.com>
989 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
990 *DIV_EXPR and *MOD_EXPR.
992 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
994 PR target/84248
995 * config/i386/i386.c (ix86_option_override_internal): Mask out
996 the CF_SET bit when checking -fcf-protection.
998 2018-02-07 Tom de Vries <tom@codesourcery.com>
1000 PR libgomp/84217
1001 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1002 enough.
1004 2018-02-07 Richard Biener <rguenther@suse.de>
1006 PR tree-optimization/84204
1007 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1008 this place.
1010 PR tree-optimization/84205
1011 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1012 special-case isl_ast_op_zdiv_r.
1014 PR tree-optimization/84223
1015 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1016 Only add conditions from within the region.
1017 (gather_bbs::after_dom_children): Adjust.
1019 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1021 PR target/84209
1022 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1023 * config/avr/avr.md: Only post-reload split REG-REG moves if
1024 either register is REGERAL_REG_P.
1026 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1028 PR tree-optimization/84235
1029 * tree-ssa-scopedtables.c
1030 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1031 if the subtraction is performed in floating point type where NaNs are
1032 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1033 build 1. Formatting fix.
1035 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1037 PR target/84146
1038 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1039 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1040 and skip it regardless of bb boundaries. Use CALL_P macro,
1041 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1042 unnecessarily, formatting fix.
1044 2018-02-06 Michael Collison <michael.collison@arm.com>
1046 * config/arm/thumb2.md:
1047 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1048 (*thumb_mov_notscc): Ditto.
1050 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1052 PR target/84154
1053 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1054 unsigned_fix, not "s".
1056 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1058 * configure.ac (gcc_fn_eh_frame_ro): New function.
1059 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1060 correct .eh_frame permissions.
1061 * configure: Regenerate.
1063 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1065 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1066 irrelevant options.
1068 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1070 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1071 Display warning message for -mno-speculate-indirect-jumps.
1073 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1075 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1076 Undocumented.
1077 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1079 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1081 PR tree-optimization/84225
1082 * tree-eh.c (find_trapping_overflow): Only call
1083 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1085 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1087 PR target/84145
1088 * config/i386/i386.c: Reimplement the check of possible options
1089 -mibt/-mshstk conbination. Change error messages.
1090 * doc/invoke.texi: Fix a typo: remove extra '='.
1092 2018-02-06 Marek Polacek <polacek@redhat.com>
1094 PR tree-optimization/84228
1095 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1097 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1099 PR target/82641
1100 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1101 emitted arch directives.
1102 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1103 __ARM_FEATURE_COPROC before changing architectures.
1105 2018-02-06 Richard Biener <rguenther@suse.de>
1107 * config/i386/i386.c (print_reg): Fix typo.
1108 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1110 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1112 * configure: Regenerate.
1114 2018-02-05 Martin Sebor <msebor@redhat.com>
1116 PR tree-optimization/83369
1117 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1118 inlining context.
1120 2018-02-05 Martin Liska <mliska@suse.cz>
1122 * doc/invoke.texi: Cherry-pick upstream r323995.
1124 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1126 * ira.c (ira_init_register_move_cost): Adjust comment.
1128 2018-02-05 Martin Liska <mliska@suse.cz>
1130 PR gcov-profile/84137
1131 * doc/gcov.texi: Fix typo in documentation.
1133 2018-02-05 Martin Liska <mliska@suse.cz>
1135 PR gcov-profile/83879
1136 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1137 using dlopen functionality.
1139 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1141 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1142 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1143 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1144 _mm_maskz_range_round_ss): New intrinsics.
1145 (__builtin_ia32_rangesd128_round)
1146 (__builtin_ia32_rangess128_round): Remove.
1147 (__builtin_ia32_rangesd128_mask_round,
1148 __builtin_ia32_rangess128_mask_round): New builtins.
1149 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1150 __builtin_ia32_rangess128_round): Remove.
1151 (__builtin_ia32_rangesd128_mask_round,
1152 __builtin_ia32_rangess128_mask_round): New builtins.
1153 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1154 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1155 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1156 "<round_saeonly_constraint>")): Changed to ...
1157 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1158 "<round_saeonly_scalar_constraint>")): ... this.
1159 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1160 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1161 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1162 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1163 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1165 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1167 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1168 options.
1169 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1170 Remove all values except native, 8540 and 8548.
1172 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1174 * config/i386/i386.c (ix86_output_function_return): Pass
1175 INVALID_REGNUM, instead of -1, as invalid register number to
1176 indirect_thunk_name and output_indirect_thunk.
1178 2018-02-02 Julia Koval <julia.koval@intel.com>
1180 * config.gcc: Add -march=icelake.
1181 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1182 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1183 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1184 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1185 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1186 (processor_target_table): Add icelake.
1187 (ix86_option_override_internal): Handle new PTAs.
1188 (get_builtin_code_for_version): Handle icelake.
1189 (M_INTEL_COREI7_ICELAKE): New.
1190 (fold_builtin_cpu): Handle icelake.
1191 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1192 * doc/invoke.texi: Add -march=icelake.
1194 2018-02-02 Julia Koval <julia.koval@intel.com>
1196 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1197 to wide_int_bitmask.
1198 * wide-int-bitmask.h: New.
1200 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1202 PR target/84066
1203 * config/i386/i386.md: Replace Pmode with word_mode in
1204 builtin_setjmp_setup and builtin_longjmp to support x32.
1206 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1208 PR target/56010
1209 PR target/83743
1210 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1211 #include "opts.h".
1212 (rs6000_supported_cpu_names): New static variable.
1213 (linux_cpu_translation_table): Likewise.
1214 (elf_platform) <cpu>: Define new static variable and use it.
1215 Translate kernel AT_PLATFORM name to canonical name if needed.
1216 Error if platform name is unknown.
1218 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1220 PR target/84089
1221 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1223 2018-02-01 Jeff Law <law@redhat.com>
1225 PR target/84128
1226 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1227 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1228 the scratch if RELEASE_VIA_POP is false.
1229 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1230 If we have to save a temporary register, decrement SIZE appropriately.
1231 Pass new arguments to release_scratch_register_on_entry.
1232 (ix86_adjust_stack_and_probe): Likewise.
1233 (ix86_emit_probe_stack_range): Pass new arguments to
1234 release_scratch_register_on_entry.
1236 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1238 PR rtl-optimization/84157
1239 * combine.c (change_zero_ext): Use REG_P predicate in
1240 front of HARD_REGISTER_P predicate.
1242 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1244 * config/avr/avr.c (avr_option_override): Move disabling of
1245 -fdelete-null-pointer-checks to...
1246 * common/config/avr/avr-common.c (avr_option_optimization_table):
1247 ...here.
1249 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1251 PR tree-optimization/81635
1252 * tree-data-ref.c (split_constant_offset_1): For types that
1253 wrap on overflow, try to use range info to prove that wrapping
1254 cannot occur.
1256 2018-02-01 Renlin Li <renlin.li@arm.com>
1258 PR target/83370
1259 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1260 TAILCALL_ADDR_REGS.
1261 (aarch64_register_move_cost): Likewise.
1262 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1263 TAILCALL_ADDR_REGS.
1264 (REG_CLASS_NAMES): Likewise.
1265 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1266 TAILCALL_ADDR_REGS. Remove IP registers.
1267 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1269 2018-02-01 Richard Biener <rguenther@suse.de>
1271 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1272 for specifying RPO order and allow NULL for that.
1273 * domwalk.c (dom_walker::dom_walker): Likewise.
1274 (dom_walker::walk): Handle NULL RPO order.
1275 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1276 in RPO order.
1277 (rewrite_update_dom_walker): Likewise.
1278 (mark_def_dom_walker): Likewise.
1280 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1282 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1283 (aarch64_maybe_expand_sve_subreg_move): Declare.
1284 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1285 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1286 predicate.
1287 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1288 that are semantically a reverse operation.
1289 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1290 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1291 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1292 functions.
1293 (aarch64_can_change_mode_class): For big-endian, forbid changes
1294 between two SVE modes if they have different element sizes.
1296 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1298 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1299 the TImode handling for big-endian targets.
1301 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1303 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1304 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1305 not just bytes.
1306 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1307 Remove BSWAP handing for big-endian targets and use the form of
1308 LD1RQ appropariate for the mode.
1310 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1312 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1313 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1314 duplicated element.
1316 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1318 PR tearget/83845
1319 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1320 check for operands that need to go through aarch64_sve_reload_be.
1322 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1324 PR tree-optimization/81661
1325 PR tree-optimization/84117
1326 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1327 * tree-eh.c: Include gimplify.h.
1328 (find_trapping_overflow, replace_trapping_overflow,
1329 rewrite_to_non_trapping_overflow): New functions.
1330 * tree-vect-loop.c: Include tree-eh.h.
1331 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1332 * tree-data-ref.c: Include tree-eh.h.
1333 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1335 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1337 PR rtl-optimization/84123
1338 * combine.c (change_zero_ext): Check if hard register satisfies
1339 can_change_dest_mode before calling gen_lowpart_SUBREG.
1341 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1343 PR target/82444
1344 * ira.c (ira_init_register_move_cost): Remove assert.
1346 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1348 PR rtl-optimization/84071
1349 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1350 * doc/tm.texi: Regenerate.
1352 2018-01-31 Richard Biener <rguenther@suse.de>
1354 PR tree-optimization/84132
1355 * tree-data-ref.c (analyze_miv_subscript): Properly
1356 check whether evolution_function_is_affine_multivariate_p
1357 before calling gcd_of_steps_may_divide_p.
1359 2018-01-31 Julia Koval <julia.koval@intel.com>
1361 PR target/83618
1362 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1363 * config/i386/i386.md (rdpid_rex64) New.
1364 (rdpid): Make 32bit only.
1366 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1368 PR lto/84105
1369 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1370 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1372 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1374 Revert
1375 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1377 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1379 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1381 PR rtl-optimization/84071
1382 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1383 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1385 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1387 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1388 (arc_attribute_table): Add 'aux' attribute.
1389 (arc_in_small_data_p): Consider aux like variables.
1390 (arc_is_aux_reg_p): New function.
1391 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1392 (arc_get_aux_arg): New function.
1393 (prepare_move_operands): Handle aux-register access.
1394 (arc_handle_aux_attribute): New function.
1395 * doc/extend.texi (ARC Variable attributes): Add subsection.
1397 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1399 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1400 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1401 (arc_attribute_table): Add 'uncached' attribute.
1402 (arc_print_operand): Print '.di' flag for uncached memory
1403 accesses.
1404 (arc_in_small_data_p): Do not consider for small data the uncached
1405 types.
1406 (arc_is_uncached_mem_p): New function.
1407 * config/arc/predicates.md (compact_store_memory_operand): Check
1408 for uncached memory accesses.
1409 (nonvol_nonimm_operand): Likewise.
1410 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1412 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1414 PR c/84100
1415 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1416 falign-loops=): Add Optimization flag.
1418 2018-01-30 Jeff Law <law@redhat.com>
1420 PR target/84064
1421 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1422 INT_REGISTERS_SAVED. Check it prior to calling
1423 get_scratch_register_on_entry.
1424 (ix86_adjust_stack_and_probe): Similarly.
1425 (ix86_emit_probe_stack_range): Similarly.
1426 (ix86_expand_prologue): Corresponding changes.
1428 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1430 PR target/40411
1431 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1432 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1434 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1436 PR target/84112
1437 * lra-constraints.c (curr_insn_transform): Process AND in the
1438 address.
1440 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1442 PR rtl-optimization/83986
1443 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1444 dependence against last_pending_memory_flush in addition to
1445 pending_jump_insns.
1447 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1449 PR tree-optimization/81611
1450 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1451 copies.
1453 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1455 PR target/83758
1456 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1457 a reg rtx.
1459 2018-01-30 Richard Biener <rguenther@suse.de>
1460 Jakub Jelinek <jakub@redhat.com>
1462 PR tree-optimization/84111
1463 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1464 inner loops added during recursion, as they don't have up-to-date
1465 SSA form.
1467 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1469 PR ipa/81360
1470 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1471 (can_inline_edge_by_limits_p): ... here.
1472 (can_early_inline_edge_p, check_callers,
1473 update_caller_keys, update_callee_keys, recursive_inlining,
1474 add_new_edges_to_heap, speculation_useful_p,
1475 inline_small_functions,
1476 inline_small_functions, flatten_function,
1477 inline_to_all_callers_1): Update.
1479 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1481 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1482 zeros correctly.
1484 2018-01-30 Richard Biener <rguenther@suse.de>
1486 PR tree-optimization/83008
1487 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1488 invariant and constant vector uses in stmts when they need
1489 more than one stmt.
1491 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1493 PR bootstrap/84017
1494 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1495 * configure: Regenerate.
1497 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1499 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1500 pattern.
1501 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1502 Use gen_rtx_REG rather than gen_lowpart.
1504 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1506 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1507 rather than 0 when creating partial subregs.
1509 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1511 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1512 of usage.
1514 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1516 PR target/81550
1517 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1518 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1519 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1520 flags. This restores the settings used before the 2017-07-24.
1521 Turning off pre increment/decrement/modify allows IVOPTS to
1522 optimize DF/SF loops where the index is an int.
1524 2018-01-29 Richard Biener <rguenther@suse.de>
1525 Kelvin Nilsen <kelvin@gcc.gnu.org>
1527 PR bootstrap/80867
1528 * tree-vect-stmts.c (vectorizable_call): Don't call
1529 targetm.vectorize_builtin_md_vectorized_function if callee is
1530 NULL.
1532 2018-01-22 Carl Love <cel@us.ibm.com>
1534 * doc/extend.tex: Fix typo in second arg in
1535 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1537 2018-01-29 Richard Biener <rguenther@suse.de>
1539 PR tree-optimization/84086
1540 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1541 (flush_ssaname_freelist): When SSA names were released reset
1542 the SCEV hash table.
1544 2018-01-29 Richard Biener <rguenther@suse.de>
1546 PR tree-optimization/84057
1547 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1548 removed paths when removing edges.
1550 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1552 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1553 -mfunction-return=@var{choice}.
1555 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1557 PR diagnostic/84034
1558 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1559 Handle CR like TAB.
1560 (layout::print_source_line): Likewise.
1561 (test_get_line_width_without_trailing_whitespace): Add test cases.
1563 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1565 PR middle-end/84040
1566 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1567 debug insns.
1569 2018-01-26 Jim Wilson <jimw@sifive.com>
1571 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1573 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1574 specified.
1576 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1578 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1579 and CMP + SUB-immediate -> SUBS.
1581 2018-01-26 Martin Sebor <msebor@redhat.com>
1583 PR tree-optimization/83896
1584 * tree-ssa-strlen.c (get_string_len): Rename...
1585 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1586 Avoid assuming length is constant.
1587 (handle_char_store): Use HOST_WIDE_INT for string length.
1589 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1591 PR target/81763
1592 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1593 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1595 2018-01-26 Richard Biener <rguenther@suse.de>
1597 PR rtl-optimization/84003
1598 * dse.c (record_store): Only record redundant stores when
1599 the earlier store aliases at least all accesses the later one does.
1601 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1603 PR rtl-optimization/83985
1604 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1605 REG_CFA_RESTORE insns.
1606 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1607 REG_CFA_RESTORE insns here.
1609 PR c/83989
1610 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1611 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1613 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1615 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1616 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1617 (arc_init): Likewise.
1618 (arc_override_options): Likewise.
1619 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1620 value.
1621 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1622 support.
1623 * config/arc/arc.h (TARGET_DBNZ): Define.
1624 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1625 properly set the tune attribute.
1626 (dbnz): Use TARGET_DBNZ guard.
1627 * config/arc/arc.opt (mtune): Add core3 option.
1629 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1631 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1632 recognize new pic like addresses.
1633 (arc_delegitimize_address): Clean up.
1635 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1637 * config/arc/arc-arches.def: Option mrf16 valid for all
1638 architectures.
1639 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1640 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1641 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1642 * config/arc/arc-tables.opt: Regenerate.
1643 * config/arc/arc.c (arc_conditional_register_usage): Handle
1644 reduced register file case.
1645 (arc_file_start): Set must have build attributes.
1646 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1647 mrf16 option value.
1648 * config/arc/arc.opt (mrf16): Add new option.
1649 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1650 * config/arc/genmultilib.awk: Handle new mrf16 option.
1651 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1652 * config/arc/t-multilib: Regenerate.
1653 * doc/invoke.texi (ARC Options): Document mrf16 option.
1655 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1657 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1658 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1659 (arc_attribute_table): Add 'secure_call' attribute.
1660 (arc_print_operand): Print secure call operand.
1661 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1662 secure.
1663 (arc_is_secure_call_p): New function. * config/arc/arc.md
1664 (call_i): Add support for sjli instruction.
1665 (call_value_i): Likewise.
1666 * config/arc/constraints.md (Csc): New constraint.
1668 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1669 John Eric Martin <John.Martin@emmicro-us.com>
1671 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1672 * config/arc/arc.c (_arc_jli_section): New struct.
1673 (arc_jli_section): New type.
1674 (rc_jli_sections): New static variable.
1675 (arc_handle_jli_attribute): New function.
1676 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1677 (arc_file_end): New function.
1678 (TARGET_ASM_FILE_END): Define.
1679 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1680 (arc_add_jli_section): New function.
1681 (jli_call_scan): Likewise.
1682 (arc_reorg): Call jli_call_scan.
1683 (arc_output_addsi): Remove 'S' from printing asm operand.
1684 (arc_is_jli_call_p): New function.
1685 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1686 operand.
1687 (movhi_insn): Likewise.
1688 (movsi_insn): Likewise.
1689 (movsi_set_cc_insn): Likewise.
1690 (loadqi_update): Likewise.
1691 (load_zeroextendqisi_update): Likewise.
1692 (load_signextendqisi_update): Likewise.
1693 (loadhi_update): Likewise.
1694 (load_zeroextendhisi_update): Likewise.
1695 (load_signextendhisi_update): Likewise.
1696 (loadsi_update): Likewise.
1697 (loadsf_update): Likewise.
1698 (movsicc_insn): Likewise.
1699 (bset_insn): Likewise.
1700 (bxor_insn): Likewise.
1701 (bclr_insn): Likewise.
1702 (bmsk_insn): Likewise.
1703 (bicsi3_insn): Likewise.
1704 (cmpsi_cc_c_insn): Likewise.
1705 (movsi_ne): Likewise.
1706 (movsi_cond_exec): Likewise.
1707 (clrsbsi2): Likewise.
1708 (norm_f): Likewise.
1709 (normw): Likewise.
1710 (swap): Likewise.
1711 (divaw): Likewise.
1712 (flag): Likewise.
1713 (sr): Likewise.
1714 (kflag): Likewise.
1715 (ffs): Likewise.
1716 (ffs_f): Likewise.
1717 (fls): Likewise.
1718 (call_i): Remove 'S' asm letter, add jli instruction.
1719 (call_value_i): Likewise.
1720 * config/arc/arc.op (mjli-always): New option.
1721 * config/arc/constraints.md (Cji): New constraint.
1722 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1723 operand.
1724 (subsf3_fpx): Likewise.
1725 (mulsf3_fpx): Likewise.
1726 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1727 asm operand.
1728 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1729 function attrbutes.
1730 * doc/invoke.texi (ARC): Document mjli-always option.
1732 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1734 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
1735 avoid addition with 0 and use incw and decw where possible.
1737 2018-01-26 Richard Biener <rguenther@suse.de>
1739 PR tree-optimization/81082
1740 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1741 association if it requires casting to unsigned.
1742 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1743 from fold_plusminus_mult_expr to catch important cases late when
1744 range info is available.
1746 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1748 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1749 * configure.ac (hidden_linkonce): New test.
1750 * configure: Regenerate.
1751 * config.in: Regenerate.
1753 2018-01-26 Julia Koval <julia.koval@intel.com>
1755 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1756 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1757 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1758 _mm_mask_bitshuffle_epi64_mask): Fix type.
1759 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1760 USI_FTYPE_V4DI_V4DI_USI): Remove.
1761 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1762 __builtin_ia32_vpshufbitqmb256_mask,
1763 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1764 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1765 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1767 2018-01-26 Alan Modra <amodra@gmail.com>
1769 PR target/84033
1770 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1771 UNSPEC_VBPERMQ. Sort other unspecs.
1773 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1775 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1777 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1779 PR middle-end/83055
1780 * predict.c (drop_profile): Do not push/pop cfun; update also
1781 node->count.
1782 (handle_missing_profiles): Fix logic looking for zero profiles.
1784 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1786 PR middle-end/83977
1787 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1788 on functions with #pragma omp declare simd or functions with simd
1789 attribute.
1790 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1791 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1792 Remove trailing \n from warning_at calls.
1794 2018-01-25 Tom de Vries <tom@codesourcery.com>
1796 PR target/84028
1797 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1798 for neutered workers.
1800 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1802 PR target/68467
1803 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1804 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1806 2018-01-24 Jeff Law <law@redhat.com>
1808 PR target/83994
1809 * i386.c (get_probe_interval): Move to earlier point.
1810 (ix86_compute_frame_layout): If -fstack-clash-protection and
1811 the frame is larger than the probe interval, then use pushes
1812 to save registers rather than reg->mem moves.
1813 (ix86_expand_prologue): Remove conditional for int_registers_saved
1814 assertion.
1816 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1818 PR target/84014
1819 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1820 min/max for never referenced object.
1822 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1824 PR middle-end/83977
1825 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1826 here.
1827 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1828 attributes from DECL_ATTRIBUTES (decl) without affecting
1829 DECL_ATTRIBUTES (current_function_decl).
1830 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1831 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1833 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1835 PR tree-optimization/83979
1836 * fold-const.c (fold_comparison): Use constant_boolean_node
1837 instead of boolean_{true,false}_node.
1839 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1841 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1842 with zero counts.
1844 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1846 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1847 Simplify the clause that sets the length attribute.
1848 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1849 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1850 clause that sets the length attribute.
1851 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1853 2018-01-24 Tom de Vries <tom@codesourcery.com>
1855 PR target/83589
1856 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1857 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1858 Add strict parameter.
1859 (prevent_branch_around_nothing): Insert dummy insn between branch to
1860 label and label with no ptx insn inbetween.
1861 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1863 2018-01-24 Tom de Vries <tom@codesourcery.com>
1865 PR target/81352
1866 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1867 for neutered threads in warp.
1868 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1870 2018-01-24 Richard Biener <rguenther@suse.de>
1872 PR tree-optimization/83176
1873 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1874 operands.
1876 2018-01-24 Richard Biener <rguenther@suse.de>
1878 PR tree-optimization/82819
1879 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1880 code generating pluses that are no-ops in the target precision.
1882 2018-01-24 Richard Biener <rguenther@suse.de>
1884 PR middle-end/84000
1885 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1887 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1889 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1890 to merge probabilities.
1891 * predict.c (probably_never_executed): Also mark as cold functions
1892 with global 0 profile and guessed local profile.
1893 * profile-count.c (profile_probability::combine_with_count): New
1894 member function.
1895 * profile-count.h (profile_probability::operator*,
1896 profile_probability::operator*=, profile_probability::operator/,
1897 profile_probability::operator/=): Reduce precision to adjusted
1898 and set value to guessed on contradictory divisions.
