| blob | 6b09207797980b8de508f564f9124212ed317478 |
1 2018-02-14 Martin Sebor <msebor@redhat.com>
3 PR tree-optimization/83698
4 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
5 arrays constrain the offset range to their bounds.
6 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
7 (builtin_access::overlap): Avoid setting the size of overlap if it's
8 already been set.
9 (maybe_diag_overlap): Also consider arrays when deciding what values
10 of offsets to include in diagnostics.
12 2018-02-14 Martin Sebor <msebor@redhat.com>
14 PR c/84108
15 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
16 that correspond to the kind of a declaration.
18 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
20 PR target/83984
21 * config/pa/pa.md: Load address of PIC label using the linkage table
22 if the label is nonlocal.
24 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
26 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
27 warning message if user requests -maltivec=be.
28 * doc/invoke.texi: Document deprecation of -maltivec=be.
30 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
32 PR target/84220
33 * config/rs6000/rs6000-c.c: Update definitions for
34 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
35 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
37 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
39 PR target/84239
40 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
41 add _get_ssp intrinsics. Remove argument from
42 __builtin_ia32_rdssp[d|q].
43 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
44 * config/i386/i386-builtin.def: Remove argument from
45 __builtin_ia32_rdssp[d|q].
46 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
47 ix86_expand_special_args_builtin for _rdssp[d|q].
48 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
49 Clear register before usage.
50 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
51 Add documentation for new _get_ssp and _inc_ssp intrinsics.
53 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
55 PR tree-optimization/84357
56 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
57 operand 1 of an ARRAY_REF too.
59 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
61 PR target/83831
62 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
63 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
64 declarations.
65 (set_of_reg): New struct.
66 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
67 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
68 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
69 functions.
70 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
71 Split into bitclr, bitset, bitinvert patterns if appropriate.
72 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
73 use rx_fuse_in_memory_bitop.
74 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
75 to named insn, correct maximum insn length.
77 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
79 PR target/79242
80 * machmode.def: Define a complex mode for PARTIAL_INT.
81 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
82 MODE_PARTIAL_INT.
83 * doc/rtl.texi: Document CSPImode.
84 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
85 handling.
86 (msp430_hard_regno_nregs_with_padding): Likewise.
88 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
90 PR target/84279
91 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
93 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
95 PR rtl-optimization/84169
96 * combine.c (try_combine): New variable split_i2i3. Set it to true if
97 we generated a parallel as new i3 and we split that to new i2 and i3
98 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
99 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
100 those to i2, not i1. Partially rewrite this scan code.
102 2018-02-13 Jakub Jelinek <jakub@redhat.com>
104 PR c/82210
105 * stor-layout.c (place_field): For variable length fields, adjust
106 offset_align afterwards not just based on the field's alignment,
107 but also on the size.
109 PR middle-end/84309
110 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
111 of exps and logs in the use_exp2 case.
113 2018-02-13 Jeff Law <law@redhat.com>
115 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
116 entry for "vector".
118 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
119 ARGS as unused.
121 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
123 PR debug/84342
124 PR debug/84319
125 * common.opt (gas-loc-support, gas-locview-support): New.
126 (ginline-points, ginternal-reset-location-views): New.
127 * doc/invoke.texi: Document them. Use @itemx where intended.
128 (gvariable-location-views): Adjust.
129 * target.def (reset_location_view): New.
130 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
131 (TARGET_RESET_LOCATION_VIEW): New.
132 * doc/tm.texi: Rebuilt.
133 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
134 (dwarf2out_default_as_locview_support): New.
135 (output_asm_line_debug_info): Use option variables.
136 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
137 (output_loc_list): Likewise.
138 (add_high_low_attributes): Check option variables.
139 Don't output entry view attribute in strict mode.
140 (gen_inlined_subroutine_die): Check option variables.
141 (dwarf2out_inline_entry): Likewise.
142 (init_sections_and_labels): Likewise.
143 (dwarf2out_early_finish): Likewise.
144 (maybe_reset_location_view): New, from...
145 (dwarf2out_var_location): ... here. Call it.
146 * debug.h (dwarf2out_default_as_loc_support): Declare.
147 (dwarf2out_default_as_locview_support): Declare.
148 * hooks.c (hook_int_rtx_insn_0): New.
149 * hooks.h (hook_int_rtx_insn_0): Declare.
150 * toplev.c (process_options): Take -gas-loc-support and
151 -gas-locview-support from dwarf2out. Enable
152 -gvariable-location-views by default only with locview
153 assembler support. Enable -ginternal-reset-location-views by
154 default only if the target defines the corresponding hook.
155 Enable -ginline-points by default if location views are
156 enabled; force it disabled if statement frontiers are
157 disabled.
158 * tree-inline.c (expand_call_inline): Check option variables.
159 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
161 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
163 PR tree-optimization/84321
164 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
165 handling. Also check whether the anti-range contains any values
166 that satisfy the mask; switch to a VR_RANGE if not.
168 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
170 PR sanitizer/84340
171 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
173 2018-02-13 Martin Jambor <mjambor@suse.cz>
175 PR c++/83990
176 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
177 of call statements, also set location of a load to a temporary.
179 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
181 * config/rl78/rl78.c (add_vector_labels): New function.
182 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
183 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
184 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
185 which checks that no arguments are passed.
186 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
187 * doc/extend.texi: Documentation for the new attribute.
189 2018-02-13 Andreas Schwab <schwab@suse.de>
191 * config/riscv/linux.h (CPP_SPEC): Define.
193 2018-02-13 Jakub Jelinek <jakub@redhat.com>
195 PR target/84335
196 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
197 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
198 OPTION_MASK_ISA_AES as first argument to def_builtin_const
199 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
200 instead of OPTION_MASK_ISA_PCLMUL as first argument to
201 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
202 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
203 temporarily for AES and PCLMUL builtins.
205 PR tree-optimization/84339
206 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
207 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
208 Formatting fixes.
210 PR middle-end/84309
211 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
212 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
213 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
214 inline function.
215 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
216 inline function.
217 * omp-simd-clone.h: New file.
218 * omp-simd-clone.c: Include omp-simd-clone.h.
219 (expand_simd_clones): No longer static.
220 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
221 cgraph.h and omp-simd-clone.h.
222 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
223 (vect_recog_widen_shift_pattern): Formatting fix.
224 (vect_pattern_recog_1): Don't check optab for calls.
226 PR target/84336
227 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
228 operands[2] into a REG before using gen_lowpart on it.
230 2018-02-12 Jeff Law <law@redhat.com>
232 PR target/83760
233 * config/sh/sh.c (find_barrier): Consider a sibling call
234 a barrier as well.
236 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
237 successfully back substituting a reg.
239 2018-02-12 Richard Biener <rguenther@suse.de>
241 PR tree-optimization/84037
242 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
243 parameter, move visited init to caller.
244 (vect_slp_analyze_operations): Separate cost from validity
245 check, initialize visited once for all instances.
246 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
247 for all instances.
248 * tree-vect-stmts.c (vect_model_simple_cost): Make early
249 out an assert.
250 (vect_model_promotion_demotion_cost): Likewise.
251 (vectorizable_bswap): Guard cost modeling with !slp_node
252 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
253 SLP stmts.
254 (vectorizable_call): Likewise.
255 (vectorizable_conversion): Likewise.
256 (vectorizable_assignment): Likewise.
257 (vectorizable_shift): Likewise.
258 (vectorizable_operation): Likewise.
259 (vectorizable_store): Likewise.
260 (vectorizable_load): Likewise.
261 (vectorizable_condition): Likewise.
262 (vectorizable_comparison): Likewise.
264 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
266 PR sanitizer/84307
267 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
268 (ASAN_MARK): Fix fnspec to account for return value, change pointer
269 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
271 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
273 PR middle-end/83665
274 * params.def (inline-min-speedup): Increase from 8 to 15.
275 (max-inline-insns-auto): Decrease from 40 to 30.
276 * ipa-split.c (consider_split): Add some buffer for function to
277 be considered inlining candidate.
278 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
279 default values.
281 2018-02-12 Richard Biener <rguenther@suse.de>
283 PR tree-optimization/84037
284 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
285 matched stmts if we cannot swap the non-matched ones.
287 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
289 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
290 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
291 _mm_maskz_scalef_round_ss): New intrinsics.
292 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
293 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
294 __builtin_ia32_scalefss_round): Remove.
295 (__builtin_ia32_scalefsd_mask_round,
296 __builtin_ia32_scalefss_mask_round): New intrinsics.
297 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
298 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
299 ((match_operand:VF_128 2 "<round_nimm_predicate>"
300 "<round_constraint>")): Changed to ...
301 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
302 "<round_scalar_constraint>")): ... this.
303 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
304 %0, %1, %2<round_op3>}"): Changed to ...
305 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
306 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
307 %2<round_scalar_mask_op3>}"): ... this.
308 * config/i386/subst.md (round_scalar_nimm_predicate): New.
310 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
312 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
313 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
314 (_mm_maskz_sqrt_round_ss): New intrinsics.
315 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
316 (__builtin_ia32_sqrtsd_mask_round)
317 (__builtin_ia32_sqrtss_mask_round): New builtins.
318 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
319 (__builtin_ia32_sqrtss_round): Remove.
320 (__builtin_ia32_sqrtsd_mask_round)
321 (__builtin_ia32_sqrtss_mask_round): New builtins.
322 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
323 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
324 ((match_operand:VF_128 1 "vector_operand"
325 "xBm,<round_constraint>")): Changed to ...
326 ((match_operand:VF_128 1 "vector_operand"
327 "xBm,<round_scalar_constraint>")): ... this.
328 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
329 %0, %2, %<iptr>1<round_op3>}): Changed to ...
330 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
331 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
332 %<iptr>1<round_scalar_mask_op3>}): ... this.
333 ((set_attr "prefix" "<round_prefix>")): Changed to ...
334 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
336 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
338 PR target/84266
339 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
340 Cast vec_cmpeq result to correct type.
341 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
342 Cast vec_cmpgt result to correct type.
344 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
346 * final.c (final_scan_insn_1): Renamed from...
347 (final_scan_insn): ... this. New wrapper, to recover
348 seen from the outermost call in recursive ones.
349 * config/sparc/sparc.c (output_return): Drop seen from call.
350 (output_sibcall): Likewise.
351 * config/visium/visium.c (output_branch): Likewise.
353 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
355 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
356 function label.
358 2018-02-10 Alan Modra <amodra@gmail.com>
360 PR target/84300
361 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
362 Specify LR as an input.
364 2018-02-10 Jakub Jelinek <jakub@redhat.com>
366 PR sanitizer/83987
367 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
368 remove_member_access_dummy_vars): New functions.
369 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
370 lower_omp_1, execute_lower_omp): Use them.
372 PR rtl-optimization/84308
373 * shrink-wrap.c (spread_components): Release todo vector.
375 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
377 PR rtl-optimization/57193
378 * ira-color.c (struct allocno_color_data): Add member
379 conflict_allocno_hard_prefs.
380 (update_conflict_allocno_hard_prefs): New.
381 (bucket_allocno_compare_func): Add a preference based on
382 conflict_allocno_hard_prefs.
383 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
384 (color_allocnos): Remove a dead code. Initiate
385 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
387 2018-02-09 Jakub Jelinek <jakub@redhat.com>
389 PR target/84226
390 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
391 constraint from =wa to wa. Avoid a subreg on the output operand,
392 instead use a pseudo and subreg it in a move.
393 (p9_xxbrd_<mode>): Changed to ...
394 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
395 (p9_xxbrd_v2df): New expander.
396 (p9_xxbrw_<mode>): Changed to ...
397 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
398 (p9_xxbrw_v4sf): New expander.
400 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
402 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
404 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
406 PR target/83926
407 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
408 multiply in 32-bit mode.
409 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
410 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
411 mode.
413 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
415 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
416 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
417 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
418 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
420 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
422 PR lto/84213
423 * dwarf2out.c (is_trivial_indirect_ref): New function.
424 (dwarf2out_late_global_decl): Do not generate a location
425 attribute for variables that have a non-trivial DECL_VALUE_EXPR
426 and that are not defined in the current unit.
428 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
430 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
431 instead of a libcall for UNORDERED.
433 2018-02-09 Tamar Christina <tamar.christina@arm.com>
435 PR target/82641
436 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
437 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
439 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
441 PR target/PR84295
442 * config/s390/s390.c (s390_set_current_function): Invoke
443 s390_indirect_branch_settings also if fndecl didn't change.
445 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
447 * config/rs6000/rs6000.md (blockage): Set length to zero.
449 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
451 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
453 2018-02-09 Jakub Jelinek <jakub@redhat.com>
455 PR sanitizer/84285
456 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
457 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
458 -static-lib*san.
460 PR debug/84252
461 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
462 PARALLEL incoming that failed vt_get_decl_and_offset check.
464 PR middle-end/84237
465 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
466 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
467 TREE_READONLY bit.
468 (get_variable_section): For decls in named .bss* sections pass true as
469 second argument to bss_initializer_p.
471 2018-02-09 Marek Polacek <polacek@redhat.com>
472 Jakub Jelinek <jakub@redhat.com>
474 PR c++/83659
475 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
476 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
477 Sync some changes from cxx_fold_indirect_ref.
479 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
481 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
482 markers.
483 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
484 (BLOCK_INLINE_ENTRY_LABEL): New.
485 (dwarf2out_var_location): Disregard inline entry markers.
486 (inline_entry_data): New struct.
487 (inline_entry_data_hasher): New hashtable type.
488 (inline_entry_data_hasher::hash): New.
489 (inline_entry_data_hasher::equal): New.
490 (inline_entry_data_table): New variable.
491 (add_high_low_attributes): Add DW_AT_entry_pc and
492 DW_AT_GNU_entry_view attributes if a pending entry is found
493 in inline_entry_data_table. Add old entry_pc attribute only
494 if debug nonbinding markers are disabled.
495 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
496 markers are enabled.
497 (block_within_block_p, dwarf2out_inline_entry): New.
498 (dwarf2out_finish): Check that no entries remained in
499 inline_entry_data_table.
500 * final.c (reemit_insn_block_notes): Handle inline entry notes.
501 (final_scan_insn, notice_source_line): Likewise.
502 (rest_of_clean_state): Skip inline entry markers.
503 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
504 markers.
505 * gimple.c (gimple_build_debug_inline_entry): New.
506 * gimple.h (enum gimple_debug_subcode): Add
507 GIMPLE_DEBUG_INLINE_ENTRY.
508 (gimple_build_debug_inline_entry): Declare.
509 (gimple_debug_inline_entry_p): New.
510 (gimple_debug_nonbind_marker_p): Adjust.
511 * insn-notes.def (INLINE_ENTRY): New.
512 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
513 inline entry marker notes.
514 (print_insn): Likewise.
515 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
516 (INSN_DEBUG_MARKER_KIND): Likewise.
517 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
518 * tree-inline.c (expand_call_inline): Build and insert
519 debug_inline_entry stmt.
520 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
521 inline entry blocks early, if nonbind markers are enabled.
522 (dump_scope_block): Dump fragment info.
523 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
524 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
525 (gimple_build_debug_inline_entry): New.
526 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
527 Enable/disable inline entry points too.
528 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
529 (DEBUG_INSN): Describe inline entry markers.
531 * common.opt (gvariable-location-views): New.
532 (gvariable-location-views=incompat5): New.
533 * config.in: Rebuilt.
534 * configure: Rebuilt.
535 * configure.ac: Test assembler for view support.
536 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
537 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
538 * dwarf2out.c (var_loc_view): New typedef.
539 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
540 (dwarf2out_locviews_in_attribute): New.
541 (dwarf2out_locviews_in_loclist): New.
542 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
543 (enum dw_line_info_opcode): Add LI_adv_address.
544 (struct dw_line_info_table): Add view.
545 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
546 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
547 (zero_view_p): New variable.
548 (ZERO_VIEW_P): New macro.
549 (output_asm_line_debug_info): New.
550 (struct var_loc_node): Add view.
551 (add_AT_view_list, AT_loc_list): New.
552 (add_var_loc_to_decl): Add view param. Test it against last.
553 (new_loc_list): Add view params. Record them.
554 (AT_loc_list_ptr): Handle loc and view lists.
555 (view_list_to_loc_list_val_node): New.
556 (print_dw_val): Handle dw_val_class_view_list.
557 (size_of_die): Likewise.
558 (value_format): Likewise.
559 (loc_list_has_views): New.
560 (gen_llsym): Set vl_symbol too.
561 (maybe_gen_llsym, skip_loc_list_entry): New.
562 (dwarf2out_maybe_output_loclist_view_pair): New.
563 (output_loc_list): Output view list or entries too.
564 (output_view_list_offset): New.