1899 (profile_probability::combine_with_freq): Remove.
1900 (profile_probability::combine_wiht_count): Declare.
1901 (profile_count::force_nonzero):: Set to adjusted.
1902 (profile_count::probability_in):: Set quality to adjusted.
1903 * tree-ssa-tail-merge.c (replace_block_by): Use
1904 combine_with_count.
1906 2018-01-23 Andrew Waterman <andrew@sifive.com>
1907 Jim Wilson <jimw@sifive.com>
1909 * config/riscv/riscv.c (riscv_stack_boundary): New.
1910 (riscv_option_override): Set riscv_stack_boundary. Handle
1911 riscv_preferred_stack_boundary_arg.
1912 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1913 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1914 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1915 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1916 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1917 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1919 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1921 PR target/83905
1922 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1923 of struct ix86_frame.
1924 (ix86_expand_epilogue): Likewise. Add a local variable for
1925 the reg_save_offset field in struct ix86_frame.
1927 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1929 PR tree-optimization/82604
1930 * tree-loop-distribution.c (enum partition_kind): New enum item
1931 PKIND_PARTIAL_MEMSET.
1932 (partition_builtin_p): Support above new enum item.
1933 (generate_code_for_partition): Ditto.
1934 (compute_access_range): Differentiate cases that equality can be
1935 proven at all loops, the innermost loops or no loops.
1936 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1937 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1938 (finalize_partitions, distribute_loop): Don't fuse partition of
1939 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1940 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1941 parloop is enabled.
1943 2018-01-23 Martin Liska <mliska@suse.cz>
1945 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1946 order to ignore the predictor.
1947 (PRED_POLYMORPHIC_CALL): Likewise.
1948 (PRED_RECURSIVE_CALL): Likewise.
1950 2018-01-23 Martin Liska <mliska@suse.cz>
1952 * tree-profile.c (tree_profiling): Print function header to
1953 aware reader which function we are working on.
1954 * value-prof.c (gimple_find_values_to_profile): Do not print
1955 not interesting value histograms.
1957 2018-01-23 Martin Liska <mliska@suse.cz>
1959 * profile-count.h (enum profile_quality): Add
1960 profile_uninitialized as the first value. Do not number values
1961 as they are zero based.
1962 (profile_count::verify): Update sanity check.
1963 (profile_probability::verify): Likewise.
1965 2018-01-23 Nathan Sidwell <nathan@acm.org>
1967 * doc/invoke.texi (ffor-scope): Deprecate.
1969 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1971 PR tree-optimization/83510
1972 * domwalk.c (set_all_edges_as_executable): New function.
1973 (dom_walker::dom_walker): Convert bool param
1974 "skip_unreachable_blocks" to enum reachability. Move setup of
1975 edge flags to set_all_edges_as_executable and only do it when
1976 reachability is REACHABLE_BLOCKS.
1977 * domwalk.h (enum dom_walker::reachability): New enum.
1978 (dom_walker::dom_walker): Convert bool param
1979 "skip_unreachable_blocks" to enum reachability.
1980 (set_all_edges_as_executable): New decl.
1981 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1982 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1983 "reachability".
1984 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1985 but converting true to REACHABLE_BLOCKS.
1986 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1987 * tree-vrp.c
1988 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1989 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1990 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1991 REACHABLE_BLOCKS.
1992 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1993 if check_all_array_refs will be called.
1995 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1997 * tree.c (selftest::test_location_wrappers): Add more test
1998 coverage.
2000 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2002 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2003 (selftest::test_bit_in_range): Likewise.
2005 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2007 PR testsuite/83888
2008 * doc/sourcebuild.texi (vect_float): Say that the selector
2009 only describes the situation when -funsafe-math-optimizations is on.
2010 (vect_float_strict): Document.
2012 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2014 PR tree-optimization/83965
2015 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2016 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2017 instead of checking only for a reduction.
2018 (vect_recog_widen_sum_pattern): Likewise.
2020 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2022 * predict.c (probably_never_executed): Only use precise profile info.
2023 (compute_function_frequency): Skip after inlining hack since we now
2024 have quality checking.
2026 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2028 * profile-count.h (profile_probability::very_unlikely,
2029 profile_probability::unlikely, profile_probability::even): Set
2030 precision to guessed.
2032 2018-01-23 Richard Biener <rguenther@suse.de>
2034 PR tree-optimization/83963
2035 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2036 Properly terminate dominator walk when crossing the exit edge not
2037 when visiting its source block.
2039 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2041 PR c++/83918
2042 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2043 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2045 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2047 PR tree-optimization/83957
2048 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2049 semicolon after for body surrounded by braces.
2051 PR tree-optimization/83081
2052 * profile-count.h (profile_probability::split): New method.
2053 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2054 Use profile_probability::split.
2055 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2056 when splitting a single conditional jump into 2.
2058 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2060 PR tree-optimization/69452
2061 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2062 decl.
2064 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2066 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2067 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2068 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2070 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2072 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2073 * config/rl78/rl78.md (movdi): New define_expand.
2074 * config/rl78/rl78.c (rl78_split_movdi): New function.
2076 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2078 PR target/83862
2079 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2080 no longer used.
2081 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2082 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2083 128-bit to produce an UNSPEC move to get the double word with the
2084 signbit and then a shift directly to do signbit.
2085 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2086 implementation with a new version that just does either a direct
2087 move or a regular move. Move memory interface to separate insns.
2088 Move insns so they are next to the expander.
2089 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2090 with signbit move. Split big and little endian case.
2091 (signbit<mode>2_dm_mem_le): Likewise.
2092 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2093 (signbit<mode>2_dm2): Likewise.
2095 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2097 * config/rl78/rl78.md (anddi3): New define_expand.
2099 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2101 * config/rl78/rl78.md (umindi3): New define_expand.
2103 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2105 * config/rl78/rl78.md (smindi3): New define_expand.
2107 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2109 * config/rl78/rl78.md (smaxdi3): New define_expand.
2111 2018-01-22 Carl Love <cel@us.ibm.com>
2113 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2114 LVX_V1TI): Add macro expansion.
2115 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2116 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2117 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2118 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2119 Change check to determine if the instruction is a byte reversing
2120 entry. Fix typo in comment.
2121 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2122 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2123 Add def_builtin calls for new builtins.
2124 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2125 Add define_insn expansion.
2127 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2129 * config/rl78/rl78.md (umaxdi3): New define_expand.
2131 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2133 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2134 for non-QImode registers.
2136 2018-01-22 Richard Biener <rguenther@suse.de>
2138 PR tree-optimization/83963
2139 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2140 including the loop exit block.
2141 (scop_detection::merge_sese): Likewise.
2142 (scop_detection::add_scop): Do it here instead.
2144 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2146 * doc/sourcebuild.texi (arm_softfloat): Document.
2148 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2150 PR gcc/77734
2151 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2152 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2153 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2155 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2156 David Edelsohn <dje.gcc@gmail.com>
2158 PR target/83946
2159 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2160 Change "crset eq" to "crset 2".
2161 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2162 (*call_indirect_aix<mode>_nospec): Likewise.
2163 (*call_value_indirect_aix<mode>_nospec): Likewise.
2164 (*call_indirect_elfv2<mode>_nospec): Likewise.
2165 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2166 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2167 change assembly output from . to $.
2168 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2169 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2170 (*tablejump<mode>_internal1_nospec): Likewise.
2172 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2174 PR target/80870
2175 * config/sh/sh_optimize_sett_clrt.cc:
2176 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2178 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2180 PR tree-optimization/83940
2181 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2182 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2183 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2184 use it to pass back the definition type.
2185 (vect_check_store_rhs): Likewise rhs_dt_out.
2186 (vect_build_gather_load_calls): Add a mask_dt argument and use
2187 it instead of a call to vect_is_simple_use.
2188 (vectorizable_store): Update calls to vect_check_load_store_mask
2189 and vect_check_store_rhs. Use the dt returned by the latter instead
2190 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2191 instead of calls to vect_is_simple_use. Pass the scalar rather
2192 than the vector operand to vect_is_simple_use when handling
2193 second and subsequent copies of an rhs value.
2194 (vectorizable_load): Update calls to vect_check_load_store_mask
2195 and vect_build_gather_load_calls. Use the cached mask_dt and
2196 gs_info.offset_dt instead of calls to vect_is_simple_use.
2198 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2200 PR middle-end/83945
2201 * tree-emutls.c: Include gimplify.h.
2202 (lower_emutls_2): New function.
2203 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2204 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2205 it before further processing.
2207 PR target/83930
2208 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2209 UINTVAL (trueop1) instead of INTVAL (op1).
2211 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2213 PR debug/81570
2214 PR debug/83728
2215 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2216 INCOMING_FRAME_SP_OFFSET if not defined.
2217 (scan_trace): Add ENTRY argument. If true and
2218 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2219 emit a note to adjust the CFA offset.
2220 (create_cfi_notes): Adjust scan_trace callers.
2221 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2222 INCOMING_FRAME_SP_OFFSET in the CIE.
2223 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2224 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2225 Likewise.
2226 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2227 * doc/tm.texi: Regenerated.
2229 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2231 PR rtl-optimization/83147
2232 * lra-constraints.c (remove_inheritance_pseudos): Use
2233 lra_substitute_pseudo_within_insn.
2235 2018-01-19 Tom de Vries <tom@codesourcery.com>
2236 Cesar Philippidis <cesar@codesourcery.com>
2238 PR target/83920
2239 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2241 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2243 PR target/83790
2244 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2245 spaces for function labels.
2247 2018-01-19 Martin Liska <mliska@suse.cz>
2249 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2250 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2251 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2252 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2253 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2254 (PRED_CONST_RETURN): Change from 69 to 65.
2255 (PRED_NULL_RETURN): Change from 91 to 71.
2256 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2257 (PRED_LOOP_GUARD): Change from 66 to 73.
2259 2018-01-19 Martin Liska <mliska@suse.cz>
2261 * predict.c (predict_insn_def): Add new assert.
2262 (struct branch_predictor): Change type to signed integer.
2263 (test_prediction_value_range): Amend test to cover
2264 PROB_UNINITIALIZED.
2265 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2266 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2267 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2268 (PRED_LOOP_IV_COMPARE): Likewise.
2269 * predict.h (PROB_UNINITIALIZED): Define new constant.
2271 2018-01-19 Martin Liska <mliska@suse.cz>
2273 * predict.c (dump_prediction): Add new format for
2274 analyze_brprob.py script which is enabled with -details
2275 suboption.
2276 * profile-count.h (precise_p): New function.
2278 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2280 PR tree-optimization/83922
2281 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2282 there are no statements that need masking.
2283 (vect_active_double_reduction_p): New function.
2284 (vect_analyze_loop_operations): Use it when handling phis that
2285 are not in the loop header.
2287 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2289 PR tree-optimization/83914
2290 * tree-vect-loop.c (vectorizable_induction): Don't convert
2291 init_expr or apply the peeling adjustment for inductions
2292 that are nested within the vectorized loop.
2294 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2296 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2297 instead of NEG.
2299 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2301 PR sanitizer/81715
2302 PR testsuite/83882
2303 * function.h (gimplify_parameters): Add gimple_seq * argument.
2304 * function.c: Include gimple.h and options.h.
2305 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2306 for the added local temporaries if needed.
2307 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2308 if there are any parameter cleanups, wrap whole body into a
2309 try/finally with the cleanups.
2311 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2313 PR target/82964
2314 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2315 Use GET_MODE_CLASS for scalar floating point.
2317 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2319 PR ipa/82256
2320 patch by PaX Team
2321 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2322 Fix call of call_cgraph_insertion_hooks.
2324 2018-01-18 Martin Sebor <msebor@redhat.com>
2326 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2328 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2330 PR ipa/83619
2331 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2332 frequencies.
2334 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2336 PR other/70268
2337 * common.opt: (-ffile-prefix-map): New option.
2338 * opts.c (common_handle_option): Defer it.
2339 * opts-global.c (handle_common_deferred_options): Handle it.
2340 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2341 * file-prefix-map.h: New file.
2342 (remap_debug_filename, add_debug_prefix_map): ...here.
2343 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2344 * final.c (debug_prefix_map, add_debug_prefix_map
2345 remap_debug_filename): Move to...
2346 * file-prefix-map.c: New file.
2347 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2348 generalize, get rid of alloca(), use strrchr() instead of strchr().
2349 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2350 Implement in terms of add_prefix_map().
2351 (remap_macro_filename, remap_debug_filename): Implement in term of
2352 remap_filename().
2353 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2354 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2355 * dbxout.c: Include file-prefix-map.h.
2356 * varasm.c: Likewise.
2357 * vmsdbgout.c: Likewise.
2358 * xcoffout.c: Likewise.
2359 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2360 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2361 * doc/invoke.texi (-ffile-prefix-map): Document.
2362 (-fdebug-prefix-map): Update description.
2364 2018-01-18 Martin Liska <mliska@suse.cz>
2366 * config/i386/i386.c (indirect_thunk_name): Document that also
2367 lfence is emitted.
2368 (output_indirect_thunk): Document why both instructions
2369 (pause and lfence) are generated.
2371 2018-01-18 Richard Biener <rguenther@suse.de>
2373 PR tree-optimization/83887
2374 * graphite-scop-detection.c
2375 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2376 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2377 (scop_detection::merge_sese): Re-implement with a flood-fill
2378 algorithm that properly finds a SESE region if it exists.
2380 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2382 PR c/61240
2383 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2384 pointer_diff optimizations use view_convert instead of convert.
2386 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2388 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2389 Generate different code for -mno-speculate-indirect-jumps.
2390 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2391 (*call_indirect_aix<mode>): Disable for
2392 -mno-speculate-indirect-jumps.
2393 (*call_indirect_aix<mode>_nospec): New define_insn.
2394 (*call_value_indirect_aix<mode>): Disable for
2395 -mno-speculate-indirect-jumps.
2396 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2397 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2398 -mno-speculate-indirect-jumps.
2399 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2401 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2403 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2404 long double type, set the flags for noting the default long double
2405 type, even if we don't pass or return a long double type.
2407 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2409 PR ipa/83051
2410 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2411 failure.
2413 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2415 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2416 support for merge[hl].
2417 (fold_mergehl_helper): New helper function.
2418 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2419 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2420 (altivec_vmrglw_direct): Add xxmrglw insn.
2422 2018-01-17 Andrew Waterman <andrew@sifive.com>
2424 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2425 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2427 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2429 PR lto/83121
2430 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2431 call the lto_location_cache before reading the
2432 DECL_SOURCE_LOCATION of the types.
2434 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2435 Richard Sandiford <richard.sandiford@linaro.org>
2437 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2438 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2439 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2440 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2441 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2442 Add declaration.
2443 * config/aarch64/constraints.md (aarch64_movti_operand):
2444 Limit immediates.
2445 * config/aarch64/predicates.md (Uti): Add new constraint.
2447 2018-01-17 Carl Love <cel@us.ibm.com>
2449 * config/rs6000/vsx.md (define_expand xl_len_r,
2450 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2451 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2452 lxvll.
2453 (define_expand, define_insn): Move the shift left from the
2454 define_insn to the define_expand for lxvl and stxvl instructions.
2455 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2456 and XL_LEN_R definitions to PURE.
2458 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2460 * config/i386/i386.c (indirect_thunk_name): Declare regno
2461 as unsigned int. Compare regno with INVALID_REGNUM.
2462 (output_indirect_thunk): Ditto.
2463 (output_indirect_thunk_function): Ditto.
2464 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2465 in the call to output_indirect_thunk_function.
2467 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2469 PR middle-end/83884
2470 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2471 rather than the size of inner_type to determine the stack slot size
2472 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2474 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2476 PR target/83546
2477 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2478 to PTA_SILVERMONT.
2480 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2482 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2483 endian Linux systems to optionally enable multilibs for selecting
2484 the long double type if the user configured an explicit type.
2485 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2486 have no long double multilibs if not defined.
2487 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2488 warn if the user used -mabi={ieee,ibm}longdouble and we built
2489 multilibs for long double.
2490 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2491 appropriate multilib option.
2492 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2493 multilib options.
2494 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2495 for building long double multilibs.
2496 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2498 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2500 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2501 copies.
2503 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2504 64 bits.
2505 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2506 128 bits.
2508 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2509 variables.
2511 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2512 return value.
2514 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2516 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2517 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2519 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2521 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2522 different rtl trees depending on TARGET_64BIT.
2523 (rs6000_gen_lvx): Likewise.
2525 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2527 * config/visium/visium.md (nop): Tweak comment.
2528 (hazard_nop): Likewise.
2530 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2532 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2533 -mspeculate-indirect-jumps.
2534 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2535 for -mno-speculate-indirect-jumps.
2536 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2537 (*call_value_indirect_elfv2<mode>): Disable for
2538 -mno-speculate-indirect-jumps.
2539 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2540 (indirect_jump): Emit different RTL for
2541 -mno-speculate-indirect-jumps.
2542 (*indirect_jump<mode>): Disable for
2543 -mno-speculate-indirect-jumps.
2544 (*indirect_jump<mode>_nospec): New define_insn.
2545 (tablejump): Emit different RTL for
2546 -mno-speculate-indirect-jumps.
2547 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2548 (tablejumpsi_nospec): New define_expand.
2549 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2550 (tablejumpdi_nospec): New define_expand.
2551 (*tablejump<mode>_internal1): Disable for
2552 -mno-speculate-indirect-jumps.
2553 (*tablejump<mode>_internal1_nospec): New define_insn.
2554 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2555 option.
2557 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2559 * caller-save.c (insert_save): Drop unnecessary parameter. All
2560 callers updated.
2562 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2563 Richard Biener <rguenth@suse.de>
2565 PR libgomp/83590
2566 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2567 return early, inline manually is_gimple_sizepos. Make sure if we
2568 call gimplify_expr we don't end up with a gimple constant.
2569 * tree.c (variably_modified_type_p): Don't return true for
2570 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2571 * gimplify.h (is_gimple_sizepos): Remove.
2573 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2575 PR tree-optimization/83857
2576 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2577 vectorizable_live_operation for pure SLP statements.
2578 (vectorizable_live_operation): Handle PHIs.
2580 2018-01-16 Richard Biener <rguenther@suse.de>
2582 PR tree-optimization/83867
2583 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2584 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2586 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2588 PR c/83844
2589 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2590 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2591 If off is not INTEGER_CST, issue a may not be aligned warning
2592 rather than isn't aligned. Use isn%'t rather than isn't.
2593 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2594 into MULT_EXPR.
2595 <case MULT_EXPR>: Improve the case when bottom and one of the
2596 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2597 operand, in that case check if the other operand is multiple of
2598 bottom divided by the INTEGER_CST operand.
2600 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2602 PR target/83858
2603 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2604 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2605 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2606 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2607 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2608 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2609 (pa_function_arg_size): New function.
2611 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2613 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2614 in a separate statement.
2616 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2618 PR tree-optimization/83847
2619 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2620 group gathers and scatters.
2622 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2624 PR rtl-optimization/86620
2625 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2627 PR rtl-optimization/83213
2628 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2629 to last if both are JUMP_INSNs.
2631 PR tree-optimization/83843
2632 * gimple-ssa-store-merging.c
2633 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2634 store_immediate_info for bswap/nop orig_stores.
2636 2018-01-15 Andrew Waterman <andrew@sifive.com>
2638 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2639 !TARGET_MUL.
2640 <UDIV>: Increase cost if !TARGET_DIV.
2642 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2644 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2645 (define_attr "cr_logical_3op"): New.
2646 (cceq_ior_compare): Adjust.
2647 (cceq_ior_compare_complement): Adjust.
2648 (*cceq_rev_compare): Adjust.
2649 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2650 (is_cracked_insn): Adjust.
2651 (insn_must_be_first_in_group): Adjust.
2652 * config/rs6000/40x.md: Adjust.
2653 * config/rs6000/440.md: Adjust.
2654 * config/rs6000/476.md: Adjust.
2655 * config/rs6000/601.md: Adjust.
2656 * config/rs6000/603.md: Adjust.
2657 * config/rs6000/6xx.md: Adjust.
2658 * config/rs6000/7450.md: Adjust.
2659 * config/rs6000/7xx.md: Adjust.
2660 * config/rs6000/8540.md: Adjust.
2661 * config/rs6000/cell.md: Adjust.
2662 * config/rs6000/e300c2c3.md: Adjust.
2663 * config/rs6000/e500mc.md: Adjust.
2664 * config/rs6000/e500mc64.md: Adjust.
2665 * config/rs6000/e5500.md: Adjust.
2666 * config/rs6000/e6500.md: Adjust.
2667 * config/rs6000/mpc.md: Adjust.
2668 * config/rs6000/power4.md: Adjust.
2669 * config/rs6000/power5.md: Adjust.
2670 * config/rs6000/power6.md: Adjust.
2671 * config/rs6000/power7.md: Adjust.
2672 * config/rs6000/power8.md: Adjust.
2673 * config/rs6000/power9.md: Adjust.
2674 * config/rs6000/rs64.md: Adjust.
2675 * config/rs6000/titan.md: Adjust.
2677 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2679 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2680 ix86_indirect_branch_register logic.
2682 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2684 * config/i386/constraints.md (Bs): Update
2685 ix86_indirect_branch_register check. Don't check
2686 ix86_indirect_branch_register with GOT_memory_operand.
2687 (Bw): Likewise.
2688 * config/i386/predicates.md (GOT_memory_operand): Don't check
2689 ix86_indirect_branch_register here.
2690 (GOT32_symbol_operand): Likewise.
2692 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2694 * config/i386/predicates.md (constant_call_address_operand):
2695 Rewrite ix86_indirect_branch_register logic.
2696 (sibcall_insn_operand): Likewise.
2698 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2700 * config/i386/constraints.md (Bs): Replace
2701 ix86_indirect_branch_thunk_register with
2702 ix86_indirect_branch_register.
2703 (Bw): Likewise.
2704 * config/i386/i386.md (indirect_jump): Likewise.
2705 (tablejump): Likewise.
2706 (*sibcall_memory): Likewise.
2707 (*sibcall_value_memory): Likewise.
2708 Peepholes of indirect call and jump via memory: Likewise.
2709 * config/i386/i386.opt: Likewise.
2710 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2711 (GOT_memory_operand): Likewise.
2712 (call_insn_operand): Likewise.
2713 (sibcall_insn_operand): Likewise.
2714 (GOT32_symbol_operand): Likewise.
2716 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2718 PR middle-end/83837
2719 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2720 type rather than type addr's type points to.
2721 (expand_omp_atomic_mutex): Likewise.
2722 (expand_omp_atomic): Likewise.
2724 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2726 PR target/83839
2727 * config/i386/i386.c (output_indirect_thunk_function): Use
2728 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2729 for __x86_return_thunk.
2731 2018-01-15 Richard Biener <rguenther@suse.de>
2733 PR middle-end/83850
2734 * expmed.c (extract_bit_field_1): Fix typo.
2736 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2738 PR target/83687
2739 * config/arm/iterators.md (VF): New mode iterator.
2740 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2741 Remove integer-related logic from pattern.
2742 (neon_vabd<mode>_3): Likewise.
2744 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2746 PR middle-end/82694
2747 * common.opt (fstrict-overflow): No longer an alias.
2748 (fwrapv-pointer): New option.
2749 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2750 also for pointer types based on flag_wrapv_pointer.
2751 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2752 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2753 opts->x_flag_wrapv got set.
2754 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2755 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2756 POINTER_TYPE_OVERFLOW_UNDEFINED.
2757 * match.pd: Likewise in address comparison pattern.
2758 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2760 2018-01-15 Richard Biener <rguenther@suse.de>
2762 PR lto/83804
2763 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2764 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2765 Reset type names to their identifier if their TYPE_DECL doesn't
2766 have linkage (and thus is used for ODR and devirt).
2767 (save_debug_info_for_decl): Remove.
2768 (save_debug_info_for_type): Likewise.
2769 (add_tree_to_fld_list): Adjust.
2770 * tree-pretty-print.c (dump_generic_node): Make dumping of
2771 type names more robust.
2773 2018-01-15 Richard Biener <rguenther@suse.de>
2775 * BASE-VER: Bump to 8.0.1.
2777 2018-01-14 Martin Sebor <msebor@redhat.com>
2779 PR other/83508
2780 * builtins.c (check_access): Avoid warning when the no-warning bit
2781 is set.
2783 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2785 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2786 * ira-color (allocno_hard_regs_compare): Likewise.
2788 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2790 PR target/83013
2791 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2792 Use .pushsection/.popsection.
2794 2018-01-14 Martin Sebor <msebor@redhat.com>
2796 PR c++/81327
2797 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2799 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2801 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2802 entry from extra_headers.
2803 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2804 extra_headers, make the list bitwise identical to the i?86-*-* one.
2806 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2808 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2809 -mcmodel=large with -mindirect-branch=thunk,
2810 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2811 -mfunction-return=thunk-extern.
2812 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2813 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2814 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2816 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2818 * config/i386/i386.c (print_reg): Print the name of the full
2819 integer register without '%'.
2820 (ix86_print_operand): Handle 'V'.
2821 * doc/extend.texi: Document 'V' modifier.
2823 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2825 * config/i386/constraints.md (Bs): Disallow memory operand for
2826 -mindirect-branch-register.
2827 (Bw): Likewise.
2828 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2829 (GOT_memory_operand): Likewise.
2830 (call_insn_operand): Likewise.
2831 (sibcall_insn_operand): Likewise.
2832 (GOT32_symbol_operand): Likewise.
2833 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2834 for -mindirect-branch-register.
2835 (tablejump): Likewise.
2836 (*sibcall_memory): Likewise.
2837 (*sibcall_value_memory): Likewise.
2838 Disallow peepholes of indirect call and jump via memory for
2839 -mindirect-branch-register.
2840 (*call_pop): Replace m with Bw.
2841 (*call_value_pop): Likewise.
2842 (*sibcall_pop_memory): Replace m with Bs.
2843 * config/i386/i386.opt (mindirect-branch-register): New option.
2844 * doc/invoke.texi: Document -mindirect-branch-register option.
2846 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2848 * config/i386/i386-protos.h (ix86_output_function_return): New.
2849 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2850 set function_return_type.
2851 (indirect_thunk_name): Add ret_p to indicate thunk for function
2852 return.
2853 (output_indirect_thunk_function): Pass false to
2854 indirect_thunk_name.
2855 (ix86_output_indirect_branch_via_reg): Likewise.
2856 (ix86_output_indirect_branch_via_push): Likewise.
2857 (output_indirect_thunk_function): Create alias for function
2858 return thunk if regno < 0.
2859 (ix86_output_function_return): New function.
2860 (ix86_handle_fndecl_attribute): Handle function_return.
2861 (ix86_attribute_table): Add function_return.
2862 * config/i386/i386.h (machine_function): Add
2863 function_return_type.
2864 * config/i386/i386.md (simple_return_internal): Use
2865 ix86_output_function_return.
2866 (simple_return_internal_long): Likewise.
2867 * config/i386/i386.opt (mfunction-return=): New option.
2868 (indirect_branch): Mention -mfunction-return=.
2869 * doc/extend.texi: Document function_return function attribute.
2870 * doc/invoke.texi: Document -mfunction-return= option.
2872 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2874 * config/i386/i386-opts.h (indirect_branch): New.
2875 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2876 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2877 with local indirect jump when converting indirect call and jump.
2878 (ix86_set_indirect_branch_type): New.
2879 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2880 (indirectlabelno): New.
2881 (indirect_thunk_needed): Likewise.
2882 (indirect_thunk_bnd_needed): Likewise.
2883 (indirect_thunks_used): Likewise.
2884 (indirect_thunks_bnd_used): Likewise.
2885 (INDIRECT_LABEL): Likewise.
2886 (indirect_thunk_name): Likewise.
2887 (output_indirect_thunk): Likewise.
2888 (output_indirect_thunk_function): Likewise.
2889 (ix86_output_indirect_branch_via_reg): Likewise.
2890 (ix86_output_indirect_branch_via_push): Likewise.
2891 (ix86_output_indirect_branch): Likewise.
2892 (ix86_output_indirect_jmp): Likewise.
2893 (ix86_code_end): Call output_indirect_thunk_function if needed.
2894 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2895 needed.
2896 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2897 (ix86_attribute_table): Add indirect_branch.
2898 * config/i386/i386.h (machine_function): Add indirect_branch_type
2899 and has_local_indirect_jump.
2900 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2901 to true.
2902 (tablejump): Likewise.
2903 (*indirect_jump): Use ix86_output_indirect_jmp.
2904 (*tablejump_1): Likewise.
2905 (simple_return_indirect_internal): Likewise.
2906 * config/i386/i386.opt (mindirect-branch=): New option.
2907 (indirect_branch): New.
2908 (keep): Likewise.
2909 (thunk): Likewise.
2910 (thunk-inline): Likewise.
2911 (thunk-extern): Likewise.
2912 * doc/extend.texi: Document indirect_branch function attribute.
2913 * doc/invoke.texi: Document -mindirect-branch= option.
2915 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2917 PR ipa/83051
2918 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2920 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2922 * ipa-inline.c (want_inline_small_function_p): Return false if
2923 inlining has already failed with CIF_FINAL_ERROR.
2924 (update_caller_keys): Call want_inline_small_function_p before
2925 can_inline_edge_p.
2926 (update_callee_keys): Likewise.
2928 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2930 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2931 New function.
2932 (rs6000_quadword_masked_address_p): Likewise.
2933 (quad_aligned_load_p): Likewise.
2934 (quad_aligned_store_p): Likewise.
2935 (const_load_sequence_p): Add comment to describe the outer-most loop.
2936 (mimic_memory_attributes_and_flags): New function.
2937 (rs6000_gen_stvx): Likewise.
2938 (replace_swapped_aligned_store): Likewise.
2939 (rs6000_gen_lvx): Likewise.
2940 (replace_swapped_aligned_load): Likewise.
2941 (replace_swapped_load_constant): Capitalize argument name in
2942 comment describing this function.
2943 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2944 and stores that access quad-word aligned addresses and replace
2945 with stvx or lvx instructions when appropriate.
2946 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2947 New function prototype.
2948 (rs6000_quadword_masked_address_p): Likewise.
2949 (rs6000_gen_lvx): Likewise.
2950 (rs6000_gen_stvx): Likewise.
2951 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2952 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2953 when memory address is aligned.
2954 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2955 this split to select lvx instruction when memory address is aligned.
2956 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2957 instruction when memory address is aligned.
2958 (*vsx_le_perm_load_v16qi): Likewise.
2959 (four unnamed splitters): Modify to select the stvx instruction
2960 when memory is aligned.
2962 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2964 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2965 which appears in the queue multiple times.
2967 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2968 Alan Hayward <alan.hayward@arm.com>
2969 David Sherwood <david.sherwood@arm.com>
2971 * tree-vectorizer.h (vec_lower_bound): New structure.
2972 (_loop_vec_info): Add check_nonzero and lower_bounds.
2973 (LOOP_VINFO_CHECK_NONZERO): New macro.
2974 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2975 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2976 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2977 fields. Make seg_len the distance travelled, not including the
2978 access size.
2979 (dr_direction_indicator): Declare.
2980 (dr_zero_step_indicator): Likewise.
2981 (dr_known_forward_stride_p): Likewise.
2982 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2983 tree-ssanames.h.
2984 (runtime_alias_check_p): Allow runtime alias checks with
2985 variable strides.
2986 (operator ==): Compare access_size and align.
2987 (prune_runtime_alias_test_list): Rework for new distinction between
2988 the access_size and seg_len.
2989 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2990 segment lengths.
2991 (get_segment_min_max): New function.
2992 (create_intersect_range_checks): Use it.
2993 (dr_step_indicator): New function.
2994 (dr_direction_indicator): Likewise.
2995 (dr_zero_step_indicator): Likewise.
2996 (dr_known_forward_stride_p): Likewise.
2997 * tree-loop-distribution.c (data_ref_segment_size): Return
2998 DR_STEP * (niters - 1).
2999 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3000 constructor.
3001 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3002 (vect_preserves_scalar_order_p): New function, split out from...
3003 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3004 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3005 (vect_vfa_access_size): New function.
3006 (vect_vfa_align): Likewise.
3007 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3008 (dump_lower_bound): New function.
3009 (vect_check_lower_bound): Likewise.
3010 (vect_small_gap_p): Likewise.
3011 (vectorizable_with_step_bound_p): Likewise.
3012 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3013 depencies if the vectorization factor is 1. Convert the checks
3014 for nonzero steps into checks on the bounds of DR_STEP. Try using
3015 a bunds check for variable steps if the minimum required step is
3016 relatively small. Update calls to the dr_with_seg_len
3017 constructor and to vect_compile_time_alias.
3018 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3019 function.
3020 (vect_loop_versioning): Call it.
3021 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3022 when retrying.
3023 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3025 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3026 Alan Hayward <alan.hayward@arm.com>
3027 David Sherwood <david.sherwood@arm.com>
3029 * doc/sourcebuild.texi (vect_scatter_store): Document.
3030 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3031 optabs.
3032 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3033 Document.
3034 * genopinit.c (main): Add supports_vec_scatter_store and
3035 supports_vec_scatter_store_cached to target_optabs.
3036 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3037 IFN_MASK_SCATTER_STORE.
3038 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3039 functions.
3040 * internal-fn.h (internal_store_fn_p): Declare.
3041 (internal_fn_stored_value_index): Likewise.
3042 * internal-fn.c (scatter_store_direct): New macro.
3043 (expand_scatter_store_optab_fn): New function.
3044 (direct_scatter_store_optab_supported_p): New macro.
3045 (internal_store_fn_p): New function.
3046 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3047 IFN_MASK_SCATTER_STORE.
3048 (internal_fn_mask_index): Likewise.
3049 (internal_fn_stored_value_index): New function.
3050 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3051 for scatter stores.
3052 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3053 * optabs-query.c (supports_vec_scatter_store_p): New function.
3054 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3055 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3056 true for scatter stores.
3057 (vect_gather_scatter_fn_p): Handle scatter stores too.
3058 (vect_check_gather_scatter): Consider using scatter stores if
3059 supports_vec_scatter_store_p.
3060 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3061 scatter stores too.
3062 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3063 internal_fn_stored_value_index.
3064 (check_load_store_masking): Handle scatter stores too.
3065 (vect_get_store_rhs): Make public.
3066 (vectorizable_call): Use internal_store_fn_p.
3067 (vectorizable_store): Handle scatter store internal functions.
3068 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3069 when deciding whether the end of the group has been reached.
3070 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3071 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3072 (mask_scatter_store<mode>): New insns.
3074 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3075 Alan Hayward <alan.hayward@arm.com>
3076 David Sherwood <david.sherwood@arm.com>
3078 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3079 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3080 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3081 function.
3082 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3083 Use vect_truncate_gather_scatter_offset if we can't treat the
3084 operation as a normal gather load or scatter store.
3085 (get_group_load_store_type): Take the gather_scatter_info
3086 as argument. Try using a gather load or scatter store for
3087 single-element groups.
3088 (get_load_store_type): Update calls to get_group_load_store_type
3089 and vect_use_strided_gather_scatters_p.
3091 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3092 Alan Hayward <alan.hayward@arm.com>
3093 David Sherwood <david.sherwood@arm.com>
3095 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3096 optional tree argument.
3097 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3098 null target hooks.
3099 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3100 but continue to use the current value as a fallback.
3101 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3102 to compare the updates.
3103 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3104 (get_load_store_type): Use it when handling a strided access.
3105 (vect_get_strided_load_store_ops): New function.
3106 (vect_get_data_ptr_increment): Likewise.
3107 (vectorizable_load): Handle strided gather loads. Always pass
3108 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3110 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3111 Alan Hayward <alan.hayward@arm.com>
3112 David Sherwood <david.sherwood@arm.com>
3114 * doc/md.texi (gather_load@var{m}): Document.
3115 (mask_gather_load@var{m}): Likewise.
3116 * genopinit.c (main): Add supports_vec_gather_load and
3117 supports_vec_gather_load_cached to target_optabs.
3118 * optabs-tree.c (init_tree_optimization_optabs): Use
3119 ggc_cleared_alloc to allocate target_optabs.
3120 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3121 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3122 functions.
3123 * internal-fn.h (internal_load_fn_p): Declare.
3124 (internal_gather_scatter_fn_p): Likewise.
3125 (internal_fn_mask_index): Likewise.
3126 (internal_gather_scatter_fn_supported_p): Likewise.
3127 * internal-fn.c (gather_load_direct): New macro.
3128 (expand_gather_load_optab_fn): New function.
3129 (direct_gather_load_optab_supported_p): New macro.
3130 (direct_internal_fn_optab): New function.
3131 (internal_load_fn_p): Likewise.
3132 (internal_gather_scatter_fn_p): Likewise.
3133 (internal_fn_mask_index): Likewise.
3134 (internal_gather_scatter_fn_supported_p): Likewise.
3135 * optabs-query.c (supports_at_least_one_mode_p): New function.
3136 (supports_vec_gather_load_p): Likewise.
3137 * optabs-query.h (supports_vec_gather_load_p): Declare.
3138 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3139 and memory_type field.
3140 (NUM_PATTERNS): Bump to 15.
3141 * tree-vect-data-refs.c: Include internal-fn.h.
3142 (vect_gather_scatter_fn_p): New function.
3143 (vect_describe_gather_scatter_call): Likewise.
3144 (vect_check_gather_scatter): Try using internal functions for
3145 gather loads. Recognize existing calls to a gather load function.
3146 (vect_analyze_data_refs): Consider using gather loads if
3147 supports_vec_gather_load_p.
3148 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3149 (vect_get_gather_scatter_offset_type): Likewise.
3150 (vect_convert_mask_for_vectype): Likewise.
3151 (vect_add_conversion_to_patterm): Likewise.
3152 (vect_try_gather_scatter_pattern): Likewise.
3153 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3154 (vect_vect_recog_func_ptrs): Add it.
3155 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3156 internal_fn_mask_index and internal_gather_scatter_fn_p.
3157 (check_load_store_masking): Take the gather_scatter_info as an
3158 argument and handle gather loads.
3159 (vect_get_gather_scatter_ops): New function.
3160 (vectorizable_call): Check internal_load_fn_p.
3161 (vectorizable_load): Likewise. Handle gather load internal
3162 functions.
3163 (vectorizable_store): Update call to check_load_store_masking.
3164 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3165 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3166 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3167 (aarch64_gather_scale_operand_d): New predicates.
3168 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3169 (mask_gather_load<mode>): New insns.
3171 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3172 Alan Hayward <alan.hayward@arm.com>
3173 David Sherwood <david.sherwood@arm.com>
3175 * optabs.def (fold_left_plus_optab): New optab.
3176 * doc/md.texi (fold_left_plus_@var{m}): Document.
3177 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3178 * internal-fn.c (fold_left_direct): Define.
3179 (expand_fold_left_optab_fn): Likewise.
3180 (direct_fold_left_optab_supported_p): Likewise.
3181 * fold-const-call.c (fold_const_fold_left): New function.
3182 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3183 * tree-parloops.c (valid_reduction_p): New function.
3184 (gather_scalar_reductions): Use it.
3185 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3186 (vect_finish_replace_stmt): Declare.
3187 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3188 (needs_fold_left_reduction_p): New function, split out from...
3189 (vect_is_simple_reduction): ...here. Accept reductions that
3190 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3191 (vect_force_simple_reduction): Also store the reduction type in
3192 the assignment's STMT_VINFO_REDUC_TYPE.
3193 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3194 (merge_with_identity): New function.
3195 (vect_expand_fold_left): Likewise.
3196 (vectorize_fold_left_reduction): Likewise.
3197 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3198 scalar phi in place for it. Check for target support and reject
3199 cases that would reassociate the operation. Defer the transform
3200 phase to vectorize_fold_left_reduction.
3201 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3202 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3203 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3205 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3207 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3208 call to ifc_temp_var.
3210 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3211 Alan Hayward <alan.hayward@arm.com>
3212 David Sherwood <david.sherwood@arm.com>
3214 * target.def (legitimize_address_displacement): Take the original
3215 offset as a poly_int.
3216 * targhooks.h (default_legitimize_address_displacement): Update
3217 accordingly.
3218 * targhooks.c (default_legitimize_address_displacement): Likewise.
3219 * doc/tm.texi: Regenerate.
3220 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3221 as an argument, moving assert of ad->disp == ad->disp_term to...
3222 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3223 Try calling targetm.legitimize_address_displacement before expanding
3224 the address rather than afterwards, and adjust for the new interface.
3225 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3226 Match the new hook interface. Handle SVE addresses.
3227 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3228 new hook interface.
3230 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3232 * Makefile.in (OBJS): Add early-remat.o.
3233 * target.def (select_early_remat_modes): New hook.
3234 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3235 * doc/tm.texi: Regenerate.
3236 * targhooks.h (default_select_early_remat_modes): Declare.
3237 * targhooks.c (default_select_early_remat_modes): New function.
3238 * timevar.def (TV_EARLY_REMAT): New timevar.
3239 * passes.def (pass_early_remat): New pass.
3240 * tree-pass.h (make_pass_early_remat): Declare.
3241 * early-remat.c: New file.
3242 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3243 function.
3244 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3246 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3247 Alan Hayward <alan.hayward@arm.com>
3248 David Sherwood <david.sherwood@arm.com>
3250 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3251 vfm1 with a bound_epilog parameter.
3252 (vect_do_peeling): Update calls accordingly, and move the prologue
3253 call earlier in the function. Treat the base bound_epilog as 0 for
3254 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3255 this base when peeling for gaps.
3256 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3257 with fully-masked loops.
3258 (vect_estimate_min_profitable_iters): Handle the single peeled
3259 iteration in that case.
3261 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3262 Alan Hayward <alan.hayward@arm.com>
3263 David Sherwood <david.sherwood@arm.com>
3265 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3266 single-element interleaving even if the size is not a power of 2.
3267 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3268 accesses for single-element interleaving if the group size is
3269 not a power of 2.
3271 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3272 Alan Hayward <alan.hayward@arm.com>
3273 David Sherwood <david.sherwood@arm.com>
3275 * doc/md.texi (fold_extract_last_@var{m}): Document.
3276 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3277 * optabs.def (fold_extract_last_optab): New optab.
3278 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3279 * internal-fn.c (fold_extract_direct): New macro.
3280 (expand_fold_extract_optab_fn): Likewise.
3281 (direct_fold_extract_optab_supported_p): Likewise.
3282 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3283 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3284 EXTRACT_LAST_REDUCTION.
3285 (get_initial_def_for_reduction): Do not create an initial vector
3286 for EXTRACT_LAST_REDUCTION reductions.
3287 (vectorizable_reduction): Leave the scalar phi in place for
3288 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3289 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3290 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3291 transform phase to vectorizable_condition.
3292 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3293 split out from...
3294 (vect_finish_stmt_generation): ...here.
3295 (vect_finish_replace_stmt): New function.
3296 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3297 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3298 pattern.
3299 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3301 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3302 Alan Hayward <alan.hayward@arm.com>
3303 David Sherwood <david.sherwood@arm.com>
3305 * doc/md.texi (extract_last_@var{m}): Document.
3306 * optabs.def (extract_last_optab): New optab.
3307 * internal-fn.def (EXTRACT_LAST): New internal function.
3308 * internal-fn.c (cond_unary_direct): New macro.
3309 (expand_cond_unary_optab_fn): Likewise.
3310 (direct_cond_unary_optab_supported_p): Likewise.
3311 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3312 loops using EXTRACT_LAST.
3313 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3314 (extract_last_<mode>): ...this optab.
3315 (vec_extract<mode><Vel>): Update accordingly.
3317 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3318 Alan Hayward <alan.hayward@arm.com>
3319 David Sherwood <david.sherwood@arm.com>
3321 * target.def (empty_mask_is_expensive): New hook.
3322 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3323 * doc/tm.texi: Regenerate.
3324 * targhooks.h (default_empty_mask_is_expensive): Declare.
3325 * targhooks.c (default_empty_mask_is_expensive): New function.
3326 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3327 if the target says that empty masks are expensive.
3328 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3329 New function.
3330 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3332 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3333 Alan Hayward <alan.hayward@arm.com>
3334 David Sherwood <david.sherwood@arm.com>
3336 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3337 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3338 (vect_use_loop_mask_for_alignment_p): New function.
3339 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3340 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3341 niters_skip argument. Make sure that the first niters_skip elements
3342 of the first iteration are inactive.
3343 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3344 Update call to vect_set_loop_masks_directly.
3345 (get_misalign_in_elems): New function, split out from...
3346 (vect_gen_prolog_loop_niters): ...here.
3347 (vect_update_init_of_dr): Take a code argument that specifies whether
3348 the adjustment should be added or subtracted.
3349 (vect_update_init_of_drs): Likewise.
3350 (vect_prepare_for_masked_peels): New function.
3351 (vect_do_peeling): Skip prologue peeling if we're using a mask
3352 instead. Update call to vect_update_inits_of_drs.
3353 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3354 mask_skip_niters.
3355 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3356 alignment. Do not include the number of peeled iterations in
3357 the minimum threshold in that case.
3358 (vectorizable_induction): Adjust the start value down by
3359 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3360 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3361 Take the number of skipped iterations into account when calculating
3362 the loop bounds.
3363 * tree-vect-stmts.c (vect_gen_while_not): New function.
3365 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3366 Alan Hayward <alan.hayward@arm.com>
3367 David Sherwood <david.sherwood@arm.com>
3369 * doc/sourcebuild.texi (vect_fully_masked): Document.
3370 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3371 default value to 0.
3372 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3373 split out from...
3374 (vect_analyze_loop_2): ...here. Don't check the vectorization
3375 factor against the number of loop iterations if the loop is
3376 fully-masked.
3378 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3379 Alan Hayward <alan.hayward@arm.com>
3380 David Sherwood <david.sherwood@arm.com>
3382 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3383 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3384 (dump_groups): Update accordingly.
3385 (iv_use::mem_type): New member variable.
3386 (address_p): New function.
3387 (record_use): Add a mem_type argument and initialize the new
3388 mem_type field.
3389 (record_group_use): Add a mem_type argument. Use address_p.
3390 Remove obsolete null checks of base_object. Update call to record_use.
3391 (find_interesting_uses_op): Update call to record_group_use.
3392 (find_interesting_uses_cond): Likewise.