565 (output_die): Handle dw_val_class_view_list.
566 (output_dwarf_version): New.
567 (output_compilation_unit_header): Use it.
568 (output_skeleton_debug_sections): Likewise.
569 (output_rnglists, output_line_info): Likewise.
570 (output_pubnames, output_aranges): Update version comments.
571 (output_one_line_info_table): Output view numbers in asm comments.
572 (dw_loc_list): Determine current endview, pass it to new_loc_list.
573 Call maybe_gen_llsym.
574 (loc_list_from_tree_1): Adjust.
575 (add_AT_location_description): Create view list attribute if
576 needed, check it's absent otherwise.
577 (convert_cfa_to_fb_loc_list): Adjust.
578 (maybe_emit_file): Call output_asm_line_debug_info for test.
579 (dwarf2out_var_location): Reset views as needed. Precompute
580 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
581 attribute. Set view.
582 (new_line_info_table): Reset next view.
583 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
584 (dwarf2out_source_line): Likewise. Output view resets and labels to
585 the assembler, or select appropriate line info opcodes.
586 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
587 (optimize_string_length): Catch it. Adjust.
588 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
589 dw_val_class_view_list, and remove it if no longer needed.
590 (hash_loc_list): Hash view numbers.
591 (loc_list_hasher::equal): Compare them.
592 (optimize_location_lists): Check whether a view list symbol is
593 needed, and whether the locview attribute is present, and
594 whether they match. Remove the locview attribute if no longer
595 needed.
596 (index_location_lists): Call skip_loc_list_entry for test.
597 (dwarf2out_finish): Call output_asm_line_debug_info for test.
598 Use output_dwarf_version.
599 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
600 (struct dw_val_node): Add val_view_list.
601 * final.c (SEEN_NEXT_VIEW): New.
602 (set_next_view_needed): New.
603 (clear_next_view_needed): New.
604 (maybe_output_next_view): New.
605 (final_start_function): Rename to...
606 (final_start_function_1): ... this. Take pointer to FIRST,
607 add SEEN parameter. Emit param bindings in the initial view.
608 (final_start_function): Reintroduce SEEN-less interface.
609 (final): Rename to...
610 (final_1): ... this. Take SEEN parameter. Output final pending
611 next view at the end.
612 (final): Reintroduce seen-less interface.
613 (final_scan_insn): Output pending next view before switching
614 sections or ending a block. Mark the next view as needed when
615 outputting variable locations. Notify debug backend of section
616 changes, and of location view changes.
617 (rest_of_handle_final): Adjust.
618 * toplev.c (process_options): Autodetect value for debug variable
619 location views option. Warn on incompat5 without -gdwarf-5.
620 * doc/invoke.texi (gvariable-location-views): New.
621 (gvariable-location-views=incompat5): New.
622 (gno-variable-location-views): New.
624 2018-02-08 David Malcolm <dmalcolm@redhat.com>
626 PR tree-optimization/84136
627 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
628 that the result of find_edge is non-NULL.
630 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
632 PR target/83008
633 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
634 storing integer register in SImode. Fix cost of 256 and 512
635 byte aligned SSE register store.
637 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
639 * config/i386/i386.c (ix86_multiplication_cost): Fix
640 multiplication cost for TARGET_AVX512DQ.
642 2018-02-08 Marek Polacek <polacek@redhat.com>
644 PR tree-optimization/84238
645 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
646 get_range_strlen.
648 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
650 PR tree-optimization/84265
651 * tree-vect-stmts.c (vectorizable_store): Don't treat
652 VMAT_CONTIGUOUS accesses as grouped.
653 (vectorizable_load): Likewise.
655 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
657 PR tree-optimization/81635
658 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
659 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
660 (test_round_for_mask): New functions.
661 (wide_int_cc_tests): Call test_round_for_mask.
662 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
663 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
664 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
665 range returned by get_range_info.
667 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
669 PR ipa/81360
670 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
671 * symtab.c: Include builtins.h
672 (symtab_node::output_to_lto_symbol_table_p): Move here
673 from lto-streamer-out.c:output_symbol_p.
674 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
675 (output_symbol_p): Move all logic to symtab.c
676 (produce_symtab): Update.
678 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
680 * config/s390/s390-opts.h (enum indirect_branch): Define.
681 * config/s390/s390-protos.h (s390_return_addr_from_memory)
682 (s390_indirect_branch_via_thunk)
683 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
684 (enum s390_indirect_branch_type): Define.
685 * config/s390/s390.c (struct s390_frame_layout, struct
686 machine_function): Remove.
687 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
688 (indirect_branch_table_label_no, indirect_branch_table_name):
689 Define variables.
690 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
691 (enum s390_indirect_branch_option): Define.
692 (s390_return_addr_from_memory): New function.
693 (s390_handle_string_attribute): New function.
694 (s390_attribute_table): Add new attribute handler.
695 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
696 (s390_indirect_branch_via_thunk): New function.
697 (s390_indirect_branch_via_inline_thunk): New function.
698 (s390_function_ok_for_sibcall): When jumping via thunk disallow
699 sibling call optimization for non z10 compiles.
700 (s390_emit_call): Force indirect branch target to be a single
701 register. Add r1 clobber for non-z10 compiles.
702 (s390_emit_epilogue): Emit return jump via return_use expander.
703 (s390_reorg): Handle JUMP_INSNs as execute targets.
704 (s390_option_override_internal): Perform validity checks for the
705 new command line options.
706 (s390_indirect_branch_attrvalue): New function.
707 (s390_indirect_branch_settings): New function.
708 (s390_set_current_function): Invoke s390_indirect_branch_settings.
709 (s390_output_indirect_thunk_function): New function.
710 (s390_code_end): Implement target hook.
711 (s390_case_values_threshold): Implement target hook.
712 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
713 macros.
714 * config/s390/s390.h (struct s390_frame_layout)
715 (struct machine_function): Move here from s390.c.
716 (TARGET_INDIRECT_BRANCH_NOBP_RET)
717 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
718 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
719 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
720 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
721 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
722 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
723 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
724 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
725 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
726 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
727 (mnemonic attribute): Add values which aren't recognized
728 automatically.
729 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
730 pattern for branch conversion. Fix mnemonic attribute.
731 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
732 indirect branch via thunk if requested.
733 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
734 ("*indirect_jump"): Disable for branch conversion using out of
735 line thunks.
736 ("indirect_jump_via_thunk<mode>_z10")
737 ("indirect_jump_via_thunk<mode>")
738 ("indirect_jump_via_inlinethunk<mode>_z10")
739 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
740 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
741 ("casesi_jump_via_inlinethunk<mode>_z10")
742 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
743 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
744 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
745 ("*indirect2_jump"): Disable for branch conversion.
746 ("casesi_jump"): Turn into expander and expand patterns for branch
747 conversion.
748 ("return_use"): New expander.
749 ("*return"): Emit return via thunk and rename it to ...
750 ("*return<mode>"): ... this one.
751 * config/s390/s390.opt: Add new options and and enum for the
752 option values.
754 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
756 * lra-constraints.c (match_reload): Unconditionally use
757 gen_lowpart_SUBREG, rather than selecting between that
758 and equivalent gen_rtx_SUBREG code.
760 2018-02-08 Richard Biener <rguenther@suse.de>
762 PR tree-optimization/84233
763 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
764 changed flag instead of boguously re-using phi_inserted.
766 2018-02-08 Martin Jambor <mjambor@suse.cz>
768 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
769 static local variables.
771 2018-02-08 Richard Biener <rguenther@suse.de>
773 PR tree-optimization/84278
774 * tree-vect-stmts.c (vectorizable_store): When looking for
775 smaller vector types to perform grouped strided loads/stores
776 make sure the mode is supported by the target.
777 (vectorizable_load): Likewise.
779 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
781 * config/aarch64/aarch64.c (aarch64_components_for_bb):
782 Increase LDP/STP opportunities by adding adjacent callee-saves.
784 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
786 PR rtl-optimization/84068
787 PR rtl-optimization/83459
788 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
790 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
792 PR tree-optimization/84224
793 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
794 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
795 non-zero arguments.
797 2018-02-07 Iain Sandoe <iain@codesourcery.com>
799 PR target/84113
800 * config/rs6000/altivec.md (*restore_world): Remove LR use.
801 * config/rs6000/predicates.md (restore_world_operation): Adjust op
802 count, remove one USE.
804 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
806 * doc/install.texi (Configuration): Document the
807 --with-long-double-format={ibm,ieee} PowerPC configuration
808 options.
810 PR target/84154
811 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
812 Convert from define_expand to be define_insn_and_split. Rework
813 float/double/_Float128 conversions to QI/HI/SImode to work with
814 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
815 conversions to QI/HImode types did a store and then a load to
816 truncate the value. For conversions to VSX registers, don't split
817 the insn, instead emit the code directly. Use the code iterator
818 any_fix to combine signed and unsigned conversions.
819 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
820 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
821 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
822 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
823 (fix_<mode>di2_hw): Likewise.
824 (fixuns_<mode>di2_hw): Likewise.
825 (fix_<mode>si2_hw): Likewise.
826 (fixuns_<mode>si2_hw): Likewise.
827 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
828 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
829 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
830 fix<uns>_trunc<SFDF:mode>si2_p8.
831 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
832 used.
833 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
834 (fix<uns>_<mode>_mem): Likewise.
835 (fctiw<u>z_<mode>_mem): Likewise.
836 (fix<uns>_<mode>_mem): Likewise.
837 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
838 the register allocator from doing a direct move to the GPRs to do
839 a store, and instead use the ISA 3.0 store byte/half-word from
840 vector register instruction. For IEEE 128-bit floating point,
841 also optimize stores of 32-bit ints.
842 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
844 2018-02-07 Alan Hayward <alan.hayward@arm.com>
846 * genextract.c (push_pathstr_operand): New function to support
847 [a-zA-Z].
848 (walk_rtx): Call push_pathstr_operand.
849 (print_path): Support [a-zA-Z].
851 2018-02-07 Richard Biener <rguenther@suse.de>
853 PR tree-optimization/84037
854 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
855 (cse_and_gimplify_to_preheader): Declare.
856 (vect_get_place_in_interleaving_chain): Likewise.
857 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
858 ivexpr_map.
859 (_loop_vec_info::~_loop_vec_info): Delete it.
860 (cse_and_gimplify_to_preheader): New function.
861 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
862 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
863 (vectorizable_load): Likewise. For grouped stores always base
864 the IV on the first element.
865 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
866 condition before gimplifying.
868 2018-02-07 Jakub Jelinek <jakub@redhat.com>
870 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
871 *DIV_EXPR and *MOD_EXPR.
873 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
875 PR target/84248
876 * config/i386/i386.c (ix86_option_override_internal): Mask out
877 the CF_SET bit when checking -fcf-protection.
879 2018-02-07 Tom de Vries <tom@codesourcery.com>
881 PR libgomp/84217
882 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
883 enough.
885 2018-02-07 Richard Biener <rguenther@suse.de>
887 PR tree-optimization/84204
888 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
889 this place.
891 PR tree-optimization/84205
892 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
893 special-case isl_ast_op_zdiv_r.
895 PR tree-optimization/84223
896 * graphite-scop-detection.c (gather_bbs::before_dom_children):
897 Only add conditions from within the region.
898 (gather_bbs::after_dom_children): Adjust.
900 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
902 PR target/84209
903 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
904 * config/avr/avr.md: Only post-reload split REG-REG moves if
905 either register is REGERAL_REG_P.
907 2018-02-07 Jakub Jelinek <jakub@redhat.com>
909 PR tree-optimization/84235
910 * tree-ssa-scopedtables.c
911 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
912 if the subtraction is performed in floating point type where NaNs are
913 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
914 build 1. Formatting fix.
916 2018-02-06 Jakub Jelinek <jakub@redhat.com>
918 PR target/84146
919 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
920 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
921 and skip it regardless of bb boundaries. Use CALL_P macro,
922 don't test INSN_P (insn) together with CALL_P or JUMP_P check
923 unnecessarily, formatting fix.
925 2018-02-06 Michael Collison <michael.collison@arm.com>
927 * config/arm/thumb2.md:
928 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
929 (*thumb_mov_notscc): Ditto.
931 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
933 PR target/84154
934 * config/rs6000/rs6000.md (su code attribute): Use "u" for
935 unsigned_fix, not "s".
937 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
939 * configure.ac (gcc_fn_eh_frame_ro): New function.
940 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
941 correct .eh_frame permissions.
942 * configure: Regenerate.
944 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
946 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
947 irrelevant options.
949 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
951 * config/rs6000/rs6000.c (rs6000_option_override_internal):
952 Display warning message for -mno-speculate-indirect-jumps.
954 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
956 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
957 Undocumented.
958 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
960 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
962 PR tree-optimization/84225
963 * tree-eh.c (find_trapping_overflow): Only call
964 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
966 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
968 PR target/84145
969 * config/i386/i386.c: Reimplement the check of possible options
970 -mibt/-mshstk conbination. Change error messages.
971 * doc/invoke.texi: Fix a typo: remove extra '='.
973 2018-02-06 Marek Polacek <polacek@redhat.com>
975 PR tree-optimization/84228
976 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
978 2018-02-06 Tamar Christina <tamar.christina@arm.com>
980 PR target/82641
981 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
982 emitted arch directives.
983 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
984 __ARM_FEATURE_COPROC before changing architectures.
986 2018-02-06 Richard Biener <rguenther@suse.de>
988 * config/i386/i386.c (print_reg): Fix typo.
989 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
991 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
993 * configure: Regenerate.
995 2018-02-05 Martin Sebor <msebor@redhat.com>
997 PR tree-optimization/83369
998 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
999 inlining context.
1001 2018-02-05 Martin Liska <mliska@suse.cz>
1003 * doc/invoke.texi: Cherry-pick upstream r323995.
1005 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1007 * ira.c (ira_init_register_move_cost): Adjust comment.
1009 2018-02-05 Martin Liska <mliska@suse.cz>
1011 PR gcov-profile/84137
1012 * doc/gcov.texi: Fix typo in documentation.
1014 2018-02-05 Martin Liska <mliska@suse.cz>
1016 PR gcov-profile/83879
1017 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1018 using dlopen functionality.
1020 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1022 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1023 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1024 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1025 _mm_maskz_range_round_ss): New intrinsics.
1026 (__builtin_ia32_rangesd128_round)
1027 (__builtin_ia32_rangess128_round): Remove.
1028 (__builtin_ia32_rangesd128_mask_round,
1029 __builtin_ia32_rangess128_mask_round): New builtins.
1030 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1031 __builtin_ia32_rangess128_round): Remove.
1032 (__builtin_ia32_rangesd128_mask_round,
1033 __builtin_ia32_rangess128_mask_round): New builtins.
1034 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1035 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1036 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1037 "<round_saeonly_constraint>")): Changed to ...
1038 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1039 "<round_saeonly_scalar_constraint>")): ... this.
1040 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1041 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1042 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1043 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1044 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1046 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1048 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1049 options.
1050 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1051 Remove all values except native, 8540 and 8548.
1053 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1055 * config/i386/i386.c (ix86_output_function_return): Pass
1056 INVALID_REGNUM, instead of -1, as invalid register number to
1057 indirect_thunk_name and output_indirect_thunk.
1059 2018-02-02 Julia Koval <julia.koval@intel.com>
1061 * config.gcc: Add -march=icelake.
1062 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1063 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1064 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1065 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1066 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1067 (processor_target_table): Add icelake.
1068 (ix86_option_override_internal): Handle new PTAs.
1069 (get_builtin_code_for_version): Handle icelake.
1070 (M_INTEL_COREI7_ICELAKE): New.
1071 (fold_builtin_cpu): Handle icelake.
1072 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1073 * doc/invoke.texi: Add -march=icelake.
1075 2018-02-02 Julia Koval <julia.koval@intel.com>
1077 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1078 to wide_int_bitmask.
1079 * wide-int-bitmask.h: New.
1081 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1083 PR target/84066
1084 * config/i386/i386.md: Replace Pmode with word_mode in
1085 builtin_setjmp_setup and builtin_longjmp to support x32.
1087 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1089 PR target/56010
1090 PR target/83743
1091 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1092 #include "opts.h".
1093 (rs6000_supported_cpu_names): New static variable.
1094 (linux_cpu_translation_table): Likewise.
1095 (elf_platform) <cpu>: Define new static variable and use it.
1096 Translate kernel AT_PLATFORM name to canonical name if needed.
1097 Error if platform name is unknown.
1099 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1101 PR target/84089
1102 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1104 2018-02-01 Jeff Law <law@redhat.com>
1106 PR target/84128
1107 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1108 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1109 the scratch if RELEASE_VIA_POP is false.
1110 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1111 If we have to save a temporary register, decrement SIZE appropriately.