3393 (find_interesting_uses_address): Likewise.
3394 (get_mem_type_for_internal_fn): New function.
3395 (find_address_like_use): Likewise.
3396 (find_interesting_uses_stmt): Try find_address_like_use before
3397 calling find_interesting_uses_op.
3398 (addr_offset_valid_p): Use the iv mem_type field as the type
3399 of the addressed memory.
3400 (add_autoinc_candidates): Likewise.
3401 (get_address_cost): Likewise.
3402 (split_small_address_groups_p): Use address_p.
3403 (split_address_groups): Likewise.
3404 (add_iv_candidate_for_use): Likewise.
3405 (autoinc_possible_for_pair): Likewise.
3406 (rewrite_groups): Likewise.
3407 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3408 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3409 (get_alias_ptr_type_for_ptr_address): New function.
3410 (rewrite_use_address): Rewrite address uses in calls that were
3411 identified by find_address_like_use.
3413 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3414 Alan Hayward <alan.hayward@arm.com>
3415 David Sherwood <david.sherwood@arm.com>
3417 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3418 TARGET_MEM_REFs.
3419 * gimple-expr.h (is_gimple_addressable: Likewise.
3420 * gimple-expr.c (is_gimple_address): Likewise.
3421 * internal-fn.c (expand_call_mem_ref): New function.
3422 (expand_mask_load_optab_fn): Use it.
3423 (expand_mask_store_optab_fn): Likewise.
3425 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3426 Alan Hayward <alan.hayward@arm.com>
3427 David Sherwood <david.sherwood@arm.com>
3429 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3430 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3431 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3432 (cond_umax@var{mode}): Document.
3433 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3434 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3435 (cond_umin_optab, cond_umax_optab): New optabs.
3436 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3437 (COND_IOR, COND_XOR): New internal functions.
3438 * internal-fn.h (get_conditional_internal_fn): Declare.
3439 * internal-fn.c (cond_binary_direct): New macro.
3440 (expand_cond_binary_optab_fn): Likewise.
3441 (direct_cond_binary_optab_supported_p): Likewise.
3442 (get_conditional_internal_fn): New function.
3443 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3444 Cope with reduction statements that are vectorized as calls rather
3445 than assignments.
3446 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3447 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3448 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3449 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3450 (UNSPEC_COND_EOR): New unspecs.
3451 (optab): Add mappings for them.
3452 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3453 (sve_int_op, sve_fp_op): New int attributes.
3455 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3456 Alan Hayward <alan.hayward@arm.com>
3457 David Sherwood <david.sherwood@arm.com>
3459 * optabs.def (while_ult_optab): New optab.
3460 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3461 * internal-fn.def (WHILE_ULT): New internal function.
3462 * internal-fn.h (direct_internal_fn_supported_p): New override
3463 that takes two types as argument.
3464 * internal-fn.c (while_direct): New macro.
3465 (expand_while_optab_fn): New function.
3466 (convert_optab_supported_p): Likewise.
3467 (direct_while_optab_supported_p): New macro.
3468 * wide-int.h (wi::udiv_ceil): New function.
3469 * tree-vectorizer.h (rgroup_masks): New structure.
3470 (vec_loop_masks): New typedef.
3471 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3472 and fully_masked_p.
3473 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3474 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3475 (vect_max_vf): New function.
3476 (slpeel_make_loop_iterate_ntimes): Delete.
3477 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3478 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3479 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3480 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3481 internal-fn.h, stor-layout.h and optabs-query.h.
3482 (vect_set_loop_mask): New function.
3483 (add_preheader_seq): Likewise.
3484 (add_header_seq): Likewise.
3485 (interleave_supported_p): Likewise.
3486 (vect_maybe_permute_loop_masks): Likewise.
3487 (vect_set_loop_masks_directly): Likewise.
3488 (vect_set_loop_condition_masked): Likewise.
3489 (vect_set_loop_condition_unmasked): New function, split out from
3490 slpeel_make_loop_iterate_ntimes.
3491 (slpeel_make_loop_iterate_ntimes): Rename to..
3492 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3493 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3494 (vect_do_peeling): Update call accordingly.
3495 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3496 loops.
3497 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3498 mask_compare_type, can_fully_mask_p and fully_masked_p.
3499 (release_vec_loop_masks): New function.
3500 (_loop_vec_info): Use it to free the loop masks.
3501 (can_produce_all_loop_masks_p): New function.
3502 (vect_get_max_nscalars_per_iter): Likewise.
3503 (vect_verify_full_masking): Likewise.
3504 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3505 retries, and free the mask rgroups before retrying. Check loop-wide
3506 reasons for disallowing fully-masked loops. Make the final decision
3507 about whether use a fully-masked loop or not.
3508 (vect_estimate_min_profitable_iters): Do not assume that peeling
3509 for the number of iterations will be needed for fully-masked loops.
3510 (vectorizable_reduction): Disable fully-masked loops.
3511 (vectorizable_live_operation): Likewise.
3512 (vect_halve_mask_nunits): New function.
3513 (vect_double_mask_nunits): Likewise.
3514 (vect_record_loop_mask): Likewise.
3515 (vect_get_loop_mask): Likewise.
3516 (vect_transform_loop): Handle the case in which the final loop
3517 iteration might handle a partial vector. Call vect_set_loop_condition
3518 instead of slpeel_make_loop_iterate_ntimes.
3519 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3520 (check_load_store_masking): New function.
3521 (prepare_load_store_mask): Likewise.
3522 (vectorizable_store): Handle fully-masked loops.
3523 (vectorizable_load): Likewise.
3524 (supportable_widening_operation): Use vect_halve_mask_nunits for
3525 booleans.
3526 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3527 (vect_gen_while): New function.
3528 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3529 (aarch64_uqdec<mode>): New insn.
3531 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3532 Alan Hayward <alan.hayward@arm.com>
3533 David Sherwood <david.sherwood@arm.com>
3535 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3536 (reduc_xor_scal_optab): New optabs.
3537 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3538 (reduc_xor_scal_@var{m}): Document.
3539 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3540 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3541 internal functions.
3542 * fold-const-call.c (fold_const_call): Handle them.
3543 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3544 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3545 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3546 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3547 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3548 (UNSPEC_XORV): New unspecs.
3549 (optab): Add entries for them.
3550 (BITWISEV): New int iterator.
3551 (bit_reduc_op): New int attributes.
3553 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3554 Alan Hayward <alan.hayward@arm.com>
3555 David Sherwood <david.sherwood@arm.com>
3557 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3558 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3559 * optabs.def (vec_shl_insert_optab): New optab.
3560 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3561 (duplicate_and_interleave): Likewise.
3562 * tree-vect-loop.c: Include internal-fn.h.
3563 (neutral_op_for_slp_reduction): New function, split out from
3564 get_initial_defs_for_reduction.
3565 (get_initial_def_for_reduction): Handle option 2 for variable-length
3566 vectors by loading the neutral value into a vector and then shifting
3567 the initial value into element 0.
3568 (get_initial_defs_for_reduction): Replace the code argument with
3569 the neutral value calculated by neutral_op_for_slp_reduction.
3570 Use gimple_build_vector for constant-length vectors.
3571 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3572 but the first group_size elements have a neutral value.
3573 Use duplicate_and_interleave otherwise.
3574 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3575 Update call to get_initial_defs_for_reduction. Handle SLP
3576 reductions for variable-length vectors by creating one vector
3577 result for each scalar result, with the elements associated
3578 with other scalar results stubbed out with the neutral value.
3579 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3580 Require IFN_VEC_SHL_INSERT for double reductions on
3581 variable-length vectors, or SLP reductions that have
3582 a neutral value. Require can_duplicate_and_interleave_p
3583 support for variable-length unchained SLP reductions if there
3584 is no neutral value, such as for MIN/MAX reductions. Also require
3585 the number of vector elements to be a multiple of the number of
3586 SLP statements when doing variable-length unchained SLP reductions.
3587 Update call to vect_create_epilog_for_reduction.
3588 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3589 and remove initial values.
3590 (duplicate_and_interleave): Make public.
3591 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3592 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3594 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3595 Alan Hayward <alan.hayward@arm.com>
3596 David Sherwood <david.sherwood@arm.com>
3598 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3599 (can_duplicate_and_interleave_p): New function.
3600 (vect_get_and_check_slp_defs): Take the vector of statements
3601 rather than just the current one. Remove excess parentheses.
3602 Restriction rejectinon of vect_constant_def and vect_external_def
3603 for variable-length vectors to boolean types, or types for which
3604 can_duplicate_and_interleave_p is false.
3605 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3606 (duplicate_and_interleave): New function.
3607 (vect_get_constant_vectors): Use gimple_build_vector for
3608 constant-length vectors and suitable variable-length constant
3609 vectors. Use duplicate_and_interleave for other variable-length
3610 vectors. Don't defer the update when inserting new statements.
3612 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3613 Alan Hayward <alan.hayward@arm.com>
3614 David Sherwood <david.sherwood@arm.com>
3616 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3617 min_profitable_iters doesn't go negative.
3619 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3620 Alan Hayward <alan.hayward@arm.com>
3621 David Sherwood <david.sherwood@arm.com>
3623 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3624 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3625 * optabs.def (vec_mask_load_lanes_optab): New optab.
3626 (vec_mask_store_lanes_optab): Likewise.
3627 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3628 (MASK_STORE_LANES): Likewise.
3629 * internal-fn.c (mask_load_lanes_direct): New macro.
3630 (mask_store_lanes_direct): Likewise.
3631 (expand_mask_load_optab_fn): Handle masked operations.
3632 (expand_mask_load_lanes_optab_fn): New macro.
3633 (expand_mask_store_optab_fn): Handle masked operations.
3634 (expand_mask_store_lanes_optab_fn): New macro.
3635 (direct_mask_load_lanes_optab_supported_p): Likewise.
3636 (direct_mask_store_lanes_optab_supported_p): Likewise.
3637 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3638 parameter.
3639 (vect_load_lanes_supported): Likewise.
3640 * tree-vect-data-refs.c (strip_conversion): New function.
3641 (can_group_stmts_p): Likewise.
3642 (vect_analyze_data_ref_accesses): Use it instead of checking
3643 for a pair of assignments.
3644 (vect_store_lanes_supported): Take a masked_p parameter.
3645 (vect_load_lanes_supported): Likewise.
3646 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3647 vect_store_lanes_supported and vect_load_lanes_supported.
3648 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3649 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3650 parameter. Don't allow gaps for masked accesses.
3651 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3652 and vect_load_lanes_supported.
3653 (get_load_store_type): Take a masked_p parameter and update
3654 call to get_group_load_store_type.
3655 (vectorizable_store): Update call to get_load_store_type.
3656 Handle IFN_MASK_STORE_LANES.
3657 (vectorizable_load): Update call to get_load_store_type.
3658 Handle IFN_MASK_LOAD_LANES.
3660 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3661 Alan Hayward <alan.hayward@arm.com>
3662 David Sherwood <david.sherwood@arm.com>
3664 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3665 modes for SVE.
3666 * config/aarch64/aarch64-protos.h
3667 (aarch64_sve_struct_memory_operand_p): Declare.
3668 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3669 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3670 (VPRED, vpred): Handle SVE structure modes.
3671 * config/aarch64/constraints.md (Utx): New constraint.
3672 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3673 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3674 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3675 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3676 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3677 structure modes. Split into pieces after RA.
3678 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3679 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3680 New patterns.
3681 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3682 SVE structure modes.
3683 (aarch64_classify_address): Likewise.
3684 (sizetochar): Move earlier in file.
3685 (aarch64_print_operand): Handle SVE register lists.
3686 (aarch64_array_mode): New function.
3687 (aarch64_sve_struct_memory_operand_p): Likewise.
3688 (TARGET_ARRAY_MODE): Redefine.
3690 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3691 Alan Hayward <alan.hayward@arm.com>
3692 David Sherwood <david.sherwood@arm.com>
3694 * target.def (array_mode): New target hook.
3695 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3696 * doc/tm.texi: Regenerate.
3697 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3698 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3699 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3700 targetm.array_mode.
3701 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3702 type sizes.
3704 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3705 Alan Hayward <alan.hayward@arm.com>
3706 David Sherwood <david.sherwood@arm.com>
3708 * fold-const.c (fold_binary_loc): Check the argument types
3709 rather than the result type when testing for a vector operation.
3711 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3713 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3714 * doc/tm.texi: Regenerate.
3716 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3717 Alan Hayward <alan.hayward@arm.com>
3718 David Sherwood <david.sherwood@arm.com>
3720 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3721 (sve): Document new AArch64 extension.
3722 * doc/md.texi (w): Extend the description of the AArch64
3723 constraint to include SVE vectors.
3724 (Upl, Upa): Document new AArch64 predicate constraints.
3725 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3726 enum.
3727 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3728 (msve-vector-bits=): New option.
3729 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3730 SVE when these are disabled.
3731 (sve): New extension.
3732 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3733 modes. Adjust their number of units based on aarch64_sve_vg.
3734 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3735 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3736 aarch64_addr_query_type.
3737 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3738 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3739 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3740 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3741 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3742 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3743 (aarch64_simd_imm_zero_p): Delete.
3744 (aarch64_check_zero_based_sve_index_immediate): Declare.
3745 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3746 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3747 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3748 (aarch64_sve_float_mul_immediate_p): Likewise.
3749 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3750 rather than an rtx.
3751 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3752 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3753 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3754 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3755 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3756 (aarch64_regmode_natural_size): Likewise.
3757 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3758 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3759 left one place.
3760 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3761 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3762 for VG and the SVE predicate registers.
3763 (V_ALIASES): Add a "z"-prefixed alias.
3764 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3765 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3766 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3767 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3768 (REG_CLASS_NAMES): Add entries for them.
3769 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3770 and the predicate registers.
3771 (aarch64_sve_vg): Declare.
3772 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3773 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3774 (REGMODE_NATURAL_SIZE): Define.
3775 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3776 SVE macros.
3777 * config/aarch64/aarch64.c: Include cfgrtl.h.
3778 (simd_immediate_info): Add a constructor for series vectors,
3779 and an associated step field.
3780 (aarch64_sve_vg): New variable.
3781 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3782 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3783 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3784 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3785 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3786 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3787 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3788 (aarch64_get_mask_mode): New functions.
3789 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3790 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3791 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3792 predicate modes and predicate registers. Explicitly restrict
3793 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3794 to store a vector mode if it is recognized by
3795 aarch64_classify_vector_mode.
3796 (aarch64_regmode_natural_size): New function.
3797 (aarch64_hard_regno_caller_save_mode): Return the original mode
3798 for predicates.
3799 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3800 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3801 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3802 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3803 functions.
3804 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3805 does not overlap dest if the function is frame-related. Handle
3806 SVE constants.
3807 (aarch64_split_add_offset): New function.
3808 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3809 them aarch64_add_offset.
3810 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3811 and update call to aarch64_sub_sp.
3812 (aarch64_add_cfa_expression): New function.
3813 (aarch64_expand_prologue): Pass extra temporary registers to the
3814 functions above. Handle the case in which we need to emit new
3815 DW_CFA_expressions for registers that were originally saved
3816 relative to the stack pointer, but now have to be expressed
3817 relative to the frame pointer.
3818 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3819 functions above.
3820 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3821 IP0 and IP1 values for SVE frames.
3822 (aarch64_expand_vec_series): New function.
3823 (aarch64_expand_sve_widened_duplicate): Likewise.
3824 (aarch64_expand_sve_const_vector): Likewise.
3825 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3826 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3827 into the register, rather than emitting a SET directly.
3828 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3829 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3830 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3831 (offset_9bit_signed_scaled_p): New functions.
3832 (aarch64_replicate_bitmask_imm): New function.
3833 (aarch64_bitmask_imm): Use it.
3834 (aarch64_cannot_force_const_mem): Reject expressions involving
3835 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3836 (aarch64_classify_index): Handle SVE indices, by requiring
3837 a plain register index with a scale that matches the element size.
3838 (aarch64_classify_address): Handle SVE addresses. Assert that
3839 the mode of the address is VOIDmode or an integer mode.
3840 Update call to aarch64_classify_symbol.
3841 (aarch64_classify_symbolic_expression): Update call to
3842 aarch64_classify_symbol.
3843 (aarch64_const_vec_all_in_range_p): New function.
3844 (aarch64_print_vector_float_operand): Likewise.
3845 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3846 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3847 and the FP immediates 1.0 and 0.5.
3848 (aarch64_print_address_internal): Handle SVE addresses.
3849 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3850 (aarch64_regno_regclass): Handle predicate registers.
3851 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3852 data modes.
3853 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3854 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3855 (aarch64_convert_sve_vector_bits): New function.
3856 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3857 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3858 rather than an rtx.
3859 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3860 Handle SVE vector and predicate modes. Accept VL-based constants
3861 that need only one temporary register, and VL offsets that require
3862 no temporary registers.
3863 (aarch64_conditional_register_usage): Mark the predicate registers
3864 as fixed if SVE isn't available.
3865 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3866 Return true for SVE vector and predicate modes.
3867 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3868 rather than an unsigned int. Handle SVE modes.
3869 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3870 SVE modes.
3871 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3872 if SVE is enabled.
3873 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3874 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3875 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3876 (aarch64_sve_float_mul_immediate_p): New functions.
3877 (aarch64_sve_valid_immediate): New function.
3878 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3879 Explicitly reject structure modes. Check for INDEX constants.
3880 Handle PTRUE and PFALSE constants.
3881 (aarch64_check_zero_based_sve_index_immediate): New function.
3882 (aarch64_simd_imm_zero_p): Delete.
3883 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3884 vector modes. Accept constants in the range of CNT[BHWD].
3885 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3886 ask for an Advanced SIMD mode.
3887 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3888 (aarch64_simd_vector_alignment): Handle SVE predicates.
3889 (aarch64_vectorize_preferred_vector_alignment): New function.
3890 (aarch64_simd_vector_alignment_reachable): Use it instead of
3891 the vector size.
3892 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3893 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3894 functions.
3895 (MAX_VECT_LEN): Delete.
3896 (expand_vec_perm_d): Add a vec_flags field.
3897 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3898 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3899 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3900 for SVE modes.
3901 (aarch64_evpc_rev): Rename to...
3902 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3903 (aarch64_evpc_rev_global): New function.
3904 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3905 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3906 MAX_VECT_LEN.
3907 (aarch64_evpc_sve_tbl): New function.
3908 (aarch64_expand_vec_perm_const_1): Update after rename of
3909 aarch64_evpc_rev. Handle SVE permutes too, trying
3910 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3911 than aarch64_evpc_tbl.
3912 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3913 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3914 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3915 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3916 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3917 (aarch64_expand_sve_vcond): New functions.
3918 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3919 of aarch64_vector_mode_p.
3920 (aarch64_dwarf_poly_indeterminate_value): New function.
3921 (aarch64_compute_pressure_classes): Likewise.
3922 (aarch64_can_change_mode_class): Likewise.
3923 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3924 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3925 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3926 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3927 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3928 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3929 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3930 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3931 constraints.
3932 (Dn, Dl, Dr): Accept const as well as const_vector.
3933 (Dz): Likewise. Compare against CONST0_RTX.
3934 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3935 of "vector" where appropriate.
3936 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3937 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3938 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3939 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3940 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3941 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3942 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3943 (v_int_equiv): Extend to SVE modes.
3944 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3945 mode attributes.
3946 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3947 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3948 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3949 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3950 (SVE_COND_FP_CMP): New int iterators.
3951 (perm_hilo): Handle the new unpack unspecs.
3952 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3953 attributes.
3954 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3955 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3956 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3957 (aarch64_equality_operator, aarch64_constant_vector_operand)
3958 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3959 (aarch64_sve_nonimmediate_operand): Likewise.
3960 (aarch64_sve_general_operand): Likewise.
3961 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3962 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3963 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3964 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3965 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3966 (aarch64_sve_float_arith_immediate): Likewise.
3967 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3968 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3969 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3970 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3971 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3972 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3973 (aarch64_sve_float_arith_operand): Likewise.
3974 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3975 (aarch64_sve_float_mul_operand): Likewise.
3976 (aarch64_sve_vec_perm_operand): Likewise.
3977 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3978 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3979 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3980 as well as const_vector.
3981 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3982 in file. Use CONST0_RTX and CONSTM1_RTX.
3983 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3984 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3985 Use aarch64_simd_imm_zero.
3986 * config/aarch64/aarch64-sve.md: New file.
3987 * config/aarch64/aarch64.md: Include it.
3988 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3989 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3990 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3991 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3992 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3993 (sve): New attribute.
3994 (enabled): Disable instructions with the sve attribute unless
3995 TARGET_SVE.
3996 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3997 aarch64_expand_mov_immediate.
3998 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3999 CNT[BHSD] immediates.
4000 (movti): Split CONST_POLY_INT moves into two halves.
4001 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4002 Split additions that need a temporary here if the destination
4003 is the stack pointer.
4004 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4005 (*add<mode>3_poly_1): New instruction.
4006 (set_clobber_cc): New expander.
4008 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4010 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4011 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4012 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4013 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4014 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4015 Change innermode from fixed_mode_size to machine_mode.
4016 (simplify_subreg): Update call accordingly. Handle a constant-sized
4017 subreg of a variable-length CONST_VECTOR.
4019 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4020 Alan Hayward <alan.hayward@arm.com>
4021 David Sherwood <david.sherwood@arm.com>
4023 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4024 (add_offset_to_base): New function, split out from...
4025 (create_mem_ref): ...here. When handling a scale other than 1,
4026 check first whether the address is valid without the offset.
4027 Add it into the base if so, leaving the index and scale as-is.
4029 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4031 PR c++/83778
4032 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4033 fold_for_warn before checking if arg2 is INTEGER_CST.
4035 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4037 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4038 (store_multiple_operation): Delete.
4039 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4040 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4041 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4042 guarded by TARGET_STRING.
4043 (rs6000_output_load_multiple): Delete.
4044 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4045 OPTION_MASK_STRING / TARGET_STRING handling.
4046 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4047 (const rs6000_opt_masks) <"string">: Change mask to 0.
4048 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4049 (MASK_STRING): Delete.
4050 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4051 parts. Simplify.
4052 (load_multiple): Delete.
4053 (*ldmsi8): Delete.
4054 (*ldmsi7): Delete.
4055 (*ldmsi6): Delete.
4056 (*ldmsi5): Delete.
4057 (*ldmsi4): Delete.
4058 (*ldmsi3): Delete.
4059 (store_multiple): Delete.
4060 (*stmsi8): Delete.
4061 (*stmsi7): Delete.
4062 (*stmsi6): Delete.
4063 (*stmsi5): Delete.
4064 (*stmsi4): Delete.
4065 (*stmsi3): Delete.
4066 (movmemsi_8reg): Delete.
4067 (corresponding unnamed define_insn): Delete.
4068 (movmemsi_6reg): Delete.
4069 (corresponding unnamed define_insn): Delete.
4070 (movmemsi_4reg): Delete.
4071 (corresponding unnamed define_insn): Delete.
4072 (movmemsi_2reg): Delete.
4073 (corresponding unnamed define_insn): Delete.
4074 (movmemsi_1reg): Delete.
4075 (corresponding unnamed define_insn): Delete.
4076 * config/rs6000/rs6000.opt (mno-string): New.
4077 (mstring): Replace by deprecation warning stub.
4078 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4080 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4082 * regrename.c (regrename_do_replace): If replacing the same
4083 reg multiple times, try to reuse last created gen_raw_REG.
4085 PR debug/81155
4086 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4087 main to workaround a bug in GDB.
4089 2018-01-12 Tom de Vries <tom@codesourcery.com>
4091 PR target/83737
4092 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4094 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4096 PR rtl-optimization/80481
4097 * ira-color.c (get_cap_member): New function.
4098 (allocnos_conflict_by_live_ranges_p): Use it.
4099 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4100 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4102 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4104 PR target/83628
4105 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4106 (*saddl_se_1): Ditto.
4107 (*ssubsi_1): Ditto.
4108 (*ssubl_se_1): Ditto.
4110 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4112 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4113 rather than wi::to_widest for DR_INITs.
4114 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4115 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4116 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4117 INTEGER_CSTs.
4118 (vect_analyze_group_access_1): Note that here.
4120 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4122 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4123 polynomial type sizes.
4125 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4127 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4128 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4129 (gimple_add_tmp_var): Likewise.
4131 2018-01-12 Martin Liska <mliska@suse.cz>
4133 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4134 (gimple_alloc_sizes): Likewise.
4135 (dump_gimple_statistics): Use PRIu64 in printf format.
4136 * gimple.h: Change uint64_t to int.
4138 2018-01-12 Martin Liska <mliska@suse.cz>
4140 * tree-core.h: Use uint64_t instead of int.
4141 * tree.c (tree_node_counts): Likewise.
4142 (tree_node_sizes): Likewise.
4143 (dump_tree_statistics): Use PRIu64 in printf format.
4145 2018-01-12 Martin Liska <mliska@suse.cz>
4147 * Makefile.in: As qsort_chk is implemented in vec.c, add
4148 vec.o to linkage of gencfn-macros.
4149 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4150 passing the info to record_node_allocation_statistics.
4151 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4152 and pass the info.
4153 * ggc-common.c (struct ggc_usage): Add operator== and use
4154 it in operator< and compare function.
4155 * mem-stats.h (struct mem_usage): Likewise.
4156 * vec.c (struct vec_usage): Remove operator< and compare
4157 function. Can be simply inherited.
4159 2018-01-12 Martin Jambor <mjambor@suse.cz>
4161 PR target/81616
4162 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4163 * tree-ssa-math-opts.c: Include domwalk.h.
4164 (convert_mult_to_fma_1): New function.
4165 (fma_transformation_info): New type.
4166 (fma_deferring_state): Likewise.
4167 (cancel_fma_deferring): New function.
4168 (result_of_phi): Likewise.
4169 (last_fma_candidate_feeds_initial_phi): Likewise.
4170 (convert_mult_to_fma): Added deferring logic, split actual
4171 transformation to convert_mult_to_fma_1.
4172 (math_opts_dom_walker): New type.
4173 (math_opts_dom_walker::after_dom_children): New method, body moved
4174 here from pass_optimize_widening_mul::execute, added deferring logic
4175 bits.
4176 (pass_optimize_widening_mul::execute): Moved most of code to
4177 math_opts_dom_walker::after_dom_children.
4178 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4179 * config/i386/i386.c (ix86_option_override_internal): Added
4180 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4182 2018-01-12 Richard Biener <rguenther@suse.de>
4184 PR debug/83157
4185 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4186 inline instance vars.
4188 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4190 PR target/81819
4191 * config/rx/rx.c (rx_is_restricted_memory_address):
4192 Handle SUBREG case.
4194 2018-01-12 Richard Biener <rguenther@suse.de>
4196 PR tree-optimization/80846
4197 * target.def (split_reduction): New target hook.
4198 * targhooks.c (default_split_reduction): New function.
4199 * targhooks.h (default_split_reduction): Declare.
4200 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4201 target requests first reduce vectors by combining low and high
4202 parts.
4203 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4204 (get_vectype_for_scalar_type_and_size): Export.
4205 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4206 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4207 * doc/tm.texi: Regenerate.
4208 * config/i386/i386.c (ix86_split_reduction): Implement
4209 TARGET_VECTORIZE_SPLIT_REDUCTION.
4211 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4213 PR target/83368
4214 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4215 in PIC mode except for TARGET_VXWORKS_RTP.
4216 * config/sparc/sparc.c: Include cfgrtl.h.
4217 (TARGET_INIT_PIC_REG): Define.
4218 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4219 (sparc_pic_register_p): New predicate.
4220 (sparc_legitimate_address_p): Use it.
4221 (sparc_legitimize_pic_address): Likewise.
4222 (sparc_delegitimize_address): Likewise.
4223 (sparc_mode_dependent_address_p): Likewise.
4224 (gen_load_pcrel_sym): Remove 4th parameter.
4225 (load_got_register): Adjust call to above. Remove obsolete stuff.
4226 (sparc_expand_prologue): Do not call load_got_register here.
4227 (sparc_flat_expand_prologue): Likewise.
4228 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4229 (sparc_use_pseudo_pic_reg): New function.
4230 (sparc_init_pic_reg): Likewise.
4231 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4232 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4234 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4236 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4237 Add item for branch_cost.
4239 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4241 PR rtl-optimization/83565
4242 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4243 not extend the result to a larger mode for rotate operations.
4244 (num_sign_bit_copies1): Likewise.
4246 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4248 PR target/40411
4249 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4250 -symbolic.
4251 Use values-Xc.o for -pedantic.
4252 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4254 2018-01-12 Martin Liska <mliska@suse.cz>
4256 PR ipa/83054
4257 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4258 New function.
4259 (possible_polymorphic_call_targets): Use it.
4260 (ipa_devirt): Likewise.
4262 2018-01-12 Martin Liska <mliska@suse.cz>
4264 * profile-count.h (enum profile_quality): Use 0 as invalid
4265 enum value of profile_quality.
4267 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4269 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4270 -mext-string options.
4272 2018-01-12 Richard Biener <rguenther@suse.de>
4274 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4275 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4276 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4277 Likewise.
4278 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4280 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4282 * configure.ac (--with-long-double-format): Add support for the
4283 configuration option to change the default long double format on
4284 PowerPC systems.
4285 * config.gcc (powerpc*-linux*-*): Likewise.
4286 * configure: Regenerate.
4287 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4288 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4289 used without modification.
4291 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4293 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4294 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4295 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4296 MISC_BUILTIN_SPEC_BARRIER.
4297 (rs6000_init_builtins): Likewise.
4298 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4299 enum value.
4300 (speculation_barrier): New define_insn.
4301 * doc/extend.texi: Document __builtin_speculation_barrier.
4303 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4305 PR target/83203
4306 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4307 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4308 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4309 iterators.
4310 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4311 integral modes instead of "ss" and "sd".
4312 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4313 vectors with 32-bit and 64-bit elements.
4314 (vecdupssescalarmodesuffix): New mode attribute.
4315 (vec_dup<mode>): Use it.
4317 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4319 PR target/83330
4320 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4321 frame if argument is passed on stack.
4323 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4325 PR target/82682
4326 * ree.c (combine_reaching_defs): Optimize also
4327 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4328 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4330 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4332 PR middle-end/83189
4333 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4335 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4337 PR middle-end/83718
4338 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4339 after they are computed.
4341 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4343 PR tree-optimization/83695
4344 * gimple-loop-linterchange.cc
4345 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4346 reset cached scev information after interchange.
4347 (pass_linterchange::execute): Remove call to scev_reset_htab.
4349 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4351 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4352 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4353 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4354 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4355 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4356 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4357 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4358 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4359 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4360 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4361 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4362 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4363 (V_lane_reg): Likewise.
4364 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4365 New define_expand.
4366 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4367 (vfmal_lane_low<mode>_intrinsic,
4368 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4369 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4370 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4371 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4372 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4373 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4375 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4377 * config/arm/arm-cpus.in (fp16fml): New feature.
4378 (ALL_SIMD): Add fp16fml.
4379 (armv8.2-a): Add fp16fml as an option.
4380 (armv8.3-a): Likewise.
4381 (armv8.4-a): Add fp16fml as part of fp16.
4382 * config/arm/arm.h (TARGET_FP16FML): Define.
4383 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4384 when appropriate.
4385 * config/arm/arm-modes.def (V2HF): Define.
4386 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4387 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4388 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4389 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4390 vfmsl_low, vfmsl_high): New set of builtins.
4391 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4392 (vfml_op): New code attribute.
4393 (VFMLHALVES): New int iterator.
4394 (VFML, VFMLSEL): New mode attributes.
4395 (V_reg): Define mapping for V2HF.
4396 (V_hi, V_lo): New mode attributes.
4397 (VF_constraint): Likewise.
4398 (vfml_half, vfml_half_selector): New int attributes.
4399 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4400 define_expand.
4401 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4402 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4403 New define_insn.
4404 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4405 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4406 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4407 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4408 documentation.
4409 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4410 Document new effective target and option set.
4412 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4414 * config/arm/arm-cpus.in (armv8_4): New feature.
4415 (ARMv8_4a): New fgroup.
4416 (armv8.4-a): New arch.
4417 * config/arm/arm-tables.opt: Regenerate.
4418 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4419 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4420 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4421 Add matching rules for -march=armv8.4-a and extensions.
4422 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4424 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4426 PR target/81821
4427 * config/rx/rx.md (BW): New mode attribute.
4428 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4430 2018-01-11 Richard Biener <rguenther@suse.de>
4432 PR tree-optimization/83435
4433 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4434 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4435 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4437 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4438 Alan Hayward <alan.hayward@arm.com>
4439 David Sherwood <david.sherwood@arm.com>
4441 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4442 field.
4443 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4444 (aarch64_print_address_internal): Use it to check for a zero offset.
4446 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4447 Alan Hayward <alan.hayward@arm.com>
4448 David Sherwood <david.sherwood@arm.com>
4450 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4451 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4452 Return a poly_int64 rather than a HOST_WIDE_INT.
4453 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4454 rather than a HOST_WIDE_INT.
4455 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4456 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4457 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4458 final_offset from HOST_WIDE_INT to poly_int64.
4459 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4460 to_constant when getting the number of units in an Advanced SIMD
4461 mode.
4462 (aarch64_builtin_vectorized_function): Check for a constant number
4463 of units.
4464 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4465 GET_MODE_SIZE.
4466 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4467 attribute instead of GET_MODE_NUNITS.
4468 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4469 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4470 GET_MODE_SIZE for fixed-size registers.
4471 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4472 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4473 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4474 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4475 (aarch64_print_operand, aarch64_print_address_internal)
4476 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4477 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4478 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4479 Handle polynomial GET_MODE_SIZE.
4480 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4481 wider than SImode without modification.
4482 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4483 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4484 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4485 passing and returning SVE modes.
4486 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4487 rather than GEN_INT.
4488 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4489 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4490 (aarch64_allocate_and_probe_stack_space): Likewise.
4491 (aarch64_layout_frame): Cope with polynomial offsets.
4492 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4493 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4494 polynomial offsets.
4495 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4496 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4497 poly_int64 rather than a HOST_WIDE_INT.
4498 (aarch64_get_separate_components, aarch64_process_components)
4499 (aarch64_expand_prologue, aarch64_expand_epilogue)
4500 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4501 (aarch64_anchor_offset): New function, split out from...
4502 (aarch64_legitimize_address): ...here.
4503 (aarch64_builtin_vectorization_cost): Handle polynomial
4504 TYPE_VECTOR_SUBPARTS.
4505 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4506 GET_MODE_NUNITS.
4507 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4508 number of elements from the PARALLEL rather than the mode.
4509 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4510 rather than GET_MODE_BITSIZE.
4511 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4512 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4513 (aarch64_expand_vec_perm_const_1): Handle polynomial
4514 d->perm.length () and d->perm elements.
4515 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4516 Apply to_constant to d->perm elements.
4517 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4518 polynomial CONST_VECTOR_NUNITS.
4519 (aarch64_move_pointer): Take amount as a poly_int64 rather
4520 than an int.
4521 (aarch64_progress_pointer): Avoid temporary variable.
4522 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4523 the mode attribute instead of GET_MODE.
4525 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4526 Alan Hayward <alan.hayward@arm.com>
4527 David Sherwood <david.sherwood@arm.com>
4529 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4530 x exists before using it.
4531 (aarch64_add_constant_internal): Rename to...
4532 (aarch64_add_offset_1): ...this. Replace regnum with separate
4533 src and dest rtxes. Handle the case in which they're different,
4534 including when the offset is zero. Replace scratchreg with an rtx.
4535 Use 2 additions if there is no spare register into which we can
4536 move a 16-bit constant.
4537 (aarch64_add_constant): Delete.
4538 (aarch64_add_offset): Replace reg with separate src and dest
4539 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4540 Use aarch64_add_offset_1.
4541 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4542 an rtx rather than an int. Take the delta as a poly_int64
4543 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4544 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4545 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4546 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4547 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4548 and aarch64_add_sp.
4549 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4550 aarch64_add_constant.
4552 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4554 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4555 Use scalar_float_mode.
4557 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4559 * config/aarch64/aarch64-simd.md
4560 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4561 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4562 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4563 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4564 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4565 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4566 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4567 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4568 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4569 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4571 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4573 PR target/83514
4574 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4575 targ_options->x_arm_arch_string is non NULL.
4577 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4579 * config/aarch64/aarch64.h
4580 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4582 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4584 PR target/82096
4585 * expmed.c (emit_store_flag_force): Swap if const op0
4586 and change VOIDmode to mode of op0.
4588 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4590 PR rtl-optimization/83761
4591 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4592 than bytes to mode_for_size.
4594 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4596 PR middle-end/83189
4597 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4598 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4599 profile.
4601 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4603 PR middle-end/83575
4604 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4605 when in layout mode.
4606 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4607 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4608 partition fixup.
4610 2018-01-10 Michael Collison <michael.collison@arm.com>
4612 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4613 * config/aarch64/aarch64-option-extension.def: Add
4614 AARCH64_OPT_EXTENSION of 'fp16fml'.
4615 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4616 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4617 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4618 * config/aarch64/constraints.md (Ui7): New constraint.
4619 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4620 (VFMLA_SEL_W): Ditto.
4621 (f16quad): Ditto.
4622 (f16mac1): Ditto.
4623 (VFMLA16_LOW): New int iterator.
4624 (VFMLA16_HIGH): Ditto.
4625 (UNSPEC_FMLAL): New unspec.
4626 (UNSPEC_FMLSL): Ditto.
4627 (UNSPEC_FMLAL2): Ditto.
4628 (UNSPEC_FMLSL2): Ditto.
4629 (f16mac): New code attribute.
4630 * config/aarch64/aarch64-simd-builtins.def
4631 (aarch64_fmlal_lowv2sf): Ditto.
4632 (aarch64_fmlsl_lowv2sf): Ditto.
4633 (aarch64_fmlalq_lowv4sf): Ditto.
4634 (aarch64_fmlslq_lowv4sf): Ditto.
4635 (aarch64_fmlal_highv2sf): Ditto.
4636 (aarch64_fmlsl_highv2sf): Ditto.
4637 (aarch64_fmlalq_highv4sf): Ditto.
4638 (aarch64_fmlslq_highv4sf): Ditto.
4639 (aarch64_fmlal_lane_lowv2sf): Ditto.
4640 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4641 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4642 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4643 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4644 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4645 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4646 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4647 (aarch64_fmlal_lane_highv2sf): Ditto.
4648 (aarch64_fmlsl_lane_highv2sf): Ditto.
4649 (aarch64_fmlal_laneq_highv2sf): Ditto.
4650 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4651 (aarch64_fmlalq_lane_highv4sf): Ditto.
4652 (aarch64_fmlsl_lane_highv4sf): Ditto.
4653 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4654 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4655 * config/aarch64/aarch64-simd.md:
4656 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4657 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4658 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4659 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4660 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4661 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4662 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4663 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4664 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4665 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4666 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4667 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4668 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4669 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4670 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4671 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4672 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4673 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4674 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4675 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4676 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4677 (vfmlsl_low_u32): Ditto.
4678 (vfmlalq_low_u32): Ditto.
4679 (vfmlslq_low_u32): Ditto.
4680 (vfmlal_high_u32): Ditto.
4681 (vfmlsl_high_u32): Ditto.
4682 (vfmlalq_high_u32): Ditto.
4683 (vfmlslq_high_u32): Ditto.
4684 (vfmlal_lane_low_u32): Ditto.
4685 (vfmlsl_lane_low_u32): Ditto.
4686 (vfmlal_laneq_low_u32): Ditto.
4687 (vfmlsl_laneq_low_u32): Ditto.
4688 (vfmlalq_lane_low_u32): Ditto.
4689 (vfmlslq_lane_low_u32): Ditto.
4690 (vfmlalq_laneq_low_u32): Ditto.
4691 (vfmlslq_laneq_low_u32): Ditto.
4692 (vfmlal_lane_high_u32): Ditto.
4693 (vfmlsl_lane_high_u32): Ditto.
4694 (vfmlal_laneq_high_u32): Ditto.
4695 (vfmlsl_laneq_high_u32): Ditto.
4696 (vfmlalq_lane_high_u32): Ditto.
4697 (vfmlslq_lane_high_u32): Ditto.
4698 (vfmlalq_laneq_high_u32): Ditto.
4699 (vfmlslq_laneq_high_u32): Ditto.
4700 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4701 (AARCH64_FL_FOR_ARCH8_4): New.
4702 (AARCH64_ISA_F16FML): New ISA flag.
4703 (TARGET_F16FML): New feature flag for fp16fml.
4704 (doc/invoke.texi): Document new fp16fml option.
4706 2018-01-10 Michael Collison <michael.collison@arm.com>
4708 * config/aarch64/aarch64-builtins.c:
4709 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4710 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4711 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4712 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4713 (AARCH64_ISA_SHA3): New ISA flag.
4714 (TARGET_SHA3): New feature flag for sha3.
4715 * config/aarch64/iterators.md (sha512_op): New int attribute.
4716 (CRYPTO_SHA512): New int iterator.
4717 (UNSPEC_SHA512H): New unspec.
4718 (UNSPEC_SHA512H2): Ditto.
4719 (UNSPEC_SHA512SU0): Ditto.
4720 (UNSPEC_SHA512SU1): Ditto.
4721 * config/aarch64/aarch64-simd-builtins.def
4722 (aarch64_crypto_sha512hqv2di): New builtin.
4723 (aarch64_crypto_sha512h2qv2di): Ditto.
4724 (aarch64_crypto_sha512su0qv2di): Ditto.
4725 (aarch64_crypto_sha512su1qv2di): Ditto.
4726 (aarch64_eor3qv8hi): Ditto.
4727 (aarch64_rax1qv2di): Ditto.
4728 (aarch64_xarqv2di): Ditto.
4729 (aarch64_bcaxqv8hi): Ditto.
4730 * config/aarch64/aarch64-simd.md:
4731 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4732 (aarch64_crypto_sha512su0qv2di): Ditto.
4733 (aarch64_crypto_sha512su1qv2di): Ditto.
4734 (aarch64_eor3qv8hi): Ditto.
4735 (aarch64_rax1qv2di): Ditto.
4736 (aarch64_xarqv2di): Ditto.
4737 (aarch64_bcaxqv8hi): Ditto.
4738 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4739 (vsha512h2q_u64): Ditto.
4740 (vsha512su0q_u64): Ditto.
4741 (vsha512su1q_u64): Ditto.
4742 (veor3q_u16): Ditto.
4743 (vrax1q_u64): Ditto.
4744 (vxarq_u64): Ditto.
4745 (vbcaxq_u16): Ditto.
4746 * config/arm/types.md (crypto_sha512): New type attribute.
4747 (crypto_sha3): Ditto.
4748 (doc/invoke.texi): Document new sha3 option.
4750 2018-01-10 Michael Collison <michael.collison@arm.com>
4752 * config/aarch64/aarch64-builtins.c:
4753 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4754 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4755 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4756 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4757 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4758 (AARCH64_ISA_SM4): New ISA flag.
4759 (TARGET_SM4): New feature flag for sm4.
4760 * config/aarch64/aarch64-simd-builtins.def
4761 (aarch64_sm3ss1qv4si): Ditto.
4762 (aarch64_sm3tt1aq4si): Ditto.
4763 (aarch64_sm3tt1bq4si): Ditto.
4764 (aarch64_sm3tt2aq4si): Ditto.
4765 (aarch64_sm3tt2bq4si): Ditto.
4766 (aarch64_sm3partw1qv4si): Ditto.
4767 (aarch64_sm3partw2qv4si): Ditto.
4768 (aarch64_sm4eqv4si): Ditto.
4769 (aarch64_sm4ekeyqv4si): Ditto.
4770 * config/aarch64/aarch64-simd.md:
4771 (aarch64_sm3ss1qv4si): Ditto.
4772 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4773 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4774 (aarch64_sm4eqv4si): Ditto.
4775 (aarch64_sm4ekeyqv4si): Ditto.
4776 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4777 (sm3part_op): Ditto.
4778 (CRYPTO_SM3TT): Ditto.
4779 (CRYPTO_SM3PART): Ditto.
4780 (UNSPEC_SM3SS1): New unspec.
4781 (UNSPEC_SM3TT1A): Ditto.
4782 (UNSPEC_SM3TT1B): Ditto.
4783 (UNSPEC_SM3TT2A): Ditto.
4784 (UNSPEC_SM3TT2B): Ditto.
4785 (UNSPEC_SM3PARTW1): Ditto.
4786 (UNSPEC_SM3PARTW2): Ditto.
4787 (UNSPEC_SM4E): Ditto.
4788 (UNSPEC_SM4EKEY): Ditto.
4789 * config/aarch64/constraints.md (Ui2): New constraint.
4790 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4791 * config/arm/types.md (crypto_sm3): New type attribute.
4792 (crypto_sm4): Ditto.
4793 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4794 (vsm3tt1aq_u32): Ditto.
4795 (vsm3tt1bq_u32): Ditto.
4796 (vsm3tt2aq_u32): Ditto.
4797 (vsm3tt2bq_u32): Ditto.
4798 (vsm3partw1q_u32): Ditto.
4799 (vsm3partw2q_u32): Ditto.
4800 (vsm4eq_u32): Ditto.
4801 (vsm4ekeyq_u32): Ditto.
4802 (doc/invoke.texi): Document new sm4 option.
4804 2018-01-10 Michael Collison <michael.collison@arm.com>
4806 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4807 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4808 (AARCH64_FL_FOR_ARCH8_4): New.
4809 (AARCH64_FL_V8_4): New flag.
4810 (doc/invoke.texi): Document new armv8.4-a option.
4812 2018-01-10 Michael Collison <michael.collison@arm.com>
4814 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4815 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4816 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4817 * config/aarch64/aarch64-option-extension.def: Add
4818 AARCH64_OPT_EXTENSION of 'sha2'.
4819 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4820 (crypto): Disable sha2 and aes if crypto disabled.
4821 (crypto): Enable aes and sha2 if enabled.
4822 (simd): Disable sha2 and aes if simd disabled.
4823 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4824 New flags.
4825 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4826 (TARGET_SHA2): New feature flag for sha2.
4827 (TARGET_AES): New feature flag for aes.