1112 Pass new arguments to release_scratch_register_on_entry.
1113 (ix86_adjust_stack_and_probe): Likewise.
1114 (ix86_emit_probe_stack_range): Pass new arguments to
1115 release_scratch_register_on_entry.
1117 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1119 PR rtl-optimization/84157
1120 * combine.c (change_zero_ext): Use REG_P predicate in
1121 front of HARD_REGISTER_P predicate.
1123 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1125 * config/avr/avr.c (avr_option_override): Move disabling of
1126 -fdelete-null-pointer-checks to...
1127 * common/config/avr/avr-common.c (avr_option_optimization_table):
1128 ...here.
1130 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1132 PR tree-optimization/81635
1133 * tree-data-ref.c (split_constant_offset_1): For types that
1134 wrap on overflow, try to use range info to prove that wrapping
1135 cannot occur.
1137 2018-02-01 Renlin Li <renlin.li@arm.com>
1139 PR target/83370
1140 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1141 TAILCALL_ADDR_REGS.
1142 (aarch64_register_move_cost): Likewise.
1143 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1144 TAILCALL_ADDR_REGS.
1145 (REG_CLASS_NAMES): Likewise.
1146 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1147 TAILCALL_ADDR_REGS. Remove IP registers.
1148 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1150 2018-02-01 Richard Biener <rguenther@suse.de>
1152 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1153 for specifying RPO order and allow NULL for that.
1154 * domwalk.c (dom_walker::dom_walker): Likewise.
1155 (dom_walker::walk): Handle NULL RPO order.
1156 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1157 in RPO order.
1158 (rewrite_update_dom_walker): Likewise.
1159 (mark_def_dom_walker): Likewise.
1161 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1163 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1164 (aarch64_maybe_expand_sve_subreg_move): Declare.
1165 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1166 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1167 predicate.
1168 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1169 that are semantically a reverse operation.
1170 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1171 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1172 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1173 functions.
1174 (aarch64_can_change_mode_class): For big-endian, forbid changes
1175 between two SVE modes if they have different element sizes.
1177 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1179 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1180 the TImode handling for big-endian targets.
1182 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1184 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1185 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1186 not just bytes.
1187 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1188 Remove BSWAP handing for big-endian targets and use the form of
1189 LD1RQ appropariate for the mode.
1191 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1193 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1194 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1195 duplicated element.
1197 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1199 PR tearget/83845
1200 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1201 check for operands that need to go through aarch64_sve_reload_be.
1203 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1205 PR tree-optimization/81661
1206 PR tree-optimization/84117
1207 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1208 * tree-eh.c: Include gimplify.h.
1209 (find_trapping_overflow, replace_trapping_overflow,
1210 rewrite_to_non_trapping_overflow): New functions.
1211 * tree-vect-loop.c: Include tree-eh.h.
1212 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1213 * tree-data-ref.c: Include tree-eh.h.
1214 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1216 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1218 PR rtl-optimization/84123
1219 * combine.c (change_zero_ext): Check if hard register satisfies
1220 can_change_dest_mode before calling gen_lowpart_SUBREG.
1222 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1224 PR target/82444
1225 * ira.c (ira_init_register_move_cost): Remove assert.
1227 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1229 PR rtl-optimization/84071
1230 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1231 * doc/tm.texi: Regenerate.
1233 2018-01-31 Richard Biener <rguenther@suse.de>
1235 PR tree-optimization/84132
1236 * tree-data-ref.c (analyze_miv_subscript): Properly
1237 check whether evolution_function_is_affine_multivariate_p
1238 before calling gcd_of_steps_may_divide_p.
1240 2018-01-31 Julia Koval <julia.koval@intel.com>
1242 PR target/83618
1243 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1244 * config/i386/i386.md (rdpid_rex64) New.
1245 (rdpid): Make 32bit only.
1247 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1249 PR lto/84105
1250 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1251 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1253 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1255 Revert
1256 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1258 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1260 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1262 PR rtl-optimization/84071
1263 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1264 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1266 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1268 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1269 (arc_attribute_table): Add 'aux' attribute.
1270 (arc_in_small_data_p): Consider aux like variables.
1271 (arc_is_aux_reg_p): New function.
1272 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1273 (arc_get_aux_arg): New function.
1274 (prepare_move_operands): Handle aux-register access.
1275 (arc_handle_aux_attribute): New function.
1276 * doc/extend.texi (ARC Variable attributes): Add subsection.
1278 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1280 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1281 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1282 (arc_attribute_table): Add 'uncached' attribute.
1283 (arc_print_operand): Print '.di' flag for uncached memory
1284 accesses.
1285 (arc_in_small_data_p): Do not consider for small data the uncached
1286 types.
1287 (arc_is_uncached_mem_p): New function.
1288 * config/arc/predicates.md (compact_store_memory_operand): Check
1289 for uncached memory accesses.
1290 (nonvol_nonimm_operand): Likewise.
1291 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1293 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1295 PR c/84100
1296 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1297 falign-loops=): Add Optimization flag.
1299 2018-01-30 Jeff Law <law@redhat.com>
1301 PR target/84064
1302 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1303 INT_REGISTERS_SAVED. Check it prior to calling
1304 get_scratch_register_on_entry.
1305 (ix86_adjust_stack_and_probe): Similarly.
1306 (ix86_emit_probe_stack_range): Similarly.
1307 (ix86_expand_prologue): Corresponding changes.
1309 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1311 PR target/40411
1312 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1313 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1315 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1317 PR target/84112
1318 * lra-constraints.c (curr_insn_transform): Process AND in the
1319 address.
1321 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1323 PR rtl-optimization/83986
1324 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1325 dependence against last_pending_memory_flush in addition to
1326 pending_jump_insns.
1328 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1330 PR tree-optimization/81611
1331 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1332 copies.
1334 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1336 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1337 a reg rtx.
1339 2018-01-30 Richard Biener <rguenther@suse.de>
1340 Jakub Jelinek <jakub@redhat.com>
1342 PR tree-optimization/84111
1343 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1344 inner loops added during recursion, as they don't have up-to-date
1345 SSA form.
1347 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1349 PR ipa/81360
1350 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1351 (can_inline_edge_by_limits_p): ... here.
1352 (can_early_inline_edge_p, check_callers,
1353 update_caller_keys, update_callee_keys, recursive_inlining,
1354 add_new_edges_to_heap, speculation_useful_p,
1355 inline_small_functions,
1356 inline_small_functions, flatten_function,
1357 inline_to_all_callers_1): Update.
1359 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1361 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1362 zeros correctly.
1364 2018-01-30 Richard Biener <rguenther@suse.de>
1366 PR tree-optimization/83008
1367 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1368 invariant and constant vector uses in stmts when they need
1369 more than one stmt.
1371 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1373 PR bootstrap/84017
1374 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1375 * configure: Regenerate.
1377 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1379 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1380 pattern.
1381 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1382 Use gen_rtx_REG rather than gen_lowpart.
1384 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1386 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1387 rather than 0 when creating partial subregs.
1389 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1391 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1392 of usage.
1394 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1396 PR target/81550
1397 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1398 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1399 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1400 flags. This restores the settings used before the 2017-07-24.
1401 Turning off pre increment/decrement/modify allows IVOPTS to
1402 optimize DF/SF loops where the index is an int.
1404 2018-01-29 Richard Biener <rguenther@suse.de>
1405 Kelvin Nilsen <kelvin@gcc.gnu.org>
1407 PR bootstrap/80867
1408 * tree-vect-stmts.c (vectorizable_call): Don't call
1409 targetm.vectorize_builtin_md_vectorized_function if callee is
1410 NULL.
1412 2018-01-22 Carl Love <cel@us.ibm.com>
1414 * doc/extend.tex: Fix typo in second arg in
1415 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1417 2018-01-29 Richard Biener <rguenther@suse.de>
1419 PR tree-optimization/84086
1420 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1421 (flush_ssaname_freelist): When SSA names were released reset
1422 the SCEV hash table.
1424 2018-01-29 Richard Biener <rguenther@suse.de>
1426 PR tree-optimization/84057
1427 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1428 removed paths when removing edges.
1430 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1432 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1433 -mfunction-return=@var{choice}.
1435 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1437 PR diagnostic/84034
1438 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1439 Handle CR like TAB.
1440 (layout::print_source_line): Likewise.
1441 (test_get_line_width_without_trailing_whitespace): Add test cases.
1443 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1445 PR middle-end/84040
1446 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1447 debug insns.
1449 2018-01-26 Jim Wilson <jimw@sifive.com>
1451 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1453 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1454 specified.
1456 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1458 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1459 and CMP + SUB-immediate -> SUBS.
1461 2018-01-26 Martin Sebor <msebor@redhat.com>
1463 PR tree-optimization/83896
1464 * tree-ssa-strlen.c (get_string_len): Rename...
1465 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1466 Avoid assuming length is constant.
1467 (handle_char_store): Use HOST_WIDE_INT for string length.
1469 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1471 PR target/81763
1472 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1473 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1475 2018-01-26 Richard Biener <rguenther@suse.de>
1477 PR rtl-optimization/84003
1478 * dse.c (record_store): Only record redundant stores when
1479 the earlier store aliases at least all accesses the later one does.
1481 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1483 PR rtl-optimization/83985
1484 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1485 REG_CFA_RESTORE insns.
1486 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1487 REG_CFA_RESTORE insns here.
1489 PR c/83989
1490 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1491 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1493 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1495 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1496 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1497 (arc_init): Likewise.
1498 (arc_override_options): Likewise.
1499 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1500 value.
1501 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1502 support.
1503 * config/arc/arc.h (TARGET_DBNZ): Define.
1504 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1505 properly set the tune attribute.
1506 (dbnz): Use TARGET_DBNZ guard.
1507 * config/arc/arc.opt (mtune): Add core3 option.
1509 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1511 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1512 recognize new pic like addresses.
1513 (arc_delegitimize_address): Clean up.
1515 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1517 * config/arc/arc-arches.def: Option mrf16 valid for all
1518 architectures.
1519 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1520 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1521 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1522 * config/arc/arc-tables.opt: Regenerate.
1523 * config/arc/arc.c (arc_conditional_register_usage): Handle
1524 reduced register file case.
1525 (arc_file_start): Set must have build attributes.
1526 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1527 mrf16 option value.
1528 * config/arc/arc.opt (mrf16): Add new option.
1529 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1530 * config/arc/genmultilib.awk: Handle new mrf16 option.
1531 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1532 * config/arc/t-multilib: Regenerate.
1533 * doc/invoke.texi (ARC Options): Document mrf16 option.
1535 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1537 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1538 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1539 (arc_attribute_table): Add 'secure_call' attribute.
1540 (arc_print_operand): Print secure call operand.
1541 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1542 secure.
1543 (arc_is_secure_call_p): New function. * config/arc/arc.md
1544 (call_i): Add support for sjli instruction.
1545 (call_value_i): Likewise.
1546 * config/arc/constraints.md (Csc): New constraint.
1548 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1549 John Eric Martin <John.Martin@emmicro-us.com>
1551 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1552 * config/arc/arc.c (_arc_jli_section): New struct.
1553 (arc_jli_section): New type.
1554 (rc_jli_sections): New static variable.
1555 (arc_handle_jli_attribute): New function.
1556 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1557 (arc_file_end): New function.
1558 (TARGET_ASM_FILE_END): Define.
1559 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1560 (arc_add_jli_section): New function.
1561 (jli_call_scan): Likewise.
1562 (arc_reorg): Call jli_call_scan.
1563 (arc_output_addsi): Remove 'S' from printing asm operand.
1564 (arc_is_jli_call_p): New function.
1565 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1566 operand.
1567 (movhi_insn): Likewise.
1568 (movsi_insn): Likewise.
1569 (movsi_set_cc_insn): Likewise.
1570 (loadqi_update): Likewise.
1571 (load_zeroextendqisi_update): Likewise.
1572 (load_signextendqisi_update): Likewise.
1573 (loadhi_update): Likewise.
1574 (load_zeroextendhisi_update): Likewise.
1575 (load_signextendhisi_update): Likewise.
1576 (loadsi_update): Likewise.
1577 (loadsf_update): Likewise.
1578 (movsicc_insn): Likewise.
1579 (bset_insn): Likewise.
1580 (bxor_insn): Likewise.
1581 (bclr_insn): Likewise.
1582 (bmsk_insn): Likewise.
1583 (bicsi3_insn): Likewise.
1584 (cmpsi_cc_c_insn): Likewise.
1585 (movsi_ne): Likewise.
1586 (movsi_cond_exec): Likewise.
1587 (clrsbsi2): Likewise.
1588 (norm_f): Likewise.
1589 (normw): Likewise.
1590 (swap): Likewise.
1591 (divaw): Likewise.
1592 (flag): Likewise.
1593 (sr): Likewise.
1594 (kflag): Likewise.
1595 (ffs): Likewise.
1596 (ffs_f): Likewise.
1597 (fls): Likewise.
1598 (call_i): Remove 'S' asm letter, add jli instruction.
1599 (call_value_i): Likewise.
1600 * config/arc/arc.op (mjli-always): New option.
1601 * config/arc/constraints.md (Cji): New constraint.
1602 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1603 operand.
1604 (subsf3_fpx): Likewise.
1605 (mulsf3_fpx): Likewise.
1606 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1607 asm operand.
1608 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1609 function attrbutes.
1610 * doc/invoke.texi (ARC): Document mjli-always option.
1612 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1614 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
1615 avoid addition with 0 and use incw and decw where possible.
1617 2018-01-26 Richard Biener <rguenther@suse.de>
1619 PR tree-optimization/81082
1620 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1621 association if it requires casting to unsigned.
1622 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1623 from fold_plusminus_mult_expr to catch important cases late when
1624 range info is available.
1626 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1628 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1629 * configure.ac (hidden_linkonce): New test.
1630 * configure: Regenerate.
1631 * config.in: Regenerate.
1633 2018-01-26 Julia Koval <julia.koval@intel.com>
1635 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
1636 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
1637 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
1638 _mm_mask_bitshuffle_epi64_mask): Fix type.
1639 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
1640 USI_FTYPE_V4DI_V4DI_USI): Remove.
1641 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
1642 __builtin_ia32_vpshufbitqmb256_mask,
1643 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
1644 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
1645 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
1647 2018-01-26 Alan Modra <amodra@gmail.com>
1649 PR target/84033
1650 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
1651 UNSPEC_VBPERMQ. Sort other unspecs.
1653 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
1655 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
1657 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
1659 PR middle-end/83055
1660 * predict.c (drop_profile): Do not push/pop cfun; update also
1661 node->count.
1662 (handle_missing_profiles): Fix logic looking for zero profiles.
1664 2018-01-25 Jakub Jelinek <jakub@redhat.com>
1666 PR middle-end/83977
1667 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
1668 on functions with #pragma omp declare simd or functions with simd
1669 attribute.
1670 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
1671 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
1672 Remove trailing \n from warning_at calls.
1674 2018-01-25 Tom de Vries <tom@codesourcery.com>
1676 PR target/84028
1677 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1678 for neutered workers.
1680 2018-01-24 Joseph Myers <joseph@codesourcery.com>
1682 PR target/68467
1683 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
1684 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
1686 2018-01-24 Jeff Law <law@redhat.com>
1688 PR target/83994
1689 * i386.c (get_probe_interval): Move to earlier point.
1690 (ix86_compute_frame_layout): If -fstack-clash-protection and
1691 the frame is larger than the probe interval, then use pushes
1692 to save registers rather than reg->mem moves.
1693 (ix86_expand_prologue): Remove conditional for int_registers_saved
1694 assertion.
1696 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
1698 PR target/84014
1699 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
1700 min/max for never referenced object.
1702 2018-01-24 Jakub Jelinek <jakub@redhat.com>
1704 PR middle-end/83977
1705 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
1706 here.
1707 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
1708 attributes from DECL_ATTRIBUTES (decl) without affecting
1709 DECL_ATTRIBUTES (current_function_decl).
1710 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
1711 functions with non-NULL DECL_ABSTRACT_ORIGIN.
1713 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
1715 PR tree-optimization/83979
1716 * fold-const.c (fold_comparison): Use constant_boolean_node
1717 instead of boolean_{true,false}_node.
1719 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
1721 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
1722 with zero counts.
1724 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1726 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1727 Simplify the clause that sets the length attribute.
1728 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1729 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
1730 clause that sets the length attribute.
1731 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1733 2018-01-24 Tom de Vries <tom@codesourcery.com>
1735 PR target/83589
1736 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
1737 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
1738 Add strict parameter.
1739 (prevent_branch_around_nothing): Insert dummy insn between branch to
1740 label and label with no ptx insn inbetween.