4828 * config/aarch64/aarch64-simd.md:
4829 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4830 conditional on TARGET_AES.
4831 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4832 (aarch64_crypto_sha1hsi): Make pattern conditional
4833 on TARGET_SHA2.
4834 (aarch64_crypto_sha1hv4si): Ditto.
4835 (aarch64_be_crypto_sha1hv4si): Ditto.
4836 (aarch64_crypto_sha1su1v4si): Ditto.
4837 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4838 (aarch64_crypto_sha1su0v4si): Ditto.
4839 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4840 (aarch64_crypto_sha256su0v4si): Ditto.
4841 (aarch64_crypto_sha256su1v4si): Ditto.
4842 (doc/invoke.texi): Document new aes and sha2 options.
4844 2018-01-10 Martin Sebor <msebor@redhat.com>
4846 PR tree-optimization/83781
4847 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4848 as string arrays.
4850 2018-01-11 Martin Sebor <msebor@gmail.com>
4851 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4853 PR tree-optimization/83501
4854 PR tree-optimization/81703
4856 * tree-ssa-strlen.c (get_string_cst): Rename...
4857 (get_string_len): ...to this. Handle global constants.
4858 (handle_char_store): Adjust.
4860 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4861 Jim Wilson <jimw@sifive.com>
4863 * config/riscv/riscv-protos.h (riscv_output_return): New.
4864 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4865 (riscv_attribute_table, riscv_output_return),
4866 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4867 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4868 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4869 function.
4870 (riscv_expand_prologue): Add early return for naked function.
4871 (riscv_expand_epilogue): Likewise.
4872 (riscv_function_ok_for_sibcall): Return false for naked function.
4873 (riscv_set_current_function): New.
4874 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4875 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4876 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4877 * doc/extend.texi (RISC-V Function Attributes): New.
4879 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4881 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4882 check for 128-bit long double before checking TCmode.
4883 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4884 128-bit long doubles before checking TFmode or TCmode.
4885 (FLOAT128_IBM_P): Likewise.
4887 2018-01-10 Martin Sebor <msebor@redhat.com>
4889 PR tree-optimization/83671
4890 * builtins.c (c_strlen): Unconditionally return zero for the empty
4891 string.
4892 Use -Warray-bounds for warnings.
4893 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4894 for non-constant array indices with COMPONENT_REF, arrays of
4895 arrays, and pointers to arrays.
4896 (gimple_fold_builtin_strlen): Determine and set length range for
4897 non-constant character arrays.
4899 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4901 PR middle-end/81897
4902 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4903 empty blocks.
4905 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4907 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4909 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4911 PR target/83399
4912 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4913 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4914 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4915 indexed_or_indirect_operand predicate.
4916 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4917 (*vsx_le_perm_load_v8hi): Likewise.
4918 (*vsx_le_perm_load_v16qi): Likewise.
4919 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4920 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4921 (*vsx_le_perm_store_v8hi): Likewise.
4922 (*vsx_le_perm_store_v16qi): Likewise.
4923 (eight unnamed splitters): Likewise.
4925 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4927 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4928 * config/rs6000/emmintrin.h: Likewise.
4929 * config/rs6000/mmintrin.h: Likewise.
4930 * config/rs6000/xmmintrin.h: Likewise.
4932 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4934 PR c++/43486
4935 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4936 "public_flag".
4937 * tree.c (tree_nop_conversion): Return true for location wrapper
4938 nodes.
4939 (maybe_wrap_with_location): New function.
4940 (selftest::check_strip_nops): New function.
4941 (selftest::test_location_wrappers): New function.
4942 (selftest::tree_c_tests): Call it.
4943 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4944 (maybe_wrap_with_location): New decl.
4945 (EXPR_LOCATION_WRAPPER_P): New macro.
4946 (location_wrapper_p): New inline function.
4947 (tree_strip_any_location_wrapper): New inline function.
4949 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4951 PR target/83735
4952 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4953 stack_realign_offset for the largest alignment of stack slot
4954 actually used.
4955 (ix86_find_max_used_stack_alignment): New function.
4956 (ix86_finalize_stack_frame_flags): Use it. Set
4957 max_used_stack_alignment if we don't realign stack.
4958 * config/i386/i386.h (machine_function): Add
4959 max_used_stack_alignment.
4961 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4963 * config/arm/arm.opt (-mbranch-cost): New option.
4964 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4965 account.
4967 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4969 PR target/83629
4970 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4971 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4973 2018-01-10 Richard Biener <rguenther@suse.de>
4975 PR debug/83765
4976 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4977 early out so it also covers the case where we have a non-NULL
4978 origin.
4980 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4982 PR tree-optimization/83753
4983 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4984 for non-strided grouped accesses if the number of elements is 1.
4986 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4988 PR target/81616
4989 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4990 * i386.h (TARGET_USE_GATHER): Define.
4991 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4993 2018-01-10 Martin Liska <mliska@suse.cz>
4995 PR bootstrap/82831
4996 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4997 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4998 partitioning.
4999 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5000 CLEANUP_NO_PARTITIONING is not set.
5002 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5004 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5005 for vectors, as a partial revert of r254296.
5006 * rtl.h (const_vec_p): Delete.
5007 (const_vec_duplicate_p): Don't test for vector CONSTs.
5008 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5009 * expmed.c (make_tree): Likewise.
5011 Revert:
5012 * common.md (E, F): Use CONSTANT_P instead of checking for
5013 CONST_VECTOR.
5014 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5015 checking for CONST_VECTOR.
5017 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5019 PR middle-end/83575
5020 * predict.c (force_edge_cold): Handle in more sane way edges
5021 with no prediction.
5023 2018-01-09 Carl Love <cel@us.ibm.com>
5025 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5026 V4SI, V4SF types.
5027 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5028 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5029 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5030 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5031 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5032 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5033 * config/rs6000/rs6000-protos.h: Add extern defition for
5034 rs6000_generate_float2_double_code.
5035 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5036 function.
5037 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5038 (float2_v2df): Add define_expand.
5040 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5042 PR target/83628
5043 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5044 op_mode in the force_to_mode call.
5046 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5048 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5049 instead of checking each element individually.
5050 (aarch64_evpc_uzp): Likewise.
5051 (aarch64_evpc_zip): Likewise.
5052 (aarch64_evpc_ext): Likewise.
5053 (aarch64_evpc_rev): Likewise.
5054 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5055 instead of checking each element individually. Return true without
5056 generating rtl if
5057 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5058 whether all selected elements come from the same input, instead of
5059 checking each element individually. Remove calls to gen_rtx_REG,
5060 start_sequence and end_sequence and instead assert that no rtl is
5061 generated.
5063 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5065 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5066 order of HIGH and CONST checks.
5068 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5070 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5071 if the destination isn't an SSA_NAME.
5073 2018-01-09 Richard Biener <rguenther@suse.de>
5075 PR tree-optimization/83668
5076 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5077 move prologue...
5078 (canonicalize_loop_form): ... here, renamed from ...
5079 (canonicalize_loop_closed_ssa_form): ... this and amended to
5080 swap successor edges for loop exit blocks to make us use
5081 the RPO order we need for initial schedule generation.
5083 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5085 PR tree-optimization/64811
5086 * match.pd: When optimizing comparisons with Inf, avoid
5087 introducing or losing exceptions from comparisons with NaN.
5089 2018-01-09 Martin Liska <mliska@suse.cz>
5091 PR sanitizer/82517
5092 * asan.c (shadow_mem_size): Add gcc_assert.
5094 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5096 Don't save registers in main().
5098 PR target/83738
5099 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5100 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5101 * config/avr/avr.c (avr_set_current_function): Don't error if
5102 naked, OS_task or OS_main are specified at the same time.
5103 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5104 OS_main.
5105 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5106 attribute.
5107 * common/config/avr/avr-common.c (avr_option_optimization_table):
5108 Switch on -mmain-is-OS_task for optimizing compilations.
5110 2018-01-09 Richard Biener <rguenther@suse.de>
5112 PR tree-optimization/83572
5113 * graphite.c: Include cfganal.h.
5114 (graphite_transform_loops): Connect infinite loops to exit
5115 and remove fake edges at the end.
5117 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5119 * ipa-inline.c (edge_badness): Revert accidental checkin.
5121 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5123 PR ipa/80763
5124 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5125 symbols; not inline clones.
5127 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5129 PR target/83507
5130 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5131 hard registers. Formatting fixes.
5133 PR preprocessor/83722
5134 * gcc.c (try_generate_repro): Pass
5135 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5136 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5137 do_report_bug.
5139 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5140 Kito Cheng <kito.cheng@gmail.com>
5142 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5143 (riscv_leaf_function_p): Delete.
5144 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5146 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5148 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5149 function.
5150 (do_ifelse): New function.
5151 (do_isel): New function.
5152 (do_sub3): New function.
5153 (do_add3): New function.
5154 (do_load_mask_compare): New function.
5155 (do_overlap_load_compare): New function.
5156 (expand_compare_loop): New function.
5157 (expand_block_compare): Call expand_compare_loop() when appropriate.
5158 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5159 option description.
5160 (-mblock-compare-inline-loop-limit): New option.
5162 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5164 PR target/83677
5165 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5166 Reverse order of second and third operands in first alternative.
5167 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5168 of first and second elements in UNSPEC_VPERMR vector.
5169 (altivec_expand_vec_perm_le): Likewise.
5171 2018-01-08 Jeff Law <law@redhat.com>
5173 PR rtl-optimizatin/81308
5174 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5175 (process_switch): If group_case_labels makes a change, then set
5176 cfg_altered.
5177 (pass_convert_switch::execute): If a switch is converted, then
5178 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5180 PR rtl-optimization/81308
5181 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5182 splitting insns.
5184 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5186 PR target/83663 - Revert r255946
5187 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5188 generation for cases where splatting a value is not useful.
5189 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5190 across a vec_duplicate and a paradoxical subreg forming a vector
5191 mode to a vec_concat.
5193 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5195 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5196 -march=armv8.3-a variants.
5197 * config/arm/t-multilib: Likewise.
5198 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5200 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5202 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5203 to generate rtl.
5204 (cceq_ior_compare_complement): Give it a name so I can use it, and
5205 change boolean_or_operator predicate to boolean_operator so it can
5206 be used to generate a crand.
5207 (eqne): New code iterator.
5208 (bd/bd_neg): New code_attrs.
5209 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5210 a single define_insn.
5211 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5212 decrement (bdnzt/bdnzf/bdzt/bdzf).
5213 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5214 with the new names of the branch decrement patterns, and added the
5215 names of the branch decrement conditional patterns.
5217 2018-01-08 Richard Biener <rguenther@suse.de>
5219 PR tree-optimization/83563
5220 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5221 cache.
5223 2018-01-08 Richard Biener <rguenther@suse.de>
5225 PR middle-end/83713
5226 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5228 2018-01-08 Richard Biener <rguenther@suse.de>
5230 PR tree-optimization/83685
5231 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5232 references to abnormals.
5234 2018-01-08 Richard Biener <rguenther@suse.de>
5236 PR lto/83719
5237 * dwarf2out.c (output_indirect_strings): Handle empty
5238 skeleton_debug_str_hash.
5239 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5241 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5243 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5244 (emit_store_direct): Likewise.
5245 (arc_trampoline_adjust_address): Likewise.
5246 (arc_asm_trampoline_template): New function.
5247 (arc_initialize_trampoline): Use asm_trampoline_template.
5248 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5249 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5250 * config/arc/arc.md (flush_icache): Delete pattern.
5252 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5254 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5255 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5256 munaligned-access.
5258 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5260 PR target/83681
5261 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5262 by not USED_FOR_TARGET.
5263 (make_pass_resolve_sw_modes): Likewise.
5265 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5267 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5268 USED_FOR_TARGET.
5270 2018-01-08 Richard Biener <rguenther@suse.de>
5272 PR middle-end/83580
5273 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5275 2018-01-08 Richard Biener <rguenther@suse.de>
5277 PR middle-end/83517
5278 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5280 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5282 PR middle-end/81897
5283 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5284 basic blocks with a small number of successors.
5285 (convert_control_dep_chain_into_preds): Improve handling of
5286 forwarder blocks.
5287 (dump_predicates): Split apart into...
5288 (dump_pred_chain): ...here...
5289 (dump_pred_info): ...and here.
5290 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5291 (can_chain_union_be_invalidated_p): Improve check for invalidation
5292 of paths.
5293 (uninit_uses_cannot_happen): Avoid unnecessary if
5294 convert_control_dep_chain_into_preds yielded nothing.
5296 2018-01-06 Martin Sebor <msebor@redhat.com>
5298 PR tree-optimization/83640
5299 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5300 subtracting negative offset from size.
5301 (builtin_access::overlap): Adjust offset bounds of the access to fall
5302 within the size of the object if possible.
5304 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5306 PR rtl-optimization/83699
5307 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5308 extract_bit_field_as_subreg to cases in which the extracted
5309 value is also a vector.
5311 * lra-constraints.c (process_alt_operands): Test for the equivalence
5312 substitutions when detecting a possible reload cycle.
5314 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5316 PR debug/83480
5317 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5318 by default if flag_selective_schedling{,2}. Formatting fixes.
5320 PR rtl-optimization/83682
5321 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5322 if it has non-VECTOR_MODE element mode.
5323 (vec_duplicate_p): Likewise.
5325 PR middle-end/83694
5326 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5327 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5329 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5331 PR target/83604
5332 * config/i386/i386-builtin.def
5333 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5334 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5335 Require also OPTION_MASK_ISA_AVX512F in addition to
5336 OPTION_MASK_ISA_GFNI.
5337 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5338 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5339 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5340 to OPTION_MASK_ISA_GFNI.
5341 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5342 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5343 OPTION_MASK_ISA_AVX512BW.
5344 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5345 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5346 addition to OPTION_MASK_ISA_GFNI.
5347 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5348 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5349 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5350 to OPTION_MASK_ISA_GFNI.
5351 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5352 a requirement for all ISAs rather than any of them with a few
5353 exceptions.
5354 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5355 processing.
5356 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5357 bitmasks to be enabled with 3 exceptions, instead of requiring any
5358 enabled ISA with lots of exceptions.
5359 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5360 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5361 Change avx512bw in isa attribute to avx512f.
5362 * config/i386/sgxintrin.h: Add license boilerplate.
5363 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5364 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5365 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5366 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5367 defined.
5368 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5369 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5370 temporarily sse2 rather than sse if not enabled already.
5372 PR target/83604
5373 * config/i386/sse.md (VI248_VLBW): Rename to ...
5374 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5375 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5376 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5377 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5378 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5379 mode iterator instead of VI248_VLBW.
5381 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5383 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5384 (record_modified): Skip clobbers; add debug output.
5385 (param_change_prob): Use sreal frequencies.
5387 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5389 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5390 punt for user-aligned variables.
5392 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5394 * tree-chrec.c (chrec_contains_symbols): Return true for
5395 POLY_INT_CST.
5397 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5399 PR target/82439
5400 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5401 of (x|y) == x for BICS pattern.
5403 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5405 PR tree-optimization/83605
5406 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5407 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5408 can throw.
5410 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5412 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5413 * config/epiphany/rtems.h: New file.
5415 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5416 Uros Bizjak <ubizjak@gmail.com>
5418 PR target/83554
5419 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5420 QIreg_operand instead of register_operand predicate.
5421 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5422 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5423 comments instead of -fmitigate[-_]rop.
5425 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5427 PR bootstrap/81926
5428 * cgraphunit.c (symbol_table::compile): Switch to text_section
5429 before calling assembly_start debug hook.
5430 * run-rtl-passes.c (run_rtl_passes): Likewise.
5431 Include output.h.
5433 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5435 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5436 range_int_cst_p rather than !symbolic_range_p before calling
5437 extract_range_from_multiplicative_op_1.
5439 2018-01-04 Jeff Law <law@redhat.com>
5441 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5442 redundant test in assertion.
5444 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5446 * doc/rtl.texi: Document machine_mode wrapper classes.
5448 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5450 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5451 using tree_to_uhwi.
5453 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5455 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5456 the VEC_PERM_EXPR fold to fail.
5458 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5460 PR debug/83585
5461 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5462 to switched_sections.
5464 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5466 PR target/83680
5467 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5468 test for d.testing.
5470 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5472 PR target/83387
5473 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5474 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5476 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5478 PR debug/83666
5479 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5480 is BLKmode and bitpos not zero or mode change is needed.
5482 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5484 PR target/83675
5485 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5486 TARGET_VIS2.
5488 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5490 PR target/83628
5491 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5492 instead of MULT rtx. Update all corresponding splitters.
5493 (*saddl_se): Ditto.
5494 (*ssub<modesuffix>): Ditto.
5495 (*ssubl_se): Ditto.
5496 (*cmp_sadd_di): Update split patterns.
5497 (*cmp_sadd_si): Ditto.
5498 (*cmp_sadd_sidi): Ditto.
5499 (*cmp_ssub_di): Ditto.
5500 (*cmp_ssub_si): Ditto.
5501 (*cmp_ssub_sidi): Ditto.
5502 * config/alpha/predicates.md (const23_operand): New predicate.
5503 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5504 Look for ASHIFT, not MULT inner operand.
5505 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5507 2018-01-04 Martin Liska <mliska@suse.cz>
5509 PR gcov-profile/83669
5510 * gcov.c (output_intermediate_file): Add version to intermediate
5511 gcov file.
5512 * doc/gcov.texi: Document new field 'version' in intermediate
5513 file format. Fix location of '-k' option of gcov command.
5515 2018-01-04 Martin Liska <mliska@suse.cz>
5517 PR ipa/82352
5518 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5520 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5522 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5524 2018-01-03 Martin Sebor <msebor@redhat.com>
5526 PR tree-optimization/83655
5527 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5528 checking calls with invalid arguments.
5530 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5532 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5533 (vectorizable_mask_load_store): Delete.
5534 (vectorizable_call): Return false for masked loads and stores.
5535 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5536 instead of gimple_assign_rhs1.
5537 (vectorizable_load): Handle IFN_MASK_LOAD.
5538 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5540 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5542 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5543 split out from..,
5544 (vectorizable_mask_load_store): ...here.
5545 (vectorizable_load): ...and here.
5547 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5549 * tree-vect-stmts.c (vect_build_all_ones_mask)
5550 (vect_build_zero_merge_argument): New functions, split out from...
5551 (vectorizable_load): ...here.
5553 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5555 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5556 split out from...
5557 (vectorizable_mask_load_store): ...here.
5558 (vectorizable_store): ...and here.
5560 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5562 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5563 split out from...
5564 (vectorizable_mask_load_store): ...here.
5566 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5568 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5569 (vect_model_store_cost): Take a vec_load_store_type instead of a
5570 vect_def_type.
5571 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5572 (vect_model_store_cost): Take a vec_load_store_type instead of a
5573 vect_def_type.
5574 (vectorizable_mask_load_store): Update accordingly.
5575 (vectorizable_store): Likewise.
5576 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5578 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5580 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5581 IFN_MASK_LOAD calls here rather than...
5582 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5584 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5585 Alan Hayward <alan.hayward@arm.com>
5586 David Sherwood <david.sherwood@arm.com>
5588 * expmed.c (extract_bit_field_1): For vector extracts,
5589 fall back to extract_bit_field_as_subreg if vec_extract
5590 isn't available.
5592 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5593 Alan Hayward <alan.hayward@arm.com>
5594 David Sherwood <david.sherwood@arm.com>
5596 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5597 they are variable or constant sized.
5598 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5599 slots for constant-sized data.
5601 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5602 Alan Hayward <alan.hayward@arm.com>
5603 David Sherwood <david.sherwood@arm.com>
5605 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5606 handling COND_EXPRs with boolean comparisons, try to find a better
5607 basis for the mask type than the boolean itself.
5609 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5611 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5612 is calculated and how it can be overridden.
5613 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5614 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5615 if defined.
5616 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5617 if nonzero.
5619 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5620 Alan Hayward <alan.hayward@arm.com>
5621 David Sherwood <david.sherwood@arm.com>
5623 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5624 Remove the mode argument.
5625 (aarch64_simd_valid_immediate): Remove the mode and inverse
5626 arguments.
5627 * config/aarch64/iterators.md (bitsize): New iterator.
5628 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5629 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5630 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5631 aarch64_simd_valid_immediate.
5632 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5633 (aarch64_reg_or_bic_imm): Likewise.
5634 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5635 with an insn_type enum and msl with a modifier_type enum.
5636 Replace element_width with a scalar_mode. Change the shift
5637 to unsigned int. Add constructors for scalar_float_mode and
5638 scalar_int_mode elements.
5639 (aarch64_vect_float_const_representable_p): Delete.
5640 (aarch64_can_const_movi_rtx_p)
5641 (aarch64_simd_scalar_immediate_valid_for_move)
5642 (aarch64_simd_make_constant): Update call to
5643 aarch64_simd_valid_immediate.
5644 (aarch64_advsimd_valid_immediate_hs): New function.
5645 (aarch64_advsimd_valid_immediate): Likewise.
5646 (aarch64_simd_valid_immediate): Remove mode and inverse
5647 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5648 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5649 and aarch64_float_const_representable_p on the result.
5650 (aarch64_output_simd_mov_immediate): Remove mode argument.
5651 Update call to aarch64_simd_valid_immediate and use of
5652 simd_immediate_info.
5653 (aarch64_output_scalar_simd_mov_immediate): Update call
5654 accordingly.
5656 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5657 Alan Hayward <alan.hayward@arm.com>
5658 David Sherwood <david.sherwood@arm.com>
5660 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5661 (mode_nunits): Likewise CONST_MODE_NUNITS.
5662 * machmode.def (ADJUST_NUNITS): Document.
5663 * genmodes.c (mode_data::need_nunits_adj): New field.
5664 (blank_mode): Update accordingly.
5665 (adj_nunits): New variable.
5666 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5667 parameter.
5668 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5669 listed in adj_nunits.
5670 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5671 listed in adj_nunits. Don't emit case statements for such modes.
5672 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5673 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5674 nothing if adj_nunits is nonnull.
5675 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5676 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5677 (emit_mode_fbit): Update use of print_maybe_const_decl.
5678 (emit_move_size): Likewise. Treat the array as non-const
5679 if adj_nunits.
5680 (emit_mode_adjustments): Handle adj_nunits.
5682 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5684 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5685 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5686 (VECTOR_MODES): Use it.
5687 (make_vector_modes): Take the prefix as an argument.
5689 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5690 Alan Hayward <alan.hayward@arm.com>
5691 David Sherwood <david.sherwood@arm.com>
5693 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5694 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5695 for MODE_VECTOR_BOOL.
5696 * machmode.def (VECTOR_BOOL_MODE): Document.
5697 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5698 (make_vector_bool_mode): New function.
5699 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5700 MODE_VECTOR_BOOL.
5701 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5702 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5703 Likewise.
5704 * stor-layout.c (int_mode_for_mode): Likewise.
5705 * tree.c (build_vector_type_for_mode): Likewise.
5706 * varasm.c (output_constant_pool_2): Likewise.
5707 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5708 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5709 for MODE_VECTOR_BOOL.
5710 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5711 of mode class checks.
5712 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5713 instead of a list of mode class checks.
5714 (expand_vector_scalar_condition): Likewise.