1741 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
1743 2018-01-24 Tom de Vries <tom@codesourcery.com>
1745 PR target/81352
1746 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
1747 for neutered threads in warp.
1748 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
1750 2018-01-24 Richard Biener <rguenther@suse.de>
1752 PR tree-optimization/83176
1753 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
1754 operands.
1756 2018-01-24 Richard Biener <rguenther@suse.de>
1758 PR tree-optimization/82819
1759 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
1760 code generating pluses that are no-ops in the target precision.
1762 2018-01-24 Richard Biener <rguenther@suse.de>
1764 PR middle-end/84000
1765 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
1767 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1769 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
1770 to merge probabilities.
1771 * predict.c (probably_never_executed): Also mark as cold functions
1772 with global 0 profile and guessed local profile.
1773 * profile-count.c (profile_probability::combine_with_count): New
1774 member function.
1775 * profile-count.h (profile_probability::operator*,
1776 profile_probability::operator*=, profile_probability::operator/,
1777 profile_probability::operator/=): Reduce precision to adjusted
1778 and set value to guessed on contradictory divisions.
1779 (profile_probability::combine_with_freq): Remove.
1780 (profile_probability::combine_wiht_count): Declare.
1781 (profile_count::force_nonzero):: Set to adjusted.
1782 (profile_count::probability_in):: Set quality to adjusted.
1783 * tree-ssa-tail-merge.c (replace_block_by): Use
1784 combine_with_count.
1786 2018-01-23 Andrew Waterman <andrew@sifive.com>
1787 Jim Wilson <jimw@sifive.com>
1789 * config/riscv/riscv.c (riscv_stack_boundary): New.
1790 (riscv_option_override): Set riscv_stack_boundary. Handle
1791 riscv_preferred_stack_boundary_arg.
1792 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
1793 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
1794 (STACK_BOUNDARY): Set to riscv_stack_boundary.
1795 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1796 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1797 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1799 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1801 PR target/83905
1802 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1803 of struct ix86_frame.
1804 (ix86_expand_epilogue): Likewise. Add a local variable for
1805 the reg_save_offset field in struct ix86_frame.
1807 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1809 PR tree-optimization/82604
1810 * tree-loop-distribution.c (enum partition_kind): New enum item
1811 PKIND_PARTIAL_MEMSET.
1812 (partition_builtin_p): Support above new enum item.
1813 (generate_code_for_partition): Ditto.
1814 (compute_access_range): Differentiate cases that equality can be
1815 proven at all loops, the innermost loops or no loops.
1816 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1817 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1818 (finalize_partitions, distribute_loop): Don't fuse partition of
1819 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1820 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1821 parloop is enabled.
1823 2018-01-23 Martin Liska <mliska@suse.cz>
1825 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1826 order to ignore the predictor.
1827 (PRED_POLYMORPHIC_CALL): Likewise.
1828 (PRED_RECURSIVE_CALL): Likewise.
1830 2018-01-23 Martin Liska <mliska@suse.cz>
1832 * tree-profile.c (tree_profiling): Print function header to
1833 aware reader which function we are working on.
1834 * value-prof.c (gimple_find_values_to_profile): Do not print
1835 not interesting value histograms.
1837 2018-01-23 Martin Liska <mliska@suse.cz>
1839 * profile-count.h (enum profile_quality): Add
1840 profile_uninitialized as the first value. Do not number values
1841 as they are zero based.
1842 (profile_count::verify): Update sanity check.
1843 (profile_probability::verify): Likewise.
1845 2018-01-23 Nathan Sidwell <nathan@acm.org>
1847 * doc/invoke.texi (ffor-scope): Deprecate.
1849 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1851 PR tree-optimization/83510
1852 * domwalk.c (set_all_edges_as_executable): New function.
1853 (dom_walker::dom_walker): Convert bool param
1854 "skip_unreachable_blocks" to enum reachability. Move setup of
1855 edge flags to set_all_edges_as_executable and only do it when
1856 reachability is REACHABLE_BLOCKS.
1857 * domwalk.h (enum dom_walker::reachability): New enum.
1858 (dom_walker::dom_walker): Convert bool param
1859 "skip_unreachable_blocks" to enum reachability.
1860 (set_all_edges_as_executable): New decl.
1861 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1862 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1863 "reachability".
1864 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1865 but converting true to REACHABLE_BLOCKS.
1866 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1867 * tree-vrp.c
1868 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1869 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1870 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1871 REACHABLE_BLOCKS.
1872 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1873 if check_all_array_refs will be called.
1875 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1877 * tree.c (selftest::test_location_wrappers): Add more test
1878 coverage.
1880 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1882 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1883 (selftest::test_bit_in_range): Likewise.
1885 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1887 PR testsuite/83888
1888 * doc/sourcebuild.texi (vect_float): Say that the selector
1889 only describes the situation when -funsafe-math-optimizations is on.
1890 (vect_float_strict): Document.
1892 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1894 PR tree-optimization/83965
1895 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1896 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1897 instead of checking only for a reduction.
1898 (vect_recog_widen_sum_pattern): Likewise.
1900 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1902 * predict.c (probably_never_executed): Only use precise profile info.
1903 (compute_function_frequency): Skip after inlining hack since we now
1904 have quality checking.
1906 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1908 * profile-count.h (profile_probability::very_unlikely,
1909 profile_probability::unlikely, profile_probability::even): Set
1910 precision to guessed.
1912 2018-01-23 Richard Biener <rguenther@suse.de>
1914 PR tree-optimization/83963
1915 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1916 Properly terminate dominator walk when crossing the exit edge not
1917 when visiting its source block.
1919 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1921 PR c++/83918
1922 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1923 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1925 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1927 PR tree-optimization/83957
1928 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1929 semicolon after for body surrounded by braces.
1931 PR tree-optimization/83081
1932 * profile-count.h (profile_probability::split): New method.
1933 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1934 Use profile_probability::split.
1935 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1936 when splitting a single conditional jump into 2.
1938 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1940 PR tree-optimization/69452
1941 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1942 decl.
1944 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1946 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
1947 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
1948 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
1950 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1952 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
1953 * config/rl78/rl78.md (movdi): New define_expand.
1954 * config/rl78/rl78.c (rl78_split_movdi): New function.
1956 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1958 PR target/83862
1959 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1960 no longer used.
1961 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1962 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1963 128-bit to produce an UNSPEC move to get the double word with the
1964 signbit and then a shift directly to do signbit.
1965 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1966 implementation with a new version that just does either a direct
1967 move or a regular move. Move memory interface to separate insns.
1968 Move insns so they are next to the expander.
1969 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1970 with signbit move. Split big and little endian case.
1971 (signbit<mode>2_dm_mem_le): Likewise.
1972 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1973 (signbit<mode>2_dm2): Likewise.
1975 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1977 * config/rl78/rl78.md (anddi3): New define_expand.
1979 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1981 * config/rl78/rl78.md (umindi3): New define_expand.
1983 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1985 * config/rl78/rl78.md (smindi3): New define_expand.
1987 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1989 * config/rl78/rl78.md (smaxdi3): New define_expand.
1991 2018-01-22 Carl Love <cel@us.ibm.com>
1993 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1994 LVX_V1TI): Add macro expansion.
1995 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1996 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1997 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1998 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1999 Change check to determine if the instruction is a byte reversing
2000 entry. Fix typo in comment.
2001 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2002 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2003 Add def_builtin calls for new builtins.
2004 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2005 Add define_insn expansion.
2007 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2009 * config/rl78/rl78.md (umaxdi3): New define_expand.
2011 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2013 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2014 for non-QImode registers.
2016 2018-01-22 Richard Biener <rguenther@suse.de>
2018 PR tree-optimization/83963
2019 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2020 including the loop exit block.
2021 (scop_detection::merge_sese): Likewise.
2022 (scop_detection::add_scop): Do it here instead.
2024 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2026 * doc/sourcebuild.texi (arm_softfloat): Document.
2028 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2030 PR gcc/77734
2031 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2032 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2033 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2035 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2036 David Edelsohn <dje.gcc@gmail.com>
2038 PR target/83946
2039 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2040 Change "crset eq" to "crset 2".
2041 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2042 (*call_indirect_aix<mode>_nospec): Likewise.
2043 (*call_value_indirect_aix<mode>_nospec): Likewise.
2044 (*call_indirect_elfv2<mode>_nospec): Likewise.
2045 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2046 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2047 change assembly output from . to $.
2048 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2049 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2050 (*tablejump<mode>_internal1_nospec): Likewise.
2052 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2054 PR target/80870
2055 * config/sh/sh_optimize_sett_clrt.cc:
2056 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2058 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2060 PR tree-optimization/83940
2061 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2062 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2063 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2064 use it to pass back the definition type.
2065 (vect_check_store_rhs): Likewise rhs_dt_out.
2066 (vect_build_gather_load_calls): Add a mask_dt argument and use
2067 it instead of a call to vect_is_simple_use.
2068 (vectorizable_store): Update calls to vect_check_load_store_mask
2069 and vect_check_store_rhs. Use the dt returned by the latter instead
2070 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2071 instead of calls to vect_is_simple_use. Pass the scalar rather
2072 than the vector operand to vect_is_simple_use when handling
2073 second and subsequent copies of an rhs value.
2074 (vectorizable_load): Update calls to vect_check_load_store_mask
2075 and vect_build_gather_load_calls. Use the cached mask_dt and
2076 gs_info.offset_dt instead of calls to vect_is_simple_use.
2078 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2080 PR middle-end/83945
2081 * tree-emutls.c: Include gimplify.h.
2082 (lower_emutls_2): New function.
2083 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2084 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2085 it before further processing.
2087 PR target/83930
2088 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2089 UINTVAL (trueop1) instead of INTVAL (op1).
2091 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2093 PR debug/81570
2094 PR debug/83728
2095 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2096 INCOMING_FRAME_SP_OFFSET if not defined.
2097 (scan_trace): Add ENTRY argument. If true and
2098 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2099 emit a note to adjust the CFA offset.
2100 (create_cfi_notes): Adjust scan_trace callers.
2101 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2102 INCOMING_FRAME_SP_OFFSET in the CIE.
2103 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2104 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2105 Likewise.
2106 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2107 * doc/tm.texi: Regenerated.
2109 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2111 PR rtl-optimization/83147
2112 * lra-constraints.c (remove_inheritance_pseudos): Use
2113 lra_substitute_pseudo_within_insn.
2115 2018-01-19 Tom de Vries <tom@codesourcery.com>
2116 Cesar Philippidis <cesar@codesourcery.com>
2118 PR target/83920
2119 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2121 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2123 PR target/83790
2124 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2125 spaces for function labels.
2127 2018-01-19 Martin Liska <mliska@suse.cz>
2129 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2130 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2131 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2132 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2133 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2134 (PRED_CONST_RETURN): Change from 69 to 65.
2135 (PRED_NULL_RETURN): Change from 91 to 71.
2136 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2137 (PRED_LOOP_GUARD): Change from 66 to 73.
2139 2018-01-19 Martin Liska <mliska@suse.cz>
2141 * predict.c (predict_insn_def): Add new assert.
2142 (struct branch_predictor): Change type to signed integer.
2143 (test_prediction_value_range): Amend test to cover
2144 PROB_UNINITIALIZED.
2145 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2146 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2147 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2148 (PRED_LOOP_IV_COMPARE): Likewise.
2149 * predict.h (PROB_UNINITIALIZED): Define new constant.
2151 2018-01-19 Martin Liska <mliska@suse.cz>
2153 * predict.c (dump_prediction): Add new format for
2154 analyze_brprob.py script which is enabled with -details
2155 suboption.
2156 * profile-count.h (precise_p): New function.
2158 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2160 PR tree-optimization/83922
2161 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2162 there are no statements that need masking.
2163 (vect_active_double_reduction_p): New function.
2164 (vect_analyze_loop_operations): Use it when handling phis that
2165 are not in the loop header.
2167 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2169 PR tree-optimization/83914
2170 * tree-vect-loop.c (vectorizable_induction): Don't convert
2171 init_expr or apply the peeling adjustment for inductions
2172 that are nested within the vectorized loop.
2174 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2176 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2177 instead of NEG.
2179 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2181 PR sanitizer/81715
2182 PR testsuite/83882
2183 * function.h (gimplify_parameters): Add gimple_seq * argument.
2184 * function.c: Include gimple.h and options.h.
2185 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2186 for the added local temporaries if needed.
2187 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2188 if there are any parameter cleanups, wrap whole body into a
2189 try/finally with the cleanups.
2191 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2193 PR target/82964
2194 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2195 Use GET_MODE_CLASS for scalar floating point.
2197 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2199 PR ipa/82256
2200 patch by PaX Team
2201 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2202 Fix call of call_cgraph_insertion_hooks.
2204 2018-01-18 Martin Sebor <msebor@redhat.com>
2206 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2208 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2210 PR ipa/83619
2211 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2212 frequencies.
2214 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2216 PR other/70268
2217 * common.opt: (-ffile-prefix-map): New option.
2218 * opts.c (common_handle_option): Defer it.
2219 * opts-global.c (handle_common_deferred_options): Handle it.
2220 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2221 * file-prefix-map.h: New file.
2222 (remap_debug_filename, add_debug_prefix_map): ...here.
2223 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2224 * final.c (debug_prefix_map, add_debug_prefix_map
2225 remap_debug_filename): Move to...
2226 * file-prefix-map.c: New file.
2227 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2228 generalize, get rid of alloca(), use strrchr() instead of strchr().
2229 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2230 Implement in terms of add_prefix_map().
2231 (remap_macro_filename, remap_debug_filename): Implement in term of
2232 remap_filename().
2233 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2234 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2235 * dbxout.c: Include file-prefix-map.h.
2236 * varasm.c: Likewise.
2237 * vmsdbgout.c: Likewise.
2238 * xcoffout.c: Likewise.
2239 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2240 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2241 * doc/invoke.texi (-ffile-prefix-map): Document.
2242 (-fdebug-prefix-map): Update description.
2244 2018-01-18 Martin Liska <mliska@suse.cz>
2246 * config/i386/i386.c (indirect_thunk_name): Document that also
2247 lfence is emitted.
2248 (output_indirect_thunk): Document why both instructions
2249 (pause and lfence) are generated.
2251 2018-01-18 Richard Biener <rguenther@suse.de>
2253 PR tree-optimization/83887
2254 * graphite-scop-detection.c
2255 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2256 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2257 (scop_detection::merge_sese): Re-implement with a flood-fill
2258 algorithm that properly finds a SESE region if it exists.
2260 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2262 PR c/61240
2263 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2264 pointer_diff optimizations use view_convert instead of convert.
2266 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2268 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2269 Generate different code for -mno-speculate-indirect-jumps.
2270 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2271 (*call_indirect_aix<mode>): Disable for
2272 -mno-speculate-indirect-jumps.
2273 (*call_indirect_aix<mode>_nospec): New define_insn.
2274 (*call_value_indirect_aix<mode>): Disable for
2275 -mno-speculate-indirect-jumps.
2276 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2277 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2278 -mno-speculate-indirect-jumps.
2279 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2281 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2283 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2284 long double type, set the flags for noting the default long double
2285 type, even if we don't pass or return a long double type.
2287 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2289 PR ipa/83051
2290 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2291 failure.
2293 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2295 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2296 support for merge[hl].
2297 (fold_mergehl_helper): New helper function.
2298 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2299 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2300 (altivec_vmrglw_direct): Add xxmrglw insn.
2302 2018-01-17 Andrew Waterman <andrew@sifive.com>
2304 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2305 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2307 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2309 PR lto/83121
2310 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2311 call the lto_location_cache before reading the
2312 DECL_SOURCE_LOCATION of the types.
2314 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2315 Richard Sandiford <richard.sandiford@linaro.org>
2317 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2318 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2319 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2320 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2321 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2322 Add declaration.
2323 * config/aarch64/constraints.md (aarch64_movti_operand):
2324 Limit immediates.
2325 * config/aarch64/predicates.md (Uti): Add new constraint.
2327 2018-01-17 Carl Love <cel@us.ibm.com>
2328 * config/rs6000/vsx.md (define_expand xl_len_r,
2329 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2330 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2331 lxvll.
2332 (define_expand, define_insn): Move the shift left from the
2333 define_insn to the define_expand for lxvl and stxvl instructions.
2334 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2335 and XL_LEN_R definitions to PURE.
2337 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2339 * config/i386/i386.c (indirect_thunk_name): Declare regno
2340 as unsigned int. Compare regno with INVALID_REGNUM.
2341 (output_indirect_thunk): Ditto.
2342 (output_indirect_thunk_function): Ditto.
2343 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2344 in the call to output_indirect_thunk_function.
2346 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2348 PR middle-end/83884
2349 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2350 rather than the size of inner_type to determine the stack slot size
2351 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2353 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2355 PR target/83546
2356 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2357 to PTA_SILVERMONT.
2359 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2361 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2362 endian Linux systems to optionally enable multilibs for selecting
2363 the long double type if the user configured an explicit type.
2364 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2365 have no long double multilibs if not defined.
2366 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2367 warn if the user used -mabi={ieee,ibm}longdouble and we built
2368 multilibs for long double.
2369 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2370 appropriate multilib option.
2371 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2372 multilib options.
2373 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2374 for building long double multilibs.
2375 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2377 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2379 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2380 copies.
2382 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2383 64 bits.
2384 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2385 128 bits.
2387 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2388 variables.
2390 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2391 return value.
2393 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2395 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2396 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2398 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2400 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2401 different rtl trees depending on TARGET_64BIT.
2402 (rs6000_gen_lvx): Likewise.
2404 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2406 * config/visium/visium.md (nop): Tweak comment.
2407 (hazard_nop): Likewise.
2409 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2411 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2412 -mspeculate-indirect-jumps.
2413 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2414 for -mno-speculate-indirect-jumps.
2415 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2416 (*call_value_indirect_elfv2<mode>): Disable for
2417 -mno-speculate-indirect-jumps.
2418 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2419 (indirect_jump): Emit different RTL for
2420 -mno-speculate-indirect-jumps.
2421 (*indirect_jump<mode>): Disable for
2422 -mno-speculate-indirect-jumps.
2423 (*indirect_jump<mode>_nospec): New define_insn.
2424 (tablejump): Emit different RTL for
2425 -mno-speculate-indirect-jumps.
2426 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2427 (tablejumpsi_nospec): New define_expand.
2428 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2429 (tablejumpdi_nospec): New define_expand.
2430 (*tablejump<mode>_internal1): Disable for
2431 -mno-speculate-indirect-jumps.
2432 (*tablejump<mode>_internal1_nospec): New define_insn.
2433 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2434 option.
2436 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2438 * caller-save.c (insert_save): Drop unnecessary parameter. All
2439 callers updated.
2441 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2442 Richard Biener <rguenth@suse.de>
2444 PR libgomp/83590
2445 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2446 return early, inline manually is_gimple_sizepos. Make sure if we
2447 call gimplify_expr we don't end up with a gimple constant.
2448 * tree.c (variably_modified_type_p): Don't return true for
2449 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2450 * gimplify.h (is_gimple_sizepos): Remove.
2452 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2454 PR tree-optimization/83857
2455 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2456 vectorizable_live_operation for pure SLP statements.
2457 (vectorizable_live_operation): Handle PHIs.
2459 2018-01-16 Richard Biener <rguenther@suse.de>
2461 PR tree-optimization/83867
2462 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2463 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2465 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2467 PR c/83844
2468 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2469 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2470 If off is not INTEGER_CST, issue a may not be aligned warning
2471 rather than isn't aligned. Use isn%'t rather than isn't.
2472 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2473 into MULT_EXPR.
2474 <case MULT_EXPR>: Improve the case when bottom and one of the
2475 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2476 operand, in that case check if the other operand is multiple of
2477 bottom divided by the INTEGER_CST operand.
2479 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2481 PR target/83858
2482 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2483 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2484 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2485 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2486 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2487 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2488 (pa_function_arg_size): New function.
2490 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2492 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2493 in a separate statement.
2495 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2497 PR tree-optimization/83847
2498 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2499 group gathers and scatters.
2501 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2503 PR rtl-optimization/86620
2504 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2506 PR rtl-optimization/83213
2507 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2508 to last if both are JUMP_INSNs.
2510 PR tree-optimization/83843
2511 * gimple-ssa-store-merging.c
2512 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2513 store_immediate_info for bswap/nop orig_stores.
2515 2018-01-15 Andrew Waterman <andrew@sifive.com>
2517 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2518 !TARGET_MUL.
2519 <UDIV>: Increase cost if !TARGET_DIV.
2521 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2523 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2524 (define_attr "cr_logical_3op"): New.
2525 (cceq_ior_compare): Adjust.
2526 (cceq_ior_compare_complement): Adjust.
2527 (*cceq_rev_compare): Adjust.
2528 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2529 (is_cracked_insn): Adjust.
2530 (insn_must_be_first_in_group): Adjust.
2531 * config/rs6000/40x.md: Adjust.
2532 * config/rs6000/440.md: Adjust.
2533 * config/rs6000/476.md: Adjust.
2534 * config/rs6000/601.md: Adjust.
2535 * config/rs6000/603.md: Adjust.
2536 * config/rs6000/6xx.md: Adjust.
2537 * config/rs6000/7450.md: Adjust.
2538 * config/rs6000/7xx.md: Adjust.
2539 * config/rs6000/8540.md: Adjust.
2540 * config/rs6000/cell.md: Adjust.
2541 * config/rs6000/e300c2c3.md: Adjust.
2542 * config/rs6000/e500mc.md: Adjust.
2543 * config/rs6000/e500mc64.md: Adjust.
2544 * config/rs6000/e5500.md: Adjust.
2545 * config/rs6000/e6500.md: Adjust.
2546 * config/rs6000/mpc.md: Adjust.
2547 * config/rs6000/power4.md: Adjust.
2548 * config/rs6000/power5.md: Adjust.
2549 * config/rs6000/power6.md: Adjust.
2550 * config/rs6000/power7.md: Adjust.
2551 * config/rs6000/power8.md: Adjust.
2552 * config/rs6000/power9.md: Adjust.
2553 * config/rs6000/rs64.md: Adjust.
2554 * config/rs6000/titan.md: Adjust.
2556 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2558 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2559 ix86_indirect_branch_register logic.
2561 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2563 * config/i386/constraints.md (Bs): Update
2564 ix86_indirect_branch_register check. Don't check
2565 ix86_indirect_branch_register with GOT_memory_operand.
2566 (Bw): Likewise.
2567 * config/i386/predicates.md (GOT_memory_operand): Don't check
2568 ix86_indirect_branch_register here.
2569 (GOT32_symbol_operand): Likewise.
2571 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2573 * config/i386/predicates.md (constant_call_address_operand):
2574 Rewrite ix86_indirect_branch_register logic.
2575 (sibcall_insn_operand): Likewise.
2577 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2579 * config/i386/constraints.md (Bs): Replace
2580 ix86_indirect_branch_thunk_register with
2581 ix86_indirect_branch_register.
2582 (Bw): Likewise.
2583 * config/i386/i386.md (indirect_jump): Likewise.
2584 (tablejump): Likewise.
2585 (*sibcall_memory): Likewise.
2586 (*sibcall_value_memory): Likewise.
2587 Peepholes of indirect call and jump via memory: Likewise.
2588 * config/i386/i386.opt: Likewise.
2589 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2590 (GOT_memory_operand): Likewise.
2591 (call_insn_operand): Likewise.
2592 (sibcall_insn_operand): Likewise.
2593 (GOT32_symbol_operand): Likewise.
2595 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2597 PR middle-end/83837
2598 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2599 type rather than type addr's type points to.
2600 (expand_omp_atomic_mutex): Likewise.
2601 (expand_omp_atomic): Likewise.
2603 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2605 PR target/83839
2606 * config/i386/i386.c (output_indirect_thunk_function): Use
2607 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2608 for __x86_return_thunk.
2610 2018-01-15 Richard Biener <rguenther@suse.de>
2612 PR middle-end/83850
2613 * expmed.c (extract_bit_field_1): Fix typo.
2615 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2617 PR target/83687
2618 * config/arm/iterators.md (VF): New mode iterator.
2619 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2620 Remove integer-related logic from pattern.
2621 (neon_vabd<mode>_3): Likewise.
2623 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2625 PR middle-end/82694
2626 * common.opt (fstrict-overflow): No longer an alias.
2627 (fwrapv-pointer): New option.
2628 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2629 also for pointer types based on flag_wrapv_pointer.
2630 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
2631 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
2632 opts->x_flag_wrapv got set.
2633 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
2634 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
2635 POINTER_TYPE_OVERFLOW_UNDEFINED.
2636 * match.pd: Likewise in address comparison pattern.
2637 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
2639 2018-01-15 Richard Biener <rguenther@suse.de>
2641 PR lto/83804
2642 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
2643 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
2644 Reset type names to their identifier if their TYPE_DECL doesn't
2645 have linkage (and thus is used for ODR and devirt).
2646 (save_debug_info_for_decl): Remove.
2647 (save_debug_info_for_type): Likewise.
2648 (add_tree_to_fld_list): Adjust.
2649 * tree-pretty-print.c (dump_generic_node): Make dumping of
2650 type names more robust.
2652 2018-01-15 Richard Biener <rguenther@suse.de>
2654 * BASE-VER: Bump to 8.0.1.
2656 2018-01-14 Martin Sebor <msebor@redhat.com>
2658 PR other/83508
2659 * builtins.c (check_access): Avoid warning when the no-warning bit
2660 is set.
2662 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
2664 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
2665 * ira-color (allocno_hard_regs_compare): Likewise.
2667 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
2669 PR target/83013
2670 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
2671 Use .pushsection/.popsection.
2673 2018-01-14 Martin Sebor <msebor@redhat.com>
2675 PR c++/81327
2676 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
2678 2018-01-14 Jakub Jelinek <jakub@redhat.com>
2680 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
2681 entry from extra_headers.
2682 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
2683 extra_headers, make the list bitwise identical to the i?86-*-* one.
2685 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2687 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
2688 -mcmodel=large with -mindirect-branch=thunk,
2689 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
2690 -mfunction-return=thunk-extern.
2691 * doc/invoke.texi: Document -mcmodel=large is incompatible with
2692 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
2693 -mfunction-return=thunk and -mfunction-return=thunk-extern.
2695 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2697 * config/i386/i386.c (print_reg): Print the name of the full
2698 integer register without '%'.
2699 (ix86_print_operand): Handle 'V'.
2700 * doc/extend.texi: Document 'V' modifier.
2702 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2704 * config/i386/constraints.md (Bs): Disallow memory operand for
2705 -mindirect-branch-register.
2706 (Bw): Likewise.
2707 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2708 (GOT_memory_operand): Likewise.
2709 (call_insn_operand): Likewise.
2710 (sibcall_insn_operand): Likewise.
2711 (GOT32_symbol_operand): Likewise.
2712 * config/i386/i386.md (indirect_jump): Call convert_memory_address
2713 for -mindirect-branch-register.
2714 (tablejump): Likewise.
2715 (*sibcall_memory): Likewise.
2716 (*sibcall_value_memory): Likewise.
2717 Disallow peepholes of indirect call and jump via memory for
2718 -mindirect-branch-register.
2719 (*call_pop): Replace m with Bw.
2720 (*call_value_pop): Likewise.
2721 (*sibcall_pop_memory): Replace m with Bs.
2722 * config/i386/i386.opt (mindirect-branch-register): New option.
2723 * doc/invoke.texi: Document -mindirect-branch-register option.
2725 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2727 * config/i386/i386-protos.h (ix86_output_function_return): New.
2728 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
2729 set function_return_type.
2730 (indirect_thunk_name): Add ret_p to indicate thunk for function
2731 return.
2732 (output_indirect_thunk_function): Pass false to
2733 indirect_thunk_name.
2734 (ix86_output_indirect_branch_via_reg): Likewise.
2735 (ix86_output_indirect_branch_via_push): Likewise.
2736 (output_indirect_thunk_function): Create alias for function
2737 return thunk if regno < 0.
2738 (ix86_output_function_return): New function.
2739 (ix86_handle_fndecl_attribute): Handle function_return.
2740 (ix86_attribute_table): Add function_return.
2741 * config/i386/i386.h (machine_function): Add
2742 function_return_type.
2743 * config/i386/i386.md (simple_return_internal): Use
2744 ix86_output_function_return.
2745 (simple_return_internal_long): Likewise.
2746 * config/i386/i386.opt (mfunction-return=): New option.
2747 (indirect_branch): Mention -mfunction-return=.
2748 * doc/extend.texi: Document function_return function attribute.
2749 * doc/invoke.texi: Document -mfunction-return= option.
2751 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
2753 * config/i386/i386-opts.h (indirect_branch): New.
2754 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
2755 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
2756 with local indirect jump when converting indirect call and jump.
2757 (ix86_set_indirect_branch_type): New.
2758 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
2759 (indirectlabelno): New.
2760 (indirect_thunk_needed): Likewise.
2761 (indirect_thunk_bnd_needed): Likewise.
2762 (indirect_thunks_used): Likewise.
2763 (indirect_thunks_bnd_used): Likewise.
2764 (INDIRECT_LABEL): Likewise.
2765 (indirect_thunk_name): Likewise.
2766 (output_indirect_thunk): Likewise.
2767 (output_indirect_thunk_function): Likewise.
2768 (ix86_output_indirect_branch_via_reg): Likewise.
2769 (ix86_output_indirect_branch_via_push): Likewise.
2770 (ix86_output_indirect_branch): Likewise.
2771 (ix86_output_indirect_jmp): Likewise.
2772 (ix86_code_end): Call output_indirect_thunk_function if needed.
2773 (ix86_output_call_insn): Call ix86_output_indirect_branch if
2774 needed.
2775 (ix86_handle_fndecl_attribute): Handle indirect_branch.
2776 (ix86_attribute_table): Add indirect_branch.
2777 * config/i386/i386.h (machine_function): Add indirect_branch_type
2778 and has_local_indirect_jump.
2779 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
2780 to true.
2781 (tablejump): Likewise.
2782 (*indirect_jump): Use ix86_output_indirect_jmp.
2783 (*tablejump_1): Likewise.
2784 (simple_return_indirect_internal): Likewise.
2785 * config/i386/i386.opt (mindirect-branch=): New option.
2786 (indirect_branch): New.
2787 (keep): Likewise.
2788 (thunk): Likewise.
2789 (thunk-inline): Likewise.
2790 (thunk-extern): Likewise.
2791 * doc/extend.texi: Document indirect_branch function attribute.
2792 * doc/invoke.texi: Document -mindirect-branch= option.
2794 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2796 PR ipa/83051
2797 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2799 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2801 * ipa-inline.c (want_inline_small_function_p): Return false if
2802 inlining has already failed with CIF_FINAL_ERROR.
2803 (update_caller_keys): Call want_inline_small_function_p before
2804 can_inline_edge_p.
2805 (update_callee_keys): Likewise.
2807 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2809 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2810 New function.
2811 (rs6000_quadword_masked_address_p): Likewise.
2812 (quad_aligned_load_p): Likewise.
2813 (quad_aligned_store_p): Likewise.
2814 (const_load_sequence_p): Add comment to describe the outer-most loop.
2815 (mimic_memory_attributes_and_flags): New function.
2816 (rs6000_gen_stvx): Likewise.
2817 (replace_swapped_aligned_store): Likewise.
2818 (rs6000_gen_lvx): Likewise.
2819 (replace_swapped_aligned_load): Likewise.
2820 (replace_swapped_load_constant): Capitalize argument name in
2821 comment describing this function.
2822 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2823 and stores that access quad-word aligned addresses and replace
2824 with stvx or lvx instructions when appropriate.
2825 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2826 New function prototype.
2827 (rs6000_quadword_masked_address_p): Likewise.
2828 (rs6000_gen_lvx): Likewise.
2829 (rs6000_gen_stvx): Likewise.
2830 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2831 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2832 when memory address is aligned.
2833 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2834 this split to select lvx instruction when memory address is aligned.
2835 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2836 instruction when memory address is aligned.
2837 (*vsx_le_perm_load_v16qi): Likewise.
2838 (four unnamed splitters): Modify to select the stvx instruction
2839 when memory is aligned.
2841 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2843 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2844 which appears in the queue multiple times.
2846 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2847 Alan Hayward <alan.hayward@arm.com>
2848 David Sherwood <david.sherwood@arm.com>
2850 * tree-vectorizer.h (vec_lower_bound): New structure.
2851 (_loop_vec_info): Add check_nonzero and lower_bounds.
2852 (LOOP_VINFO_CHECK_NONZERO): New macro.
2853 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2854 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2855 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2856 fields. Make seg_len the distance travelled, not including the
2857 access size.
2858 (dr_direction_indicator): Declare.
2859 (dr_zero_step_indicator): Likewise.
2860 (dr_known_forward_stride_p): Likewise.
2861 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2862 tree-ssanames.h.
2863 (runtime_alias_check_p): Allow runtime alias checks with
2864 variable strides.
2865 (operator ==): Compare access_size and align.
2866 (prune_runtime_alias_test_list): Rework for new distinction between
2867 the access_size and seg_len.
2868 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2869 segment lengths.
2870 (get_segment_min_max): New function.
2871 (create_intersect_range_checks): Use it.