5715 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5717 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5718 Alan Hayward <alan.hayward@arm.com>
5719 David Sherwood <david.sherwood@arm.com>
5721 * machmode.h (mode_size): Change from unsigned short to
5722 poly_uint16_pod.
5723 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5724 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5725 or if measurement_type is not polynomial.
5726 (fixed_size_mode::includes_p): Check for constant-sized modes.
5727 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5728 return a poly_uint16 rather than an unsigned short.
5729 (emit_mode_size): Change the type of mode_size from unsigned short
5730 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5731 (emit_mode_adjustments): Cope with polynomial vector sizes.
5732 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5733 for GET_MODE_SIZE.
5734 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5735 for GET_MODE_SIZE.
5736 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5737 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5738 * caller-save.c (setup_save_areas): Likewise.
5739 (replace_reg_with_saved_mem): Likewise.
5740 * calls.c (emit_library_call_value_1): Likewise.
5741 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5742 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5743 (gen_lowpart_for_combine): Likewise.
5744 * convert.c (convert_to_integer_1): Likewise.
5745 * cse.c (equiv_constant, cse_insn): Likewise.
5746 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5747 (cselib_subst_to_values): Likewise.
5748 * dce.c (word_dce_process_block): Likewise.
5749 * df-problems.c (df_word_lr_mark_ref): Likewise.
5750 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5751 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5752 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5753 (rtl_for_decl_location): Likewise.
5754 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5755 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5756 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5757 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5758 (expand_expr_real_1): Likewise.
5759 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5760 (pad_below): Likewise.
5761 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5762 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5763 * ira.c (get_subreg_tracking_sizes): Likewise.
5764 * ira-build.c (ira_create_allocno_objects): Likewise.
5765 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5766 (ira_sort_regnos_for_alter_reg): Likewise.
5767 * ira-costs.c (record_operand_costs): Likewise.
5768 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5769 (resolve_simple_move): Likewise.
5770 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5771 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5772 (lra_constraints): Likewise.
5773 (CONST_POOL_OK_P): Reject variable-sized modes.
5774 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5775 (add_pseudo_to_slot, lra_spill): Likewise.
5776 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5777 * optabs-query.c (get_best_extraction_insn): Likewise.
5778 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5779 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5780 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5781 * recog.c (offsettable_address_addr_space_p): Likewise.
5782 * regcprop.c (maybe_mode_change): Likewise.
5783 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5784 * regrename.c (build_def_use): Likewise.
5785 * regstat.c (dump_reg_info): Likewise.
5786 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5787 (find_reloads, find_reloads_subreg_address): Likewise.
5788 * reload1.c (eliminate_regs_1): Likewise.
5789 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5790 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5791 (simplify_binary_operation_1, simplify_subreg): Likewise.
5792 * targhooks.c (default_function_arg_padding): Likewise.
5793 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5794 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5795 (verify_gimple_assign_ternary): Likewise.
5796 * tree-inline.c (estimate_move_cost): Likewise.
5797 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5798 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5799 (get_address_cost_ainc): Likewise.
5800 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5801 (vect_supportable_dr_alignment): Likewise.
5802 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5803 (vectorizable_reduction): Likewise.
5804 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5805 (vectorizable_operation, vectorizable_load): Likewise.
5806 * tree.c (build_same_sized_truth_vector_type): Likewise.
5807 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5808 * var-tracking.c (emit_note_insn_var_location): Likewise.
5809 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5810 (ADDR_VEC_ALIGN): Likewise.
5812 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5813 Alan Hayward <alan.hayward@arm.com>
5814 David Sherwood <david.sherwood@arm.com>
5816 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5817 unsigned short.
5818 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5819 or if measurement_type is polynomial.
5820 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5821 * combine.c (make_extraction): Likewise.
5822 * dse.c (find_shift_sequence): Likewise.
5823 * dwarf2out.c (mem_loc_descriptor): Likewise.
5824 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5825 (extract_bit_field, extract_low_bits): Likewise.
5826 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5827 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5828 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5829 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5830 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5831 * reload.c (find_reloads): Likewise.
5832 * reload1.c (alter_reg): Likewise.
5833 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5834 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5835 * tree-if-conv.c (predicate_mem_writes): Likewise.
5836 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5837 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5838 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5839 * valtrack.c (dead_debug_insert_temp): Likewise.
5840 * varasm.c (mergeable_constant_section): Likewise.
5841 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5843 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5844 Alan Hayward <alan.hayward@arm.com>
5845 David Sherwood <david.sherwood@arm.com>
5847 * expr.c (expand_assignment): Cope with polynomial mode sizes
5848 when assigning to a CONCAT.
5850 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5851 Alan Hayward <alan.hayward@arm.com>
5852 David Sherwood <david.sherwood@arm.com>
5854 * machmode.h (mode_precision): Change from unsigned short to
5855 poly_uint16_pod.
5856 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5857 short.
5858 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5859 or if measurement_type is not polynomial.
5860 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5861 in which the mode is already known to be a scalar_int_mode.
5862 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5863 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5864 initializer.
5865 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5866 for GET_MODE_PRECISION.
5867 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5868 for GET_MODE_PRECISION.
5869 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5870 as polynomial.
5871 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5872 (expand_field_assignment, make_extraction): Likewise.
5873 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5874 (get_last_value): Likewise.
5875 * convert.c (convert_to_integer_1): Likewise.
5876 * cse.c (cse_insn): Likewise.
5877 * expr.c (expand_expr_real_1): Likewise.
5878 * lra-constraints.c (simplify_operand_subreg): Likewise.
5879 * optabs-query.c (can_atomic_load_p): Likewise.
5880 * optabs.c (expand_atomic_load): Likewise.
5881 (expand_atomic_store): Likewise.
5882 * ree.c (combine_reaching_defs): Likewise.
5883 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5884 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5885 * tree.h (type_has_mode_precision_p): Likewise.
5886 * ubsan.c (instrument_si_overflow): Likewise.
5888 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5889 Alan Hayward <alan.hayward@arm.com>
5890 David Sherwood <david.sherwood@arm.com>
5892 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5893 polynomial numbers of units.
5894 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5895 (valid_vector_subparts_p): New function.
5896 (build_vector_type): Remove temporary shim and take the number
5897 of units as a poly_uint64 rather than an int.
5898 (build_opaque_vector_type): Take the number of units as a
5899 poly_uint64 rather than an int.
5900 * tree.c (build_vector_from_ctor): Handle polynomial
5901 TYPE_VECTOR_SUBPARTS.
5902 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5903 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5904 (build_vector_from_val): If the number of units is variable,
5905 use build_vec_duplicate_cst for constant operands and
5906 VEC_DUPLICATE_EXPR otherwise.
5907 (make_vector_type): Remove temporary is_constant ().
5908 (build_vector_type, build_opaque_vector_type): Take the number of
5909 units as a poly_uint64 rather than an int.
5910 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5911 VECTOR_CST_NELTS.
5912 * cfgexpand.c (expand_debug_expr): Likewise.
5913 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5914 (store_constructor, expand_expr_real_1): Likewise.
5915 (const_scalar_mask_from_tree): Likewise.
5916 * fold-const-call.c (fold_const_reduction): Likewise.
5917 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5918 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5919 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5920 (fold_relational_const): Likewise.
5921 (native_interpret_vector): Likewise. Change the size from an
5922 int to an unsigned int.
5923 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5924 TYPE_VECTOR_SUBPARTS.
5925 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5926 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5927 duplicating a non-constant operand into a variable-length vector.
5928 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5929 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5930 * ipa-icf.c (sem_variable::equals): Likewise.
5931 * match.pd: Likewise.
5932 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5933 * print-tree.c (print_node): Likewise.
5934 * stor-layout.c (layout_type): Likewise.
5935 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5936 * tree-cfg.c (verify_gimple_comparison): Likewise.
5937 (verify_gimple_assign_binary): Likewise.
5938 (verify_gimple_assign_ternary): Likewise.
5939 (verify_gimple_assign_single): Likewise.
5940 * tree-pretty-print.c (dump_generic_node): Likewise.
5941 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5942 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5943 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5944 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5945 (vect_shift_permute_load_chain): Likewise.
5946 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5947 (expand_vector_condition, optimize_vector_constructor): Likewise.
5948 (lower_vec_perm, get_compute_type): Likewise.
5949 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5950 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5951 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5952 (vect_recog_mask_conversion_pattern): Likewise.
5953 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5954 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5955 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5956 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5957 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5958 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5959 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5960 (supportable_widening_operation): Likewise.
5961 (supportable_narrowing_operation): Likewise.
5962 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5963 Likewise.
5964 * varasm.c (output_constant): Likewise.
5966 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5967 Alan Hayward <alan.hayward@arm.com>
5968 David Sherwood <david.sherwood@arm.com>
5970 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5971 so that both the length == 3 and length != 3 cases set up their
5972 own permute vectors. Add comments explaining why we know the
5973 number of elements is constant.
5974 (vect_permute_load_chain): Likewise.
5976 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5977 Alan Hayward <alan.hayward@arm.com>
5978 David Sherwood <david.sherwood@arm.com>
5980 * machmode.h (mode_nunits): Change from unsigned char to
5981 poly_uint16_pod.
5982 (ONLY_FIXED_SIZE_MODES): New macro.
5983 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5984 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5985 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5986 New typedefs.
5987 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5988 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5989 or if measurement_type is not polynomial.
5990 * genmodes.c (ZERO_COEFFS): New macro.
5991 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5992 poly_uint16.
5993 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5994 Use ZERO_COEFFS when emitting initializers.
5995 * data-streamer.h (bp_pack_poly_value): New function.
5996 (bp_unpack_poly_value): Likewise.
5997 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5998 for GET_MODE_NUNITS.
5999 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6000 for GET_MODE_NUNITS.
6001 * tree.c (make_vector_type): Remove temporary shim and make
6002 the real function take the number of units as a poly_uint64
6003 rather than an int.
6004 (build_vector_type_for_mode): Handle polynomial nunits.
6005 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6006 * emit-rtl.c (const_vec_series_p_1): Likewise.
6007 (gen_rtx_CONST_VECTOR): Likewise.
6008 * fold-const.c (test_vec_duplicate_folding): Likewise.
6009 * genrecog.c (validate_pattern): Likewise.
6010 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6011 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6012 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6013 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6014 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6015 * rtlanal.c (subreg_get_info): Likewise.
6016 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6017 (vect_grouped_load_supported): Likewise.
6018 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6019 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6020 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6021 (simplify_const_unary_operation, simplify_binary_operation_1)
6022 (simplify_const_binary_operation, simplify_ternary_operation)
6023 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6024 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6025 instead of CONST_VECTOR_NUNITS.
6026 * varasm.c (output_constant_pool_2): Likewise.
6027 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6028 explicit-encoded elements in the XVEC for variable-length vectors.
6030 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6032 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6034 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6035 Alan Hayward <alan.hayward@arm.com>
6036 David Sherwood <david.sherwood@arm.com>
6038 * coretypes.h (fixed_size_mode): Declare.
6039 (fixed_size_mode_pod): New typedef.
6040 * builtins.h (target_builtins::x_apply_args_mode)
6041 (target_builtins::x_apply_result_mode): Change type to
6042 fixed_size_mode_pod.
6043 * builtins.c (apply_args_size, apply_result_size, result_vector)
6044 (expand_builtin_apply_args_1, expand_builtin_apply)
6045 (expand_builtin_return): Update accordingly.
6047 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6049 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6050 * cselib.c (cselib_hash_rtx): Likewise.
6051 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6052 CONST_VECTOR encoding.
6054 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6055 Jeff Law <law@redhat.com>
6057 PR target/83641
6058 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6059 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6060 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6061 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6063 PR target/83641
6064 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6065 explicitly probe *sp in a noreturn function if there were any callee
6066 register saves or frame pointer is needed.
6068 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6070 PR debug/83621
6071 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6072 BLKmode for ternary, binary or unary expressions.
6074 PR debug/83645
6075 * var-tracking.c (delete_vta_debug_insn): New inline function.
6076 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6077 insns from get_insns () to NULL instead of each bb separately.
6078 Use delete_vta_debug_insn. No longer static.
6079 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6080 delete_vta_debug_insns callers.
6081 * rtl.h (delete_vta_debug_insns): Declare.
6082 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6083 instead of variable_tracking_main.
6085 2018-01-03 Martin Sebor <msebor@redhat.com>
6087 PR tree-optimization/83603
6088 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6089 arguments past the endof the argument list in functions declared
6090 without a prototype.
6091 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6092 Avoid checking when arguments are null.
6094 2018-01-03 Martin Sebor <msebor@redhat.com>
6096 PR c/83559
6097 * doc/extend.texi (attribute const): Fix a typo.
6098 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6099 issuing -Wsuggest-attribute for void functions.
6101 2018-01-03 Martin Sebor <msebor@redhat.com>
6103 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6104 offset_int::from instead of wide_int::to_shwi.
6105 (maybe_diag_overlap): Remove assertion.
6106 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6107 * gimple-ssa-sprintf.c (format_directive): Same.
6108 (parse_directive): Same.
6109 (sprintf_dom_walker::compute_format_length): Same.
6110 (try_substitute_return_value): Same.
6112 2018-01-03 Jeff Law <law@redhat.com>
6114 PR middle-end/83654
6115 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6116 non-constant residual for zero at runtime and avoid probing in
6117 that case. Reorganize code for trailing problem to mirror handling
6118 of the residual.
6120 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6122 PR tree-optimization/83501
6123 * tree-ssa-strlen.c (get_string_cst): New.
6124 (handle_char_store): Call get_string_cst.
6126 2018-01-03 Martin Liska <mliska@suse.cz>
6128 PR tree-optimization/83593
6129 * tree-ssa-strlen.c: Include tree-cfg.h.
6130 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6131 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6132 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6133 to false.
6134 (strlen_dom_walker::before_dom_children): Call
6135 gimple_purge_dead_eh_edges. Dump tranformation with details
6136 dump flags.
6137 (strlen_dom_walker::before_dom_children): Update call by adding
6138 new argument cleanup_eh.
6139 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6141 2018-01-03 Martin Liska <mliska@suse.cz>
6143 PR ipa/83549
6144 * cif-code.def (VARIADIC_THUNK): New enum value.
6145 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6146 thunks.
6148 2018-01-03 Jan Beulich <jbeulich@suse.com>
6150 * sse.md (mov<mode>_internal): Tighten condition for when to use
6151 vmovdqu<ssescalarsize> for TI and OI modes.
6153 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6155 Update copyright years.
6157 2018-01-03 Martin Liska <mliska@suse.cz>
6159 PR ipa/83594
6160 * ipa-visibility.c (function_and_variable_visibility): Skip
6161 functions with noipa attribure.
6163 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6165 * gcc.c (process_command): Update copyright notice dates.
6166 * gcov-dump.c (print_version): Ditto.
6167 * gcov.c (print_version): Ditto.
6168 * gcov-tool.c (print_version): Ditto.
6169 * gengtype.c (create_file): Ditto.
6170 * doc/cpp.texi: Bump @copying's copyright year.
6171 * doc/cppinternals.texi: Ditto.
6172 * doc/gcc.texi: Ditto.
6173 * doc/gccint.texi: Ditto.
6174 * doc/gcov.texi: Ditto.
6175 * doc/install.texi: Ditto.
6176 * doc/invoke.texi: Ditto.
6178 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6180 * vector-builder.h (vector_builder::m_full_nelts): Change from
6181 unsigned int to poly_uint64.
6182 (vector_builder::full_nelts): Update prototype accordingly.
6183 (vector_builder::new_vector): Likewise.
6184 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6185 (vector_builder::operator ==): Likewise.
6186 (vector_builder::finalize): Likewise.
6187 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6188 Take the number of elements as a poly_uint64 rather than an
6189 unsigned int.
6190 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6191 from unsigned int to poly_uint64.
6192 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6193 (vec_perm_indices::new_vector): Likewise.
6194 (vec_perm_indices::length): Likewise.
6195 (vec_perm_indices::nelts_per_input): Likewise.
6196 (vec_perm_indices::input_nelts): Likewise.
6197 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6198 number of elements per input as a poly_uint64 rather than an
6199 unsigned int. Use the original encoding for variable-length
6200 vectors, rather than clamping each individual element.
6201 For the second and subsequent elements in each pattern,
6202 clamp the step and base before clamping their sum.
6203 (vec_perm_indices::series_p): Handle polynomial element counts.
6204 (vec_perm_indices::all_in_range_p): Likewise.
6205 (vec_perm_indices_to_tree): Likewise.
6206 (vec_perm_indices_to_rtx): Likewise.
6207 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6208 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6209 (tree_vector_builder::new_binary_operation): Handle polynomial
6210 element counts. Return false if we need to know the number
6211 of elements at compile time.
6212 * fold-const.c (fold_vec_perm): Punt if the number of elements
6213 isn't known at compile time.
6215 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6217 * vec-perm-indices.h (vec_perm_builder): Change element type
6218 from HOST_WIDE_INT to poly_int64.
6219 (vec_perm_indices::element_type): Update accordingly.
6220 (vec_perm_indices::clamp): Handle polynomial element_types.
6221 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6222 (vec_perm_indices::all_in_range_p): Likewise.
6223 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6224 than shwi trees.
6225 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6226 polynomial vec_perm_indices element types.
6227 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6228 * fold-const.c (fold_vec_perm): Likewise.
6229 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6230 * tree-vect-generic.c (lower_vec_perm): Likewise.
6231 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6232 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6233 element type to HOST_WIDE_INT.
6235 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6236 Alan Hayward <alan.hayward@arm.com>
6237 David Sherwood <david.sherwood@arm.com>
6239 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6240 rather than an int. Use plus_constant.
6241 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6242 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6244 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6245 Alan Hayward <alan.hayward@arm.com>
6246 David Sherwood <david.sherwood@arm.com>
6248 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6249 a HOST_WIDE_INT to a poly_int64.
6251 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6252 Alan Hayward <alan.hayward@arm.com>
6253 David Sherwood <david.sherwood@arm.com>
6255 * calls.c (load_register_parameters): Cope with polynomial
6256 mode sizes. Require a constant size for BLKmode parameters
6257 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6258 forces a parameter to be padded at the lsb end in order to
6259 fill a complete number of words, require the parameter size
6260 to be ordered wrt UNITS_PER_WORD.
6262 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6263 Alan Hayward <alan.hayward@arm.com>
6264 David Sherwood <david.sherwood@arm.com>
6266 * reload1.c (spill_stack_slot_width): Change element type
6267 from unsigned int to poly_uint64_pod.
6268 (alter_reg): Treat mode sizes as polynomial.
6270 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6271 Alan Hayward <alan.hayward@arm.com>
6272 David Sherwood <david.sherwood@arm.com>
6274 * reload.c (complex_word_subreg_p): New function.
6275 (reload_inner_reg_of_subreg, push_reload): Use it.
6277 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6278 Alan Hayward <alan.hayward@arm.com>
6279 David Sherwood <david.sherwood@arm.com>
6281 * lra-constraints.c (process_alt_operands): Reject matched
6282 operands whose sizes aren't ordered.
6283 (match_reload): Refer to this check here.
6285 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6286 Alan Hayward <alan.hayward@arm.com>
6287 David Sherwood <david.sherwood@arm.com>
6289 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6290 that the mode size is in the set {1, 2, 4, 8, 16}.
6292 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6293 Alan Hayward <alan.hayward@arm.com>
6294 David Sherwood <david.sherwood@arm.com>
6296 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6297 Use plus_constant instead of gen_rtx_PLUS.
6299 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6300 Alan Hayward <alan.hayward@arm.com>
6301 David Sherwood <david.sherwood@arm.com>
6303 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6304 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6305 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6306 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6307 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6308 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6309 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6310 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6311 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6312 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6313 a poly_int64.
6314 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6315 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6316 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6317 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6318 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6319 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6320 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6321 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6322 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6323 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6324 function.
6325 * expr.c (emit_move_resolve_push): Treat the input and result
6326 of PUSH_ROUNDING as a poly_int64.
6327 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6328 (emit_push_insn): Likewise.
6329 * lra-eliminations.c (mark_not_eliminable): Likewise.
6330 * recog.c (push_operand): Likewise.
6331 * reload1.c (elimination_effects): Likewise.
6332 * rtlanal.c (nonzero_bits1): Likewise.
6333 * calls.c (store_one_arg): Likewise. Require the padding to be
6334 known at compile time.
6336 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6337 Alan Hayward <alan.hayward@arm.com>
6338 David Sherwood <david.sherwood@arm.com>
6340 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6341 Use plus_constant instead of gen_rtx_PLUS.
6343 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6344 Alan Hayward <alan.hayward@arm.com>
6345 David Sherwood <david.sherwood@arm.com>
6347 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6348 rather than an int.
6350 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6351 Alan Hayward <alan.hayward@arm.com>
6352 David Sherwood <david.sherwood@arm.com>
6354 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6355 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6356 via stack temporaries. Treat the mode size as polynomial too.
6358 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6359 Alan Hayward <alan.hayward@arm.com>
6360 David Sherwood <david.sherwood@arm.com>
6362 * expr.c (expand_expr_real_2): When handling conversions involving
6363 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6364 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6365 as a poly_uint64 too.
6367 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6368 Alan Hayward <alan.hayward@arm.com>
6369 David Sherwood <david.sherwood@arm.com>
6371 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6373 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6374 Alan Hayward <alan.hayward@arm.com>
6375 David Sherwood <david.sherwood@arm.com>
6377 * combine.c (can_change_dest_mode): Handle polynomial
6378 REGMODE_NATURAL_SIZE.
6379 * expmed.c (store_bit_field_1): Likewise.
6380 * expr.c (store_constructor): Likewise.
6381 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6382 and polynomial REGMODE_NATURAL_SIZE.
6383 (gen_lowpart_common): Likewise.
6384 * reginfo.c (record_subregs_of_mode): Likewise.
6385 * rtlanal.c (read_modify_subreg_p): Likewise.
6387 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6388 Alan Hayward <alan.hayward@arm.com>
6389 David Sherwood <david.sherwood@arm.com>
6391 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6392 numbers of elements.
6394 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6395 Alan Hayward <alan.hayward@arm.com>
6396 David Sherwood <david.sherwood@arm.com>
6398 * match.pd: Cope with polynomial numbers of vector elements.
6400 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6401 Alan Hayward <alan.hayward@arm.com>
6402 David Sherwood <david.sherwood@arm.com>
6404 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6405 in a POINTER_PLUS_EXPR.
6407 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6408 Alan Hayward <alan.hayward@arm.com>
6409 David Sherwood <david.sherwood@arm.com>
6411 * omp-simd-clone.c (simd_clone_subparts): New function.
6412 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6413 (ipa_simd_modify_function_body): Likewise.
6415 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6416 Alan Hayward <alan.hayward@arm.com>
6417 David Sherwood <david.sherwood@arm.com>
6419 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6420 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6421 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6422 (expand_vector_condition, vector_element): Likewise.
6423 (subparts_gt): New function.
6424 (get_compute_type): Use subparts_gt.
6425 (count_type_subparts): Delete.
6426 (expand_vector_operations_1): Use subparts_gt instead of
6427 count_type_subparts.