2872 (dr_step_indicator): New function.
2873 (dr_direction_indicator): Likewise.
2874 (dr_zero_step_indicator): Likewise.
2875 (dr_known_forward_stride_p): Likewise.
2876 * tree-loop-distribution.c (data_ref_segment_size): Return
2877 DR_STEP * (niters - 1).
2878 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2879 constructor.
2880 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2881 (vect_preserves_scalar_order_p): New function, split out from...
2882 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2883 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2884 (vect_vfa_access_size): New function.
2885 (vect_vfa_align): Likewise.
2886 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2887 (dump_lower_bound): New function.
2888 (vect_check_lower_bound): Likewise.
2889 (vect_small_gap_p): Likewise.
2890 (vectorizable_with_step_bound_p): Likewise.
2891 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2892 depencies if the vectorization factor is 1. Convert the checks
2893 for nonzero steps into checks on the bounds of DR_STEP. Try using
2894 a bunds check for variable steps if the minimum required step is
2895 relatively small. Update calls to the dr_with_seg_len
2896 constructor and to vect_compile_time_alias.
2897 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2898 function.
2899 (vect_loop_versioning): Call it.
2900 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2901 when retrying.
2902 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2904 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2905 Alan Hayward <alan.hayward@arm.com>
2906 David Sherwood <david.sherwood@arm.com>
2908 * doc/sourcebuild.texi (vect_scatter_store): Document.
2909 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2910 optabs.
2911 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2912 Document.
2913 * genopinit.c (main): Add supports_vec_scatter_store and
2914 supports_vec_scatter_store_cached to target_optabs.
2915 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2916 IFN_MASK_SCATTER_STORE.
2917 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2918 functions.
2919 * internal-fn.h (internal_store_fn_p): Declare.
2920 (internal_fn_stored_value_index): Likewise.
2921 * internal-fn.c (scatter_store_direct): New macro.
2922 (expand_scatter_store_optab_fn): New function.
2923 (direct_scatter_store_optab_supported_p): New macro.
2924 (internal_store_fn_p): New function.
2925 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2926 IFN_MASK_SCATTER_STORE.
2927 (internal_fn_mask_index): Likewise.
2928 (internal_fn_stored_value_index): New function.
2929 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2930 for scatter stores.
2931 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2932 * optabs-query.c (supports_vec_scatter_store_p): New function.
2933 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2934 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2935 true for scatter stores.
2936 (vect_gather_scatter_fn_p): Handle scatter stores too.
2937 (vect_check_gather_scatter): Consider using scatter stores if
2938 supports_vec_scatter_store_p.
2939 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2940 scatter stores too.
2941 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2942 internal_fn_stored_value_index.
2943 (check_load_store_masking): Handle scatter stores too.
2944 (vect_get_store_rhs): Make public.
2945 (vectorizable_call): Use internal_store_fn_p.
2946 (vectorizable_store): Handle scatter store internal functions.
2947 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2948 when deciding whether the end of the group has been reached.
2949 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2950 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2951 (mask_scatter_store<mode>): New insns.
2953 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2954 Alan Hayward <alan.hayward@arm.com>
2955 David Sherwood <david.sherwood@arm.com>
2957 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2958 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2959 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2960 function.
2961 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2962 Use vect_truncate_gather_scatter_offset if we can't treat the
2963 operation as a normal gather load or scatter store.
2964 (get_group_load_store_type): Take the gather_scatter_info
2965 as argument. Try using a gather load or scatter store for
2966 single-element groups.
2967 (get_load_store_type): Update calls to get_group_load_store_type
2968 and vect_use_strided_gather_scatters_p.
2970 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2971 Alan Hayward <alan.hayward@arm.com>
2972 David Sherwood <david.sherwood@arm.com>
2974 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2975 optional tree argument.
2976 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2977 null target hooks.
2978 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2979 but continue to use the current value as a fallback.
2980 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2981 to compare the updates.
2982 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2983 (get_load_store_type): Use it when handling a strided access.
2984 (vect_get_strided_load_store_ops): New function.
2985 (vect_get_data_ptr_increment): Likewise.
2986 (vectorizable_load): Handle strided gather loads. Always pass
2987 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2989 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2990 Alan Hayward <alan.hayward@arm.com>
2991 David Sherwood <david.sherwood@arm.com>
2993 * doc/md.texi (gather_load@var{m}): Document.
2994 (mask_gather_load@var{m}): Likewise.
2995 * genopinit.c (main): Add supports_vec_gather_load and
2996 supports_vec_gather_load_cached to target_optabs.
2997 * optabs-tree.c (init_tree_optimization_optabs): Use
2998 ggc_cleared_alloc to allocate target_optabs.
2999 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3000 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3001 functions.
3002 * internal-fn.h (internal_load_fn_p): Declare.
3003 (internal_gather_scatter_fn_p): Likewise.
3004 (internal_fn_mask_index): Likewise.
3005 (internal_gather_scatter_fn_supported_p): Likewise.
3006 * internal-fn.c (gather_load_direct): New macro.
3007 (expand_gather_load_optab_fn): New function.
3008 (direct_gather_load_optab_supported_p): New macro.
3009 (direct_internal_fn_optab): New function.
3010 (internal_load_fn_p): Likewise.
3011 (internal_gather_scatter_fn_p): Likewise.
3012 (internal_fn_mask_index): Likewise.
3013 (internal_gather_scatter_fn_supported_p): Likewise.
3014 * optabs-query.c (supports_at_least_one_mode_p): New function.
3015 (supports_vec_gather_load_p): Likewise.
3016 * optabs-query.h (supports_vec_gather_load_p): Declare.
3017 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3018 and memory_type field.
3019 (NUM_PATTERNS): Bump to 15.
3020 * tree-vect-data-refs.c: Include internal-fn.h.
3021 (vect_gather_scatter_fn_p): New function.
3022 (vect_describe_gather_scatter_call): Likewise.
3023 (vect_check_gather_scatter): Try using internal functions for
3024 gather loads. Recognize existing calls to a gather load function.
3025 (vect_analyze_data_refs): Consider using gather loads if
3026 supports_vec_gather_load_p.
3027 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3028 (vect_get_gather_scatter_offset_type): Likewise.
3029 (vect_convert_mask_for_vectype): Likewise.
3030 (vect_add_conversion_to_patterm): Likewise.
3031 (vect_try_gather_scatter_pattern): Likewise.
3032 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3033 (vect_vect_recog_func_ptrs): Add it.
3034 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3035 internal_fn_mask_index and internal_gather_scatter_fn_p.
3036 (check_load_store_masking): Take the gather_scatter_info as an
3037 argument and handle gather loads.
3038 (vect_get_gather_scatter_ops): New function.
3039 (vectorizable_call): Check internal_load_fn_p.
3040 (vectorizable_load): Likewise. Handle gather load internal
3041 functions.
3042 (vectorizable_store): Update call to check_load_store_masking.
3043 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3044 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3045 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3046 (aarch64_gather_scale_operand_d): New predicates.
3047 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3048 (mask_gather_load<mode>): New insns.
3050 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3051 Alan Hayward <alan.hayward@arm.com>
3052 David Sherwood <david.sherwood@arm.com>
3054 * optabs.def (fold_left_plus_optab): New optab.
3055 * doc/md.texi (fold_left_plus_@var{m}): Document.
3056 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3057 * internal-fn.c (fold_left_direct): Define.
3058 (expand_fold_left_optab_fn): Likewise.
3059 (direct_fold_left_optab_supported_p): Likewise.
3060 * fold-const-call.c (fold_const_fold_left): New function.
3061 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3062 * tree-parloops.c (valid_reduction_p): New function.
3063 (gather_scalar_reductions): Use it.
3064 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3065 (vect_finish_replace_stmt): Declare.
3066 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3067 (needs_fold_left_reduction_p): New function, split out from...
3068 (vect_is_simple_reduction): ...here. Accept reductions that
3069 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3070 (vect_force_simple_reduction): Also store the reduction type in
3071 the assignment's STMT_VINFO_REDUC_TYPE.
3072 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3073 (merge_with_identity): New function.
3074 (vect_expand_fold_left): Likewise.
3075 (vectorize_fold_left_reduction): Likewise.
3076 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3077 scalar phi in place for it. Check for target support and reject
3078 cases that would reassociate the operation. Defer the transform
3079 phase to vectorize_fold_left_reduction.
3080 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3081 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3082 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3084 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3086 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3087 call to ifc_temp_var.
3089 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3090 Alan Hayward <alan.hayward@arm.com>
3091 David Sherwood <david.sherwood@arm.com>
3093 * target.def (legitimize_address_displacement): Take the original
3094 offset as a poly_int.
3095 * targhooks.h (default_legitimize_address_displacement): Update
3096 accordingly.
3097 * targhooks.c (default_legitimize_address_displacement): Likewise.
3098 * doc/tm.texi: Regenerate.
3099 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3100 as an argument, moving assert of ad->disp == ad->disp_term to...
3101 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3102 Try calling targetm.legitimize_address_displacement before expanding
3103 the address rather than afterwards, and adjust for the new interface.
3104 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3105 Match the new hook interface. Handle SVE addresses.
3106 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3107 new hook interface.
3109 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3111 * Makefile.in (OBJS): Add early-remat.o.
3112 * target.def (select_early_remat_modes): New hook.
3113 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3114 * doc/tm.texi: Regenerate.
3115 * targhooks.h (default_select_early_remat_modes): Declare.
3116 * targhooks.c (default_select_early_remat_modes): New function.
3117 * timevar.def (TV_EARLY_REMAT): New timevar.
3118 * passes.def (pass_early_remat): New pass.
3119 * tree-pass.h (make_pass_early_remat): Declare.
3120 * early-remat.c: New file.
3121 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3122 function.
3123 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3125 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3126 Alan Hayward <alan.hayward@arm.com>
3127 David Sherwood <david.sherwood@arm.com>
3129 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3130 vfm1 with a bound_epilog parameter.
3131 (vect_do_peeling): Update calls accordingly, and move the prologue
3132 call earlier in the function. Treat the base bound_epilog as 0 for
3133 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3134 this base when peeling for gaps.
3135 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3136 with fully-masked loops.
3137 (vect_estimate_min_profitable_iters): Handle the single peeled
3138 iteration in that case.
3140 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3141 Alan Hayward <alan.hayward@arm.com>
3142 David Sherwood <david.sherwood@arm.com>
3144 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3145 single-element interleaving even if the size is not a power of 2.
3146 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3147 accesses for single-element interleaving if the group size is
3148 not a power of 2.
3150 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3151 Alan Hayward <alan.hayward@arm.com>
3152 David Sherwood <david.sherwood@arm.com>
3154 * doc/md.texi (fold_extract_last_@var{m}): Document.
3155 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3156 * optabs.def (fold_extract_last_optab): New optab.
3157 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3158 * internal-fn.c (fold_extract_direct): New macro.
3159 (expand_fold_extract_optab_fn): Likewise.
3160 (direct_fold_extract_optab_supported_p): Likewise.
3161 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3162 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3163 EXTRACT_LAST_REDUCTION.
3164 (get_initial_def_for_reduction): Do not create an initial vector
3165 for EXTRACT_LAST_REDUCTION reductions.
3166 (vectorizable_reduction): Leave the scalar phi in place for
3167 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3168 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3169 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3170 transform phase to vectorizable_condition.
3171 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3172 split out from...
3173 (vect_finish_stmt_generation): ...here.
3174 (vect_finish_replace_stmt): New function.
3175 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3176 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3177 pattern.
3178 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3180 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3181 Alan Hayward <alan.hayward@arm.com>
3182 David Sherwood <david.sherwood@arm.com>
3184 * doc/md.texi (extract_last_@var{m}): Document.
3185 * optabs.def (extract_last_optab): New optab.
3186 * internal-fn.def (EXTRACT_LAST): New internal function.
3187 * internal-fn.c (cond_unary_direct): New macro.
3188 (expand_cond_unary_optab_fn): Likewise.
3189 (direct_cond_unary_optab_supported_p): Likewise.
3190 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3191 loops using EXTRACT_LAST.
3192 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3193 (extract_last_<mode>): ...this optab.
3194 (vec_extract<mode><Vel>): Update accordingly.
3196 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3197 Alan Hayward <alan.hayward@arm.com>
3198 David Sherwood <david.sherwood@arm.com>
3200 * target.def (empty_mask_is_expensive): New hook.
3201 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3202 * doc/tm.texi: Regenerate.
3203 * targhooks.h (default_empty_mask_is_expensive): Declare.
3204 * targhooks.c (default_empty_mask_is_expensive): New function.
3205 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3206 if the target says that empty masks are expensive.
3207 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3208 New function.
3209 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3211 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3212 Alan Hayward <alan.hayward@arm.com>
3213 David Sherwood <david.sherwood@arm.com>
3215 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3216 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3217 (vect_use_loop_mask_for_alignment_p): New function.
3218 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3219 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3220 niters_skip argument. Make sure that the first niters_skip elements
3221 of the first iteration are inactive.
3222 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3223 Update call to vect_set_loop_masks_directly.
3224 (get_misalign_in_elems): New function, split out from...
3225 (vect_gen_prolog_loop_niters): ...here.
3226 (vect_update_init_of_dr): Take a code argument that specifies whether
3227 the adjustment should be added or subtracted.
3228 (vect_update_init_of_drs): Likewise.
3229 (vect_prepare_for_masked_peels): New function.
3230 (vect_do_peeling): Skip prologue peeling if we're using a mask
3231 instead. Update call to vect_update_inits_of_drs.
3232 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3233 mask_skip_niters.
3234 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3235 alignment. Do not include the number of peeled iterations in
3236 the minimum threshold in that case.
3237 (vectorizable_induction): Adjust the start value down by
3238 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3239 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3240 Take the number of skipped iterations into account when calculating
3241 the loop bounds.
3242 * tree-vect-stmts.c (vect_gen_while_not): New function.
3244 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3245 Alan Hayward <alan.hayward@arm.com>
3246 David Sherwood <david.sherwood@arm.com>
3248 * doc/sourcebuild.texi (vect_fully_masked): Document.
3249 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3250 default value to 0.
3251 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3252 split out from...
3253 (vect_analyze_loop_2): ...here. Don't check the vectorization
3254 factor against the number of loop iterations if the loop is
3255 fully-masked.
3257 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3258 Alan Hayward <alan.hayward@arm.com>
3259 David Sherwood <david.sherwood@arm.com>
3261 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3262 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3263 (dump_groups): Update accordingly.
3264 (iv_use::mem_type): New member variable.
3265 (address_p): New function.
3266 (record_use): Add a mem_type argument and initialize the new
3267 mem_type field.
3268 (record_group_use): Add a mem_type argument. Use address_p.
3269 Remove obsolete null checks of base_object. Update call to record_use.
3270 (find_interesting_uses_op): Update call to record_group_use.
3271 (find_interesting_uses_cond): Likewise.
3272 (find_interesting_uses_address): Likewise.
3273 (get_mem_type_for_internal_fn): New function.
3274 (find_address_like_use): Likewise.
3275 (find_interesting_uses_stmt): Try find_address_like_use before
3276 calling find_interesting_uses_op.
3277 (addr_offset_valid_p): Use the iv mem_type field as the type
3278 of the addressed memory.
3279 (add_autoinc_candidates): Likewise.
3280 (get_address_cost): Likewise.
3281 (split_small_address_groups_p): Use address_p.
3282 (split_address_groups): Likewise.
3283 (add_iv_candidate_for_use): Likewise.
3284 (autoinc_possible_for_pair): Likewise.
3285 (rewrite_groups): Likewise.
3286 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3287 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3288 (get_alias_ptr_type_for_ptr_address): New function.
3289 (rewrite_use_address): Rewrite address uses in calls that were
3290 identified by find_address_like_use.
3292 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3293 Alan Hayward <alan.hayward@arm.com>
3294 David Sherwood <david.sherwood@arm.com>
3296 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3297 TARGET_MEM_REFs.
3298 * gimple-expr.h (is_gimple_addressable: Likewise.
3299 * gimple-expr.c (is_gimple_address): Likewise.
3300 * internal-fn.c (expand_call_mem_ref): New function.
3301 (expand_mask_load_optab_fn): Use it.
3302 (expand_mask_store_optab_fn): Likewise.
3304 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3305 Alan Hayward <alan.hayward@arm.com>
3306 David Sherwood <david.sherwood@arm.com>
3308 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3309 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3310 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3311 (cond_umax@var{mode}): Document.
3312 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3313 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3314 (cond_umin_optab, cond_umax_optab): New optabs.
3315 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3316 (COND_IOR, COND_XOR): New internal functions.
3317 * internal-fn.h (get_conditional_internal_fn): Declare.
3318 * internal-fn.c (cond_binary_direct): New macro.
3319 (expand_cond_binary_optab_fn): Likewise.
3320 (direct_cond_binary_optab_supported_p): Likewise.
3321 (get_conditional_internal_fn): New function.
3322 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3323 Cope with reduction statements that are vectorized as calls rather
3324 than assignments.
3325 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3326 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3327 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3328 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3329 (UNSPEC_COND_EOR): New unspecs.
3330 (optab): Add mappings for them.
3331 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3332 (sve_int_op, sve_fp_op): New int attributes.
3334 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3335 Alan Hayward <alan.hayward@arm.com>
3336 David Sherwood <david.sherwood@arm.com>
3338 * optabs.def (while_ult_optab): New optab.
3339 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3340 * internal-fn.def (WHILE_ULT): New internal function.
3341 * internal-fn.h (direct_internal_fn_supported_p): New override
3342 that takes two types as argument.
3343 * internal-fn.c (while_direct): New macro.
3344 (expand_while_optab_fn): New function.
3345 (convert_optab_supported_p): Likewise.
3346 (direct_while_optab_supported_p): New macro.
3347 * wide-int.h (wi::udiv_ceil): New function.
3348 * tree-vectorizer.h (rgroup_masks): New structure.
3349 (vec_loop_masks): New typedef.
3350 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3351 and fully_masked_p.
3352 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3353 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3354 (vect_max_vf): New function.
3355 (slpeel_make_loop_iterate_ntimes): Delete.
3356 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3357 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3358 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3359 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3360 internal-fn.h, stor-layout.h and optabs-query.h.
3361 (vect_set_loop_mask): New function.
3362 (add_preheader_seq): Likewise.
3363 (add_header_seq): Likewise.
3364 (interleave_supported_p): Likewise.
3365 (vect_maybe_permute_loop_masks): Likewise.
3366 (vect_set_loop_masks_directly): Likewise.
3367 (vect_set_loop_condition_masked): Likewise.
3368 (vect_set_loop_condition_unmasked): New function, split out from
3369 slpeel_make_loop_iterate_ntimes.
3370 (slpeel_make_loop_iterate_ntimes): Rename to..
3371 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3372 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3373 (vect_do_peeling): Update call accordingly.
3374 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3375 loops.
3376 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3377 mask_compare_type, can_fully_mask_p and fully_masked_p.
3378 (release_vec_loop_masks): New function.
3379 (_loop_vec_info): Use it to free the loop masks.
3380 (can_produce_all_loop_masks_p): New function.
3381 (vect_get_max_nscalars_per_iter): Likewise.
3382 (vect_verify_full_masking): Likewise.
3383 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3384 retries, and free the mask rgroups before retrying. Check loop-wide
3385 reasons for disallowing fully-masked loops. Make the final decision
3386 about whether use a fully-masked loop or not.
3387 (vect_estimate_min_profitable_iters): Do not assume that peeling
3388 for the number of iterations will be needed for fully-masked loops.
3389 (vectorizable_reduction): Disable fully-masked loops.
3390 (vectorizable_live_operation): Likewise.
3391 (vect_halve_mask_nunits): New function.
3392 (vect_double_mask_nunits): Likewise.
3393 (vect_record_loop_mask): Likewise.
3394 (vect_get_loop_mask): Likewise.
3395 (vect_transform_loop): Handle the case in which the final loop
3396 iteration might handle a partial vector. Call vect_set_loop_condition
3397 instead of slpeel_make_loop_iterate_ntimes.
3398 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3399 (check_load_store_masking): New function.
3400 (prepare_load_store_mask): Likewise.
3401 (vectorizable_store): Handle fully-masked loops.
3402 (vectorizable_load): Likewise.
3403 (supportable_widening_operation): Use vect_halve_mask_nunits for
3404 booleans.
3405 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3406 (vect_gen_while): New function.
3407 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3408 (aarch64_uqdec<mode>): New insn.
3410 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3411 Alan Hayward <alan.hayward@arm.com>
3412 David Sherwood <david.sherwood@arm.com>
3414 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3415 (reduc_xor_scal_optab): New optabs.
3416 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3417 (reduc_xor_scal_@var{m}): Document.
3418 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3419 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3420 internal functions.
3421 * fold-const-call.c (fold_const_call): Handle them.
3422 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3423 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3424 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3425 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3426 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3427 (UNSPEC_XORV): New unspecs.
3428 (optab): Add entries for them.
3429 (BITWISEV): New int iterator.
3430 (bit_reduc_op): New int attributes.
3432 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3433 Alan Hayward <alan.hayward@arm.com>
3434 David Sherwood <david.sherwood@arm.com>
3436 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3437 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3438 * optabs.def (vec_shl_insert_optab): New optab.
3439 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3440 (duplicate_and_interleave): Likewise.
3441 * tree-vect-loop.c: Include internal-fn.h.
3442 (neutral_op_for_slp_reduction): New function, split out from
3443 get_initial_defs_for_reduction.
3444 (get_initial_def_for_reduction): Handle option 2 for variable-length
3445 vectors by loading the neutral value into a vector and then shifting
3446 the initial value into element 0.
3447 (get_initial_defs_for_reduction): Replace the code argument with
3448 the neutral value calculated by neutral_op_for_slp_reduction.
3449 Use gimple_build_vector for constant-length vectors.
3450 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3451 but the first group_size elements have a neutral value.
3452 Use duplicate_and_interleave otherwise.
3453 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3454 Update call to get_initial_defs_for_reduction. Handle SLP
3455 reductions for variable-length vectors by creating one vector
3456 result for each scalar result, with the elements associated
3457 with other scalar results stubbed out with the neutral value.
3458 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3459 Require IFN_VEC_SHL_INSERT for double reductions on
3460 variable-length vectors, or SLP reductions that have
3461 a neutral value. Require can_duplicate_and_interleave_p
3462 support for variable-length unchained SLP reductions if there
3463 is no neutral value, such as for MIN/MAX reductions. Also require
3464 the number of vector elements to be a multiple of the number of
3465 SLP statements when doing variable-length unchained SLP reductions.
3466 Update call to vect_create_epilog_for_reduction.
3467 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3468 and remove initial values.
3469 (duplicate_and_interleave): Make public.
3470 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3471 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3473 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3474 Alan Hayward <alan.hayward@arm.com>
3475 David Sherwood <david.sherwood@arm.com>
3477 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3478 (can_duplicate_and_interleave_p): New function.
3479 (vect_get_and_check_slp_defs): Take the vector of statements
3480 rather than just the current one. Remove excess parentheses.
3481 Restriction rejectinon of vect_constant_def and vect_external_def
3482 for variable-length vectors to boolean types, or types for which
3483 can_duplicate_and_interleave_p is false.
3484 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3485 (duplicate_and_interleave): New function.
3486 (vect_get_constant_vectors): Use gimple_build_vector for
3487 constant-length vectors and suitable variable-length constant
3488 vectors. Use duplicate_and_interleave for other variable-length
3489 vectors. Don't defer the update when inserting new statements.
3491 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3492 Alan Hayward <alan.hayward@arm.com>
3493 David Sherwood <david.sherwood@arm.com>
3495 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3496 min_profitable_iters doesn't go negative.
3498 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3499 Alan Hayward <alan.hayward@arm.com>
3500 David Sherwood <david.sherwood@arm.com>
3502 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3503 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3504 * optabs.def (vec_mask_load_lanes_optab): New optab.
3505 (vec_mask_store_lanes_optab): Likewise.
3506 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3507 (MASK_STORE_LANES): Likewise.
3508 * internal-fn.c (mask_load_lanes_direct): New macro.
3509 (mask_store_lanes_direct): Likewise.
3510 (expand_mask_load_optab_fn): Handle masked operations.
3511 (expand_mask_load_lanes_optab_fn): New macro.
3512 (expand_mask_store_optab_fn): Handle masked operations.
3513 (expand_mask_store_lanes_optab_fn): New macro.
3514 (direct_mask_load_lanes_optab_supported_p): Likewise.
3515 (direct_mask_store_lanes_optab_supported_p): Likewise.
3516 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3517 parameter.
3518 (vect_load_lanes_supported): Likewise.
3519 * tree-vect-data-refs.c (strip_conversion): New function.
3520 (can_group_stmts_p): Likewise.
3521 (vect_analyze_data_ref_accesses): Use it instead of checking
3522 for a pair of assignments.
3523 (vect_store_lanes_supported): Take a masked_p parameter.
3524 (vect_load_lanes_supported): Likewise.
3525 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3526 vect_store_lanes_supported and vect_load_lanes_supported.
3527 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3528 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3529 parameter. Don't allow gaps for masked accesses.
3530 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3531 and vect_load_lanes_supported.
3532 (get_load_store_type): Take a masked_p parameter and update
3533 call to get_group_load_store_type.
3534 (vectorizable_store): Update call to get_load_store_type.
3535 Handle IFN_MASK_STORE_LANES.
3536 (vectorizable_load): Update call to get_load_store_type.
3537 Handle IFN_MASK_LOAD_LANES.
3539 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3540 Alan Hayward <alan.hayward@arm.com>
3541 David Sherwood <david.sherwood@arm.com>
3543 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3544 modes for SVE.
3545 * config/aarch64/aarch64-protos.h
3546 (aarch64_sve_struct_memory_operand_p): Declare.
3547 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3548 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3549 (VPRED, vpred): Handle SVE structure modes.
3550 * config/aarch64/constraints.md (Utx): New constraint.
3551 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3552 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3553 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3554 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3555 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3556 structure modes. Split into pieces after RA.
3557 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3558 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3559 New patterns.
3560 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3561 SVE structure modes.
3562 (aarch64_classify_address): Likewise.
3563 (sizetochar): Move earlier in file.
3564 (aarch64_print_operand): Handle SVE register lists.
3565 (aarch64_array_mode): New function.
3566 (aarch64_sve_struct_memory_operand_p): Likewise.
3567 (TARGET_ARRAY_MODE): Redefine.
3569 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3570 Alan Hayward <alan.hayward@arm.com>
3571 David Sherwood <david.sherwood@arm.com>
3573 * target.def (array_mode): New target hook.
3574 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3575 * doc/tm.texi: Regenerate.
3576 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3577 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3578 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3579 targetm.array_mode.
3580 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3581 type sizes.
3583 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3584 Alan Hayward <alan.hayward@arm.com>
3585 David Sherwood <david.sherwood@arm.com>
3587 * fold-const.c (fold_binary_loc): Check the argument types
3588 rather than the result type when testing for a vector operation.
3590 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3592 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3593 * doc/tm.texi: Regenerate.
3595 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3596 Alan Hayward <alan.hayward@arm.com>
3597 David Sherwood <david.sherwood@arm.com>
3599 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3600 (sve): Document new AArch64 extension.
3601 * doc/md.texi (w): Extend the description of the AArch64
3602 constraint to include SVE vectors.
3603 (Upl, Upa): Document new AArch64 predicate constraints.
3604 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3605 enum.
3606 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3607 (msve-vector-bits=): New option.
3608 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3609 SVE when these are disabled.
3610 (sve): New extension.
3611 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3612 modes. Adjust their number of units based on aarch64_sve_vg.
3613 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3614 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3615 aarch64_addr_query_type.
3616 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3617 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3618 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3619 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3620 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3621 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3622 (aarch64_simd_imm_zero_p): Delete.
3623 (aarch64_check_zero_based_sve_index_immediate): Declare.
3624 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3625 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3626 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3627 (aarch64_sve_float_mul_immediate_p): Likewise.
3628 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3629 rather than an rtx.
3630 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
3631 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
3632 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
3633 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
3634 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
3635 (aarch64_regmode_natural_size): Likewise.
3636 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
3637 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
3638 left one place.
3639 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
3640 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
3641 for VG and the SVE predicate registers.
3642 (V_ALIASES): Add a "z"-prefixed alias.
3643 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
3644 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
3645 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
3646 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
3647 (REG_CLASS_NAMES): Add entries for them.
3648 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
3649 and the predicate registers.
3650 (aarch64_sve_vg): Declare.
3651 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
3652 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
3653 (REGMODE_NATURAL_SIZE): Define.
3654 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
3655 SVE macros.
3656 * config/aarch64/aarch64.c: Include cfgrtl.h.
3657 (simd_immediate_info): Add a constructor for series vectors,
3658 and an associated step field.
3659 (aarch64_sve_vg): New variable.
3660 (aarch64_dbx_register_number): Handle VG and the predicate registers.
3661 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
3662 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
3663 (VEC_ANY_DATA, VEC_STRUCT): New constants.
3664 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
3665 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
3666 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
3667 (aarch64_get_mask_mode): New functions.
3668 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
3669 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
3670 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
3671 predicate modes and predicate registers. Explicitly restrict
3672 GPRs to modes of 16 bytes or smaller. Only allow FP registers
3673 to store a vector mode if it is recognized by
3674 aarch64_classify_vector_mode.
3675 (aarch64_regmode_natural_size): New function.
3676 (aarch64_hard_regno_caller_save_mode): Return the original mode
3677 for predicates.
3678 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
3679 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
3680 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
3681 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
3682 functions.
3683 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
3684 does not overlap dest if the function is frame-related. Handle
3685 SVE constants.
3686 (aarch64_split_add_offset): New function.
3687 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
3688 them aarch64_add_offset.
3689 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
3690 and update call to aarch64_sub_sp.
3691 (aarch64_add_cfa_expression): New function.
3692 (aarch64_expand_prologue): Pass extra temporary registers to the
3693 functions above. Handle the case in which we need to emit new
3694 DW_CFA_expressions for registers that were originally saved
3695 relative to the stack pointer, but now have to be expressed
3696 relative to the frame pointer.
3697 (aarch64_output_mi_thunk): Pass extra temporary registers to the
3698 functions above.
3699 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
3700 IP0 and IP1 values for SVE frames.
3701 (aarch64_expand_vec_series): New function.
3702 (aarch64_expand_sve_widened_duplicate): Likewise.
3703 (aarch64_expand_sve_const_vector): Likewise.
3704 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
3705 Handle SVE constants. Use emit_move_insn to move a force_const_mem
3706 into the register, rather than emitting a SET directly.
3707 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
3708 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
3709 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
3710 (offset_9bit_signed_scaled_p): New functions.
3711 (aarch64_replicate_bitmask_imm): New function.
3712 (aarch64_bitmask_imm): Use it.
3713 (aarch64_cannot_force_const_mem): Reject expressions involving
3714 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
3715 (aarch64_classify_index): Handle SVE indices, by requiring
3716 a plain register index with a scale that matches the element size.
3717 (aarch64_classify_address): Handle SVE addresses. Assert that
3718 the mode of the address is VOIDmode or an integer mode.
3719 Update call to aarch64_classify_symbol.
3720 (aarch64_classify_symbolic_expression): Update call to
3721 aarch64_classify_symbol.
3722 (aarch64_const_vec_all_in_range_p): New function.
3723 (aarch64_print_vector_float_operand): Likewise.
3724 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
3725 "vN" for FP registers with SVE modes. Handle (const ...) vectors
3726 and the FP immediates 1.0 and 0.5.
3727 (aarch64_print_address_internal): Handle SVE addresses.
3728 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
3729 (aarch64_regno_regclass): Handle predicate registers.
3730 (aarch64_secondary_reload): Handle big-endian reloads of SVE
3731 data modes.
3732 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
3733 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
3734 (aarch64_convert_sve_vector_bits): New function.
3735 (aarch64_override_options): Use it to handle -msve-vector-bits=.
3736 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3737 rather than an rtx.
3738 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
3739 Handle SVE vector and predicate modes. Accept VL-based constants
3740 that need only one temporary register, and VL offsets that require
3741 no temporary registers.
3742 (aarch64_conditional_register_usage): Mark the predicate registers
3743 as fixed if SVE isn't available.
3744 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
3745 Return true for SVE vector and predicate modes.
3746 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
3747 rather than an unsigned int. Handle SVE modes.
3748 (aarch64_preferred_simd_mode): Update call accordingly. Handle
3749 SVE modes.
3750 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
3751 if SVE is enabled.
3752 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3753 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3754 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3755 (aarch64_sve_float_mul_immediate_p): New functions.
3756 (aarch64_sve_valid_immediate): New function.
3757 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
3758 Explicitly reject structure modes. Check for INDEX constants.
3759 Handle PTRUE and PFALSE constants.
3760 (aarch64_check_zero_based_sve_index_immediate): New function.
3761 (aarch64_simd_imm_zero_p): Delete.
3762 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
3763 vector modes. Accept constants in the range of CNT[BHWD].
3764 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
3765 ask for an Advanced SIMD mode.
3766 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
3767 (aarch64_simd_vector_alignment): Handle SVE predicates.
3768 (aarch64_vectorize_preferred_vector_alignment): New function.