6429 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6430 Alan Hayward <alan.hayward@arm.com>
6431 David Sherwood <david.sherwood@arm.com>
6433 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6434 (vect_compile_time_alias): ...this new function. Do the calculation
6435 on poly_ints rather than trees.
6436 (vect_prune_runtime_alias_test_list): Update call accordingly.
6438 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6439 Alan Hayward <alan.hayward@arm.com>
6440 David Sherwood <david.sherwood@arm.com>
6442 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6443 numbers of units.
6444 (vect_schedule_slp_instance): Likewise.
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 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6451 constant and extern definitions for variable-length vectors.
6452 (vect_get_constant_vectors): Note that the number of units
6453 is known to be constant.
6455 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6456 Alan Hayward <alan.hayward@arm.com>
6457 David Sherwood <david.sherwood@arm.com>
6459 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6460 of units as polynomial. Choose between WIDE and NARROW based
6461 on multiple_p.
6463 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6464 Alan Hayward <alan.hayward@arm.com>
6465 David Sherwood <david.sherwood@arm.com>
6467 * tree-vect-stmts.c (simd_clone_subparts): New function.
6468 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6470 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6471 Alan Hayward <alan.hayward@arm.com>
6472 David Sherwood <david.sherwood@arm.com>
6474 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6475 vectors as polynomial. Use build_index_vector for
6476 IFN_GOMP_SIMD_LANE.
6478 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6479 Alan Hayward <alan.hayward@arm.com>
6480 David Sherwood <david.sherwood@arm.com>
6482 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6483 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6484 for variable-length vectors.
6485 (vectorizable_mask_load_store): Treat the number of units as
6486 polynomial, asserting that it is constant if the condition has
6487 already been enforced.
6488 (vectorizable_store, vectorizable_load): Likewise.
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-loop.c (vectorizable_live_operation): Treat the number
6495 of units as polynomial. Punt if we can't tell at compile time
6496 which vector contains the final result.
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-loop.c (vectorizable_induction): Treat the number
6503 of units as polynomial. Punt on SLP inductions. Use an integer
6504 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6505 cast of such a series for variable-length floating-point
6506 reductions.
6508 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6509 Alan Hayward <alan.hayward@arm.com>
6510 David Sherwood <david.sherwood@arm.com>
6512 * tree.h (build_index_vector): Declare.
6513 * tree.c (build_index_vector): New function.
6514 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6515 of units as polynomial, forcibly converting it to a constant if
6516 vectorizable_reduction has already enforced the condition.
6517 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6518 to create a {1,2,3,...} vector.
6519 (vectorizable_reduction): Treat the number of units as polynomial.
6520 Choose vectype_in based on the largest scalar element size rather
6521 than the smallest number of units. Enforce the restrictions
6522 relied on above.
6524 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6525 Alan Hayward <alan.hayward@arm.com>
6526 David Sherwood <david.sherwood@arm.com>
6528 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6529 number of units as polynomial.
6531 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6532 Alan Hayward <alan.hayward@arm.com>
6533 David Sherwood <david.sherwood@arm.com>
6535 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6536 * target.def (autovectorize_vector_sizes): Return the vector sizes
6537 by pointer, using vector_sizes rather than a bitmask.
6538 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6539 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6540 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6541 Likewise.
6542 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6543 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6544 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6545 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6546 * omp-general.c (omp_max_vf): Likewise.
6547 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6548 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6549 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6550 * tree-vect-slp.c (vect_slp_bb): Likewise.
6551 * doc/tm.texi: Regenerate.
6552 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6553 to a poly_uint64.
6554 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6555 the vector size as a poly_uint64 rather than an unsigned int.
6556 (current_vector_size): Change from an unsigned int to a poly_uint64.
6557 (get_vectype_for_scalar_type): Update accordingly.
6558 * tree.h (build_truth_vector_type): Take the size and number of
6559 units as a poly_uint64 rather than an unsigned int.
6560 (build_vector_type): Add a temporary overload that takes
6561 the number of units as a poly_uint64 rather than an unsigned int.
6562 * tree.c (make_vector_type): Likewise.
6563 (build_truth_vector_type): Take the number of units as a poly_uint64
6564 rather than an unsigned int.
6566 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6567 Alan Hayward <alan.hayward@arm.com>
6568 David Sherwood <david.sherwood@arm.com>
6570 * target.def (get_mask_mode): Take the number of units and length
6571 as poly_uint64s rather than unsigned ints.
6572 * targhooks.h (default_get_mask_mode): Update accordingly.
6573 * targhooks.c (default_get_mask_mode): Likewise.
6574 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6575 * doc/tm.texi: Regenerate.
6577 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6578 Alan Hayward <alan.hayward@arm.com>
6579 David Sherwood <david.sherwood@arm.com>
6581 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6582 * omp-general.c (omp_max_vf): Likewise.
6583 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6584 (expand_omp_simd): Handle polynomial safelen.
6585 * omp-low.c (omplow_simd_context): Add a default constructor.
6586 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6587 (lower_rec_simd_input_clauses): Update accordingly.
6588 (lower_rec_input_clauses): Likewise.
6590 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6591 Alan Hayward <alan.hayward@arm.com>
6592 David Sherwood <david.sherwood@arm.com>
6594 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6595 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6596 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6597 (vect_analyze_slp_cost): Likewise.
6598 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6599 (vect_model_load_cost): Likewise.
6601 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6602 Alan Hayward <alan.hayward@arm.com>
6603 David Sherwood <david.sherwood@arm.com>
6605 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6606 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6607 from an unsigned int * to a poly_uint64_pod *.
6608 (calculate_unrolling_factor): New function.
6609 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6611 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6612 Alan Hayward <alan.hayward@arm.com>
6613 David Sherwood <david.sherwood@arm.com>
6615 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6616 from an unsigned int to a poly_uint64.
6617 (_loop_vec_info::slp_unrolling_factor): Likewise.
6618 (_loop_vec_info::vectorization_factor): Change from an int
6619 to a poly_uint64.
6620 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6621 (vect_get_num_vectors): New function.
6622 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6623 (vect_get_num_copies): Use vect_get_num_vectors.
6624 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6625 to an unsigned int *.
6626 (vect_analyze_data_refs): Change min_vf from an int * to a
6627 poly_uint64 *.
6628 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6629 than an unsigned HOST_WIDE_INT.
6630 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6631 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6632 to an unsigned int *.
6633 (vect_analyze_data_ref_dependences): Likewise.
6634 (vect_compute_data_ref_alignment): Handle polynomial vf.
6635 (vect_enhance_data_refs_alignment): Likewise.
6636 (vect_prune_runtime_alias_test_list): Likewise.
6637 (vect_shift_permute_load_chain): Likewise.
6638 (vect_supportable_dr_alignment): Likewise.
6639 (dependence_distance_ge_vf): Take the vectorization factor as a
6640 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6641 (vect_analyze_data_refs): Change min_vf from an int * to a
6642 poly_uint64 *.
6643 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6644 vfm1 as a poly_uint64 rather than an int. Make the same change
6645 for the returned bound_scalar.
6646 (vect_gen_vector_loop_niters): Handle polynomial vf.
6647 (vect_do_peeling): Likewise. Update call to
6648 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6649 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6650 be constant.
6651 * tree-vect-loop.c (vect_determine_vectorization_factor)
6652 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6653 (vect_get_known_peeling_cost): Likewise.
6654 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6655 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6656 (vect_transform_loop): Likewise. Use the lowest possible VF when
6657 updating the upper bounds of the loop.
6658 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6659 rather than an int.
6660 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6661 polynomial unroll factors.
6662 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6663 (vect_make_slp_decision): Likewise.
6664 (vect_supported_load_permutation_p): Likewise, and polynomial
6665 vf too.
6666 (vect_analyze_slp_cost): Handle polynomial vf.
6667 (vect_slp_analyze_node_operations): Likewise.
6668 (vect_slp_analyze_bb_1): Likewise.
6669 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6670 than an unsigned HOST_WIDE_INT.
6671 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6672 (vectorizable_load): Handle polynomial vf.
6673 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6674 a poly_uint64.
6675 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6677 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6678 Alan Hayward <alan.hayward@arm.com>
6679 David Sherwood <david.sherwood@arm.com>
6681 * match.pd: Handle bit operations involving three constants
6682 and try to fold one pair.
6684 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6686 * tree-vect-loop-manip.c: Include gimple-fold.h.
6687 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6688 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6689 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6690 Add a path that uses a step of VF instead of 1, but disable it
6691 for now.
6692 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6693 and niters_no_overflow parameters. Update calls to
6694 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6695 Create a new SSA name if the latter choses to use a ste other
6696 than zero, and return it via niters_vector_mult_vf_var.
6697 * tree-vect-loop.c (vect_transform_loop): Update calls to
6698 vect_do_peeling, vect_gen_vector_loop_niters and
6699 slpeel_make_loop_iterate_ntimes.
6700 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6701 (vect_gen_vector_loop_niters): Update declarations after above changes.
6703 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6705 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6706 128-bit round to integer instructions.
6707 (ceil<mode>2): Likewise.
6708 (btrunc<mode>2): Likewise.
6709 (round<mode>2): Likewise.
6711 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6713 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6714 unaligned VSX load/store on P8/P9.
6715 (expand_block_clear): Allow the use of unaligned VSX
6716 load/store on P8/P9.
6718 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6720 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6721 New function.
6722 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6723 swap associated with both a load and a store.
6725 2018-01-02 Andrew Waterman <andrew@sifive.com>
6727 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6728 * config/riscv/riscv.md (clear_cache): Use it.
6730 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6732 * web.c: Remove out-of-date comment.
6734 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6736 * expr.c (fixup_args_size_notes): Check that any existing
6737 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6738 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6739 (emit_single_push_insn): ...here.
6741 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6743 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6744 (const_vector_encoded_nelts): New function.
6745 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6746 (const_vector_int_elt, const_vector_elt): Declare.
6747 * emit-rtl.c (const_vector_int_elt_1): New function.
6748 (const_vector_elt): Likewise.
6749 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6750 of CONST_VECTOR_ELT.
6752 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6754 * expr.c: Include rtx-vector-builder.h.
6755 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6756 directly on the tree encoding.
6757 (const_vector_from_tree): Likewise.
6758 * optabs.c: Include rtx-vector-builder.h.
6759 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6760 sequence of "u" values.
6761 * vec-perm-indices.c: Include rtx-vector-builder.h.
6762 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6763 directly on the vec_perm_indices encoding.
6765 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6767 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6768 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6769 * rtx-vector-builder.h: New file.
6770 * rtx-vector-builder.c: Likewise.
6771 * rtl.h (rtx_def::u2): Add a const_vector field.
6772 (CONST_VECTOR_NPATTERNS): New macro.
6773 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6774 (CONST_VECTOR_DUPLICATE_P): Likewise.
6775 (CONST_VECTOR_STEPPED_P): Likewise.
6776 (CONST_VECTOR_ENCODED_ELT): Likewise.
6777 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6778 (unwrap_const_vec_duplicate): Likewise.
6779 (const_vec_series_p): Check for a non-duplicated vector encoding.
6780 Say that the function only returns true for integer vectors.
6781 * emit-rtl.c: Include rtx-vector-builder.h.
6782 (gen_const_vec_duplicate_1): Delete.
6783 (gen_const_vector): Call gen_const_vec_duplicate instead of
6784 gen_const_vec_duplicate_1.
6785 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6786 (gen_const_vec_duplicate): Use rtx_vector_builder.
6787 (gen_const_vec_series): Likewise.
6788 (gen_rtx_CONST_VECTOR): Likewise.
6789 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6790 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6791 Build a new vector rather than modifying a CONST_VECTOR in-place.
6792 (handle_special_swappables): Update call accordingly.
6793 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6794 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6795 Build a new vector rather than modifying a CONST_VECTOR in-place.
6796 (handle_special_swappables): Update call accordingly.
6798 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6800 * simplify-rtx.c (simplify_const_binary_operation): Use
6801 CONST_VECTOR_ELT instead of XVECEXP.
6803 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6805 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6806 the selector elements to be different from the data elements
6807 if the selector is a VECTOR_CST.
6808 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6809 ssizetype for the selector.
6811 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6813 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6814 before testing each element individually.
6815 * tree-vect-generic.c (lower_vec_perm): Likewise.
6817 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6819 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6820 * selftest-run-tests.c (selftest::run_tests): Call it.
6821 * vector-builder.h (vector_builder::operator ==): New function.
6822 (vector_builder::operator !=): Likewise.
6823 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6824 (vec_perm_indices::all_from_input_p): New function.
6825 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6826 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6827 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6828 instead of reading the VECTOR_CST directly. Detect whether both
6829 vector inputs are the same before constructing the vec_perm_indices,
6830 and update the number of inputs argument accordingly. Use the
6831 utility functions added above. Only construct sel2 if we need to.
6833 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6835 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6836 the broadcast of the low byte.
6837 (expand_mult_highpart): Use an explicit encoding for the permutes.
6838 * optabs-query.c (can_mult_highpart_p): Likewise.
6839 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6840 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6841 (vectorizable_bswap): Likewise.
6842 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6843 explicit encoding for the power-of-2 permutes.
6844 (vect_permute_store_chain): Likewise.
6845 (vect_grouped_load_supported): Likewise.
6846 (vect_permute_load_chain): Likewise.
6848 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6850 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6851 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6852 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6853 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6854 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6855 (vect_gen_perm_mask_any): Likewise.
6857 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6859 * int-vector-builder.h: New file.
6860 * vec-perm-indices.h: Include int-vector-builder.h.
6861 (vec_perm_indices): Redefine as an int_vector_builder.
6862 (auto_vec_perm_indices): Delete.
6863 (vec_perm_builder): Redefine as a stand-alone class.
6864 (vec_perm_indices::vec_perm_indices): New function.
6865 (vec_perm_indices::clamp): Likewise.
6866 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6867 (vec_perm_indices::new_vector): New function.
6868 (vec_perm_indices::new_expanded_vector): Update for new
6869 vec_perm_indices class.
6870 (vec_perm_indices::rotate_inputs): New function.
6871 (vec_perm_indices::all_in_range_p): Operate directly on the
6872 encoded form, without computing elided elements.
6873 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6874 encoding. Update for new vec_perm_indices class.
6875 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6876 the given vec_perm_builder.
6877 (expand_vec_perm_var): Update vec_perm_builder constructor.
6878 (expand_mult_highpart): Use vec_perm_builder instead of
6879 auto_vec_perm_indices.
6880 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6881 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6882 or double series encoding as appropriate.
6883 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6884 vec_perm_indices instead of auto_vec_perm_indices.
6885 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6886 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6887 (vect_permute_store_chain): Likewise.
6888 (vect_grouped_load_supported): Likewise.
6889 (vect_permute_load_chain): Likewise.
6890 (vect_shift_permute_load_chain): Likewise.
6891 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6892 (vect_transform_slp_perm_load): Likewise.
6893 (vect_schedule_slp_instance): Likewise.
6894 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6895 (vectorizable_mask_load_store): Likewise.
6896 (vectorizable_bswap): Likewise.
6897 (vectorizable_store): Likewise.
6898 (vectorizable_load): Likewise.
6899 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6900 vec_perm_indices instead of auto_vec_perm_indices. Use
6901 tree_to_vec_perm_builder to read the vector from a tree.
6902 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6903 vec_perm_builder instead of a vec_perm_indices.
6904 (have_whole_vector_shift): Use vec_perm_builder and
6905 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6906 truncation to calc_vec_perm_mask_for_shift.
6907 (vect_create_epilog_for_reduction): Likewise.
6908 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6909 from auto_vec_perm_indices to vec_perm_indices.
6910 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6911 instead of changing individual elements.
6912 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6913 the vector in d.perm.
6914 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6915 from auto_vec_perm_indices to vec_perm_indices.
6916 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6917 instead of changing individual elements.
6918 (arm_vectorize_vec_perm_const): Use new_vector to install
6919 the vector in d.perm.
6920 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6921 Update vec_perm_builder constructor.
6922 (rs6000_expand_interleave): Likewise.
6923 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6924 (rs6000_expand_interleave): Likewise.
6926 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6928 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6929 to qimode could truncate the indices.
6930 * optabs.c (expand_vec_perm_var): Likewise.
6932 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6934 * Makefile.in (OBJS): Add vec-perm-indices.o.
6935 * vec-perm-indices.h: New file.
6936 * vec-perm-indices.c: Likewise.
6937 * target.h (vec_perm_indices): Replace with a forward class
6938 declaration.
6939 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6940 * optabs.h: Include vec-perm-indices.h.
6941 (expand_vec_perm): Delete.
6942 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6943 (expand_vec_perm_const): Declare.
6944 * target.def (vec_perm_const_ok): Replace with...
6945 (vec_perm_const): ...this new hook.
6946 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6947 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6948 * doc/tm.texi: Regenerate.
6949 * optabs.def (vec_perm_const): Delete.
6950 * doc/md.texi (vec_perm_const): Likewise.
6951 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6952 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6953 expand_vec_perm for constant permutation vectors. Assert that
6954 the mode of variable permutation vectors is the integer equivalent
6955 of the mode that is being permuted.
6956 * optabs-query.h (selector_fits_mode_p): Declare.
6957 * optabs-query.c: Include vec-perm-indices.h.
6958 (selector_fits_mode_p): New function.
6959 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6960 is defined, instead of checking whether the vec_perm_const_optab
6961 exists. Use targetm.vectorize.vec_perm_const instead of
6962 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6963 fit in the vector mode before using a variable permute.
6964 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6965 vec_perm_indices instead of an rtx.
6966 (expand_vec_perm): Replace with...
6967 (expand_vec_perm_const): ...this new function. Take the selector
6968 as a vec_perm_indices rather than an rtx. Also take the mode of
6969 the selector. Update call to shift_amt_for_vec_perm_mask.
6970 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6971 Use vec_perm_indices::new_expanded_vector to expand the original
6972 selector into bytes. Check whether the indices fit in the vector
6973 mode before using a variable permute.
6974 (expand_vec_perm_var): Make global.
6975 (expand_mult_highpart): Use expand_vec_perm_const.
6976 * fold-const.c: Includes vec-perm-indices.h.
6977 * tree-ssa-forwprop.c: Likewise.
6978 * tree-vect-data-refs.c: Likewise.
6979 * tree-vect-generic.c: Likewise.
6980 * tree-vect-loop.c: Likewise.
6981 * tree-vect-slp.c: Likewise.
6982 * tree-vect-stmts.c: Likewise.
6983 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6984 Delete.
6985 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6986 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6987 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6988 (aarch64_vectorize_vec_perm_const): ...this new function.
6989 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6990 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6991 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6992 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6993 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6994 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6995 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6996 into...
6997 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6998 check for NEON modes.
6999 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7000 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7001 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7002 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7003 into...
7004 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7005 the old VEC_PERM_CONST conditions.
7006 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7007 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7008 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7009 (ia64_vectorize_vec_perm_const_ok): Merge into...
7010 (ia64_vectorize_vec_perm_const): ...this new function.
7011 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7012 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7013 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7014 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7015 * config/mips/mips.c (mips_expand_vec_perm_const)
7016 (mips_vectorize_vec_perm_const_ok): Merge into...
7017 (mips_vectorize_vec_perm_const): ...this new function.
7018 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7019 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7020 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7021 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7022 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7023 (rs6000_expand_vec_perm_const): Delete.
7024 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7025 Delete.
7026 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7027 (altivec_expand_vec_perm_const_le): Take each operand individually.
7028 Operate on constant selectors rather than rtxes.
7029 (altivec_expand_vec_perm_const): Likewise. Update call to
7030 altivec_expand_vec_perm_const_le.
7031 (rs6000_expand_vec_perm_const): Delete.
7032 (rs6000_vectorize_vec_perm_const_ok): Delete.
7033 (rs6000_vectorize_vec_perm_const): New function.
7034 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7035 an element count and rtx array.
7036 (rs6000_expand_extract_even): Update call accordingly.
7037 (rs6000_expand_interleave): Likewise.
7038 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7039 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7040 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7041 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7042 (rs6000_expand_vec_perm_const): Delete.
7043 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7044 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7045 (altivec_expand_vec_perm_const_le): Take each operand individually.
7046 Operate on constant selectors rather than rtxes.
7047 (altivec_expand_vec_perm_const): Likewise. Update call to
7048 altivec_expand_vec_perm_const_le.
7049 (rs6000_expand_vec_perm_const): Delete.
7050 (rs6000_vectorize_vec_perm_const_ok): Delete.
7051 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7052 reference to the SPE evmerge intructions.
7053 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7054 an element count and rtx array.
7055 (rs6000_expand_extract_even): Update call accordingly.
7056 (rs6000_expand_interleave): Likewise.
7057 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7058 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7059 new function.
7060 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7062 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7064 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7065 vector mode and that that mode matches the mode of the data
7066 being permuted.
7067 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7068 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7069 directly using expand_vec_perm_1 when forcing selectors into
7070 registers.
7071 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7073 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7075 * optabs-query.h (can_vec_perm_p): Delete.
7076 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7077 * optabs-query.c (can_vec_perm_p): Split into...
7078 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7079 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7080 particular selector is valid.
7081 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7082 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7083 (vect_grouped_load_supported): Likewise.
7084 (vect_shift_permute_load_chain): Likewise.
7085 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7086 (vect_transform_slp_perm_load): Likewise.
7087 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7088 (vectorizable_bswap): Likewise.
7089 (vect_gen_perm_mask_checked): Likewise.
7090 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7091 implementations of variable permutation vectors into account
7092 when deciding which selector to use.
7093 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7094 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7095 with a false third argument.
7096 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7097 to test whether the constant selector is valid and can_vec_perm_var_p
7098 to test whether a variable selector is valid.
7100 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7102 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7103 * optabs-query.c (can_vec_perm_p): Likewise.
7104 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7105 instead of vec_perm_indices.
7106 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7107 (vect_gen_perm_mask_checked): Likewise,
7108 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7109 (vect_gen_perm_mask_checked): Likewise,
7111 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7113 * optabs-query.h (qimode_for_vec_perm): Declare.
7114 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7115 (qimode_for_vec_perm): ...this new function.
7116 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7118 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7120 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7121 does not have a conditional at the top.
7123 2018-01-02 Richard Biener <rguenther@suse.de>
7125 * ipa-inline.c (big_speedup_p): Fix expression.
7127 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7129 PR target/81616
7130 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7131 for generic 4->6.
7133 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7135 PR target/81616
7136 Generic tuning.
7137 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7138 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7139 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7140 cond_taken_branch_cost 3->4.
7142 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7144 PR tree-optimization/83581
7145 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7146 TODO_cleanup_cfg if any changes have been made.
7148 PR middle-end/83608
7149 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7150 convert_modes if target mode has the right side, but different mode
7151 class.
7153 PR middle-end/83609
7154 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7155 last argument when extracting from CONCAT. If either from_real or
7156 from_imag is NULL, use expansion through memory. If result is not
7157 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7158 the parts directly to inner mode, if even that fails, use expansion
7159 through memory.
7161 PR middle-end/83623
7162 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7163 check for bswap in mode rather than HImode and use that in expand_unop
7164 too.
7165 \f
7166 Copyright (C) 2018 Free Software Foundation, Inc.
7168 Copying and distribution of this file, with or without modification,
7169 are permitted in any medium without royalty provided the copyright
7170 notice and this notice are preserved.