3769 (aarch64_simd_vector_alignment_reachable): Use it instead of
3770 the vector size.
3771 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
3772 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
3773 functions.
3774 (MAX_VECT_LEN): Delete.
3775 (expand_vec_perm_d): Add a vec_flags field.
3776 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
3777 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
3778 (aarch64_evpc_ext): Don't apply a big-endian lane correction
3779 for SVE modes.
3780 (aarch64_evpc_rev): Rename to...
3781 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
3782 (aarch64_evpc_rev_global): New function.
3783 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
3784 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
3785 MAX_VECT_LEN.
3786 (aarch64_evpc_sve_tbl): New function.
3787 (aarch64_expand_vec_perm_const_1): Update after rename of
3788 aarch64_evpc_rev. Handle SVE permutes too, trying
3789 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
3790 than aarch64_evpc_tbl.
3791 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
3792 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
3793 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
3794 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
3795 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3796 (aarch64_expand_sve_vcond): New functions.
3797 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3798 of aarch64_vector_mode_p.
3799 (aarch64_dwarf_poly_indeterminate_value): New function.
3800 (aarch64_compute_pressure_classes): Likewise.
3801 (aarch64_can_change_mode_class): Likewise.
3802 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3803 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3804 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3805 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3806 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3807 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3808 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3809 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3810 constraints.
3811 (Dn, Dl, Dr): Accept const as well as const_vector.
3812 (Dz): Likewise. Compare against CONST0_RTX.
3813 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3814 of "vector" where appropriate.
3815 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3816 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3817 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3818 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3819 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3820 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3821 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3822 (v_int_equiv): Extend to SVE modes.
3823 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3824 mode attributes.
3825 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3826 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3827 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3828 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3829 (SVE_COND_FP_CMP): New int iterators.
3830 (perm_hilo): Handle the new unpack unspecs.
3831 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3832 attributes.
3833 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3834 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3835 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3836 (aarch64_equality_operator, aarch64_constant_vector_operand)
3837 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3838 (aarch64_sve_nonimmediate_operand): Likewise.
3839 (aarch64_sve_general_operand): Likewise.
3840 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3841 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3842 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3843 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3844 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3845 (aarch64_sve_float_arith_immediate): Likewise.
3846 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3847 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3848 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3849 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3850 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3851 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3852 (aarch64_sve_float_arith_operand): Likewise.
3853 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3854 (aarch64_sve_float_mul_operand): Likewise.
3855 (aarch64_sve_vec_perm_operand): Likewise.
3856 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3857 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3858 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3859 as well as const_vector.
3860 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3861 in file. Use CONST0_RTX and CONSTM1_RTX.
3862 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3863 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3864 Use aarch64_simd_imm_zero.
3865 * config/aarch64/aarch64-sve.md: New file.
3866 * config/aarch64/aarch64.md: Include it.
3867 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3868 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3869 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3870 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3871 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3872 (sve): New attribute.
3873 (enabled): Disable instructions with the sve attribute unless
3874 TARGET_SVE.
3875 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3876 aarch64_expand_mov_immediate.
3877 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3878 CNT[BHSD] immediates.
3879 (movti): Split CONST_POLY_INT moves into two halves.
3880 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3881 Split additions that need a temporary here if the destination
3882 is the stack pointer.
3883 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3884 (*add<mode>3_poly_1): New instruction.
3885 (set_clobber_cc): New expander.
3887 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3889 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3890 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3891 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3892 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3893 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3894 Change innermode from fixed_mode_size to machine_mode.
3895 (simplify_subreg): Update call accordingly. Handle a constant-sized
3896 subreg of a variable-length CONST_VECTOR.
3898 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3899 Alan Hayward <alan.hayward@arm.com>
3900 David Sherwood <david.sherwood@arm.com>
3902 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3903 (add_offset_to_base): New function, split out from...
3904 (create_mem_ref): ...here. When handling a scale other than 1,
3905 check first whether the address is valid without the offset.
3906 Add it into the base if so, leaving the index and scale as-is.
3908 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3910 PR c++/83778
3911 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3912 fold_for_warn before checking if arg2 is INTEGER_CST.
3914 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3916 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3917 (store_multiple_operation): Delete.
3918 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3919 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3920 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3921 guarded by TARGET_STRING.
3922 (rs6000_output_load_multiple): Delete.
3923 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3924 OPTION_MASK_STRING / TARGET_STRING handling.
3925 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3926 (const rs6000_opt_masks) <"string">: Change mask to 0.
3927 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3928 (MASK_STRING): Delete.
3929 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3930 parts. Simplify.
3931 (load_multiple): Delete.
3932 (*ldmsi8): Delete.
3933 (*ldmsi7): Delete.
3934 (*ldmsi6): Delete.
3935 (*ldmsi5): Delete.
3936 (*ldmsi4): Delete.
3937 (*ldmsi3): Delete.
3938 (store_multiple): Delete.
3939 (*stmsi8): Delete.
3940 (*stmsi7): Delete.
3941 (*stmsi6): Delete.
3942 (*stmsi5): Delete.
3943 (*stmsi4): Delete.
3944 (*stmsi3): Delete.
3945 (movmemsi_8reg): Delete.
3946 (corresponding unnamed define_insn): Delete.
3947 (movmemsi_6reg): Delete.
3948 (corresponding unnamed define_insn): Delete.
3949 (movmemsi_4reg): Delete.
3950 (corresponding unnamed define_insn): Delete.
3951 (movmemsi_2reg): Delete.
3952 (corresponding unnamed define_insn): Delete.
3953 (movmemsi_1reg): Delete.
3954 (corresponding unnamed define_insn): Delete.
3955 * config/rs6000/rs6000.opt (mno-string): New.
3956 (mstring): Replace by deprecation warning stub.
3957 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3959 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3961 * regrename.c (regrename_do_replace): If replacing the same
3962 reg multiple times, try to reuse last created gen_raw_REG.
3964 PR debug/81155
3965 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3966 main to workaround a bug in GDB.
3968 2018-01-12 Tom de Vries <tom@codesourcery.com>
3970 PR target/83737
3971 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3973 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3975 PR rtl-optimization/80481
3976 * ira-color.c (get_cap_member): New function.
3977 (allocnos_conflict_by_live_ranges_p): Use it.
3978 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3979 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3981 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3983 PR target/83628
3984 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3985 (*saddl_se_1): Ditto.
3986 (*ssubsi_1): Ditto.
3987 (*ssubl_se_1): Ditto.
3989 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3991 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3992 rather than wi::to_widest for DR_INITs.
3993 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3994 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3995 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3996 INTEGER_CSTs.
3997 (vect_analyze_group_access_1): Note that here.
3999 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4001 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4002 polynomial type sizes.
4004 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4006 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4007 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4008 (gimple_add_tmp_var): Likewise.
4010 2018-01-12 Martin Liska <mliska@suse.cz>
4012 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4013 (gimple_alloc_sizes): Likewise.
4014 (dump_gimple_statistics): Use PRIu64 in printf format.
4015 * gimple.h: Change uint64_t to int.
4017 2018-01-12 Martin Liska <mliska@suse.cz>
4019 * tree-core.h: Use uint64_t instead of int.
4020 * tree.c (tree_node_counts): Likewise.
4021 (tree_node_sizes): Likewise.
4022 (dump_tree_statistics): Use PRIu64 in printf format.
4024 2018-01-12 Martin Liska <mliska@suse.cz>
4026 * Makefile.in: As qsort_chk is implemented in vec.c, add
4027 vec.o to linkage of gencfn-macros.
4028 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4029 passing the info to record_node_allocation_statistics.
4030 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4031 and pass the info.
4032 * ggc-common.c (struct ggc_usage): Add operator== and use
4033 it in operator< and compare function.
4034 * mem-stats.h (struct mem_usage): Likewise.
4035 * vec.c (struct vec_usage): Remove operator< and compare
4036 function. Can be simply inherited.
4038 2018-01-12 Martin Jambor <mjambor@suse.cz>
4040 PR target/81616
4041 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4042 * tree-ssa-math-opts.c: Include domwalk.h.
4043 (convert_mult_to_fma_1): New function.
4044 (fma_transformation_info): New type.
4045 (fma_deferring_state): Likewise.
4046 (cancel_fma_deferring): New function.
4047 (result_of_phi): Likewise.
4048 (last_fma_candidate_feeds_initial_phi): Likewise.
4049 (convert_mult_to_fma): Added deferring logic, split actual
4050 transformation to convert_mult_to_fma_1.
4051 (math_opts_dom_walker): New type.
4052 (math_opts_dom_walker::after_dom_children): New method, body moved
4053 here from pass_optimize_widening_mul::execute, added deferring logic
4054 bits.
4055 (pass_optimize_widening_mul::execute): Moved most of code to
4056 math_opts_dom_walker::after_dom_children.
4057 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4058 * config/i386/i386.c (ix86_option_override_internal): Added
4059 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4061 2018-01-12 Richard Biener <rguenther@suse.de>
4063 PR debug/83157
4064 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4065 inline instance vars.
4067 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4069 PR target/81819
4070 * config/rx/rx.c (rx_is_restricted_memory_address):
4071 Handle SUBREG case.
4073 2018-01-12 Richard Biener <rguenther@suse.de>
4075 PR tree-optimization/80846
4076 * target.def (split_reduction): New target hook.
4077 * targhooks.c (default_split_reduction): New function.
4078 * targhooks.h (default_split_reduction): Declare.
4079 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4080 target requests first reduce vectors by combining low and high
4081 parts.
4082 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4083 (get_vectype_for_scalar_type_and_size): Export.
4084 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4085 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4086 * doc/tm.texi: Regenerate.
4087 * config/i386/i386.c (ix86_split_reduction): Implement
4088 TARGET_VECTORIZE_SPLIT_REDUCTION.
4090 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4092 PR target/83368
4093 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4094 in PIC mode except for TARGET_VXWORKS_RTP.
4095 * config/sparc/sparc.c: Include cfgrtl.h.
4096 (TARGET_INIT_PIC_REG): Define.
4097 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4098 (sparc_pic_register_p): New predicate.
4099 (sparc_legitimate_address_p): Use it.
4100 (sparc_legitimize_pic_address): Likewise.
4101 (sparc_delegitimize_address): Likewise.
4102 (sparc_mode_dependent_address_p): Likewise.
4103 (gen_load_pcrel_sym): Remove 4th parameter.
4104 (load_got_register): Adjust call to above. Remove obsolete stuff.
4105 (sparc_expand_prologue): Do not call load_got_register here.
4106 (sparc_flat_expand_prologue): Likewise.
4107 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4108 (sparc_use_pseudo_pic_reg): New function.
4109 (sparc_init_pic_reg): Likewise.
4110 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4111 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4113 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4115 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4116 Add item for branch_cost.
4118 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4120 PR rtl-optimization/83565
4121 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4122 not extend the result to a larger mode for rotate operations.
4123 (num_sign_bit_copies1): Likewise.
4125 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4127 PR target/40411
4128 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4129 -symbolic.
4130 Use values-Xc.o for -pedantic.
4131 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4133 2018-01-12 Martin Liska <mliska@suse.cz>
4135 PR ipa/83054
4136 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4137 New function.
4138 (possible_polymorphic_call_targets): Use it.
4139 (ipa_devirt): Likewise.
4141 2018-01-12 Martin Liska <mliska@suse.cz>
4143 * profile-count.h (enum profile_quality): Use 0 as invalid
4144 enum value of profile_quality.
4146 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4148 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4149 -mext-string options.
4151 2018-01-12 Richard Biener <rguenther@suse.de>
4153 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4154 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4155 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4156 Likewise.
4157 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4159 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4161 * configure.ac (--with-long-double-format): Add support for the
4162 configuration option to change the default long double format on
4163 PowerPC systems.
4164 * config.gcc (powerpc*-linux*-*): Likewise.
4165 * configure: Regenerate.
4166 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4167 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4168 used without modification.
4170 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4172 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4173 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4174 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4175 MISC_BUILTIN_SPEC_BARRIER.
4176 (rs6000_init_builtins): Likewise.
4177 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4178 enum value.
4179 (speculation_barrier): New define_insn.
4180 * doc/extend.texi: Document __builtin_speculation_barrier.
4182 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4184 PR target/83203
4185 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4186 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4187 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4188 iterators.
4189 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4190 integral modes instead of "ss" and "sd".
4191 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4192 vectors with 32-bit and 64-bit elements.
4193 (vecdupssescalarmodesuffix): New mode attribute.
4194 (vec_dup<mode>): Use it.
4196 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4198 PR target/83330
4199 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4200 frame if argument is passed on stack.
4202 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4204 PR target/82682
4205 * ree.c (combine_reaching_defs): Optimize also
4206 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4207 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4209 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4211 PR middle-end/83189
4212 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4214 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4216 PR middle-end/83718
4217 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4218 after they are computed.
4220 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4222 PR tree-optimization/83695
4223 * gimple-loop-linterchange.cc
4224 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4225 reset cached scev information after interchange.
4226 (pass_linterchange::execute): Remove call to scev_reset_htab.
4228 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4230 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4231 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4232 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4233 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4234 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4235 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4236 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4237 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4238 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4239 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4240 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4241 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4242 (V_lane_reg): Likewise.
4243 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4244 New define_expand.
4245 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4246 (vfmal_lane_low<mode>_intrinsic,
4247 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4248 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4249 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4250 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4251 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4252 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4254 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4256 * config/arm/arm-cpus.in (fp16fml): New feature.
4257 (ALL_SIMD): Add fp16fml.
4258 (armv8.2-a): Add fp16fml as an option.
4259 (armv8.3-a): Likewise.
4260 (armv8.4-a): Add fp16fml as part of fp16.
4261 * config/arm/arm.h (TARGET_FP16FML): Define.
4262 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4263 when appropriate.
4264 * config/arm/arm-modes.def (V2HF): Define.
4265 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4266 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4267 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4268 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4269 vfmsl_low, vfmsl_high): New set of builtins.
4270 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4271 (vfml_op): New code attribute.
4272 (VFMLHALVES): New int iterator.
4273 (VFML, VFMLSEL): New mode attributes.
4274 (V_reg): Define mapping for V2HF.
4275 (V_hi, V_lo): New mode attributes.
4276 (VF_constraint): Likewise.
4277 (vfml_half, vfml_half_selector): New int attributes.
4278 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4279 define_expand.
4280 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4281 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4282 New define_insn.
4283 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4284 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4285 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4286 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4287 documentation.
4288 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4289 Document new effective target and option set.
4291 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4293 * config/arm/arm-cpus.in (armv8_4): New feature.
4294 (ARMv8_4a): New fgroup.
4295 (armv8.4-a): New arch.
4296 * config/arm/arm-tables.opt: Regenerate.
4297 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4298 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4299 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4300 Add matching rules for -march=armv8.4-a and extensions.
4301 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4303 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4305 PR target/81821
4306 * config/rx/rx.md (BW): New mode attribute.
4307 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4309 2018-01-11 Richard Biener <rguenther@suse.de>
4311 PR tree-optimization/83435
4312 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4313 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4314 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4316 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4317 Alan Hayward <alan.hayward@arm.com>
4318 David Sherwood <david.sherwood@arm.com>
4320 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4321 field.
4322 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4323 (aarch64_print_address_internal): Use it to check for a zero offset.
4325 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4326 Alan Hayward <alan.hayward@arm.com>
4327 David Sherwood <david.sherwood@arm.com>
4329 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4330 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4331 Return a poly_int64 rather than a HOST_WIDE_INT.
4332 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4333 rather than a HOST_WIDE_INT.
4334 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4335 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4336 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4337 final_offset from HOST_WIDE_INT to poly_int64.
4338 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4339 to_constant when getting the number of units in an Advanced SIMD
4340 mode.
4341 (aarch64_builtin_vectorized_function): Check for a constant number
4342 of units.
4343 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4344 GET_MODE_SIZE.
4345 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4346 attribute instead of GET_MODE_NUNITS.
4347 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4348 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4349 GET_MODE_SIZE for fixed-size registers.
4350 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4351 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4352 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4353 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4354 (aarch64_print_operand, aarch64_print_address_internal)
4355 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4356 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4357 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4358 Handle polynomial GET_MODE_SIZE.
4359 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4360 wider than SImode without modification.
4361 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4362 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4363 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4364 passing and returning SVE modes.
4365 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4366 rather than GEN_INT.
4367 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4368 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4369 (aarch64_allocate_and_probe_stack_space): Likewise.
4370 (aarch64_layout_frame): Cope with polynomial offsets.
4371 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4372 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4373 polynomial offsets.
4374 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4375 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4376 poly_int64 rather than a HOST_WIDE_INT.
4377 (aarch64_get_separate_components, aarch64_process_components)
4378 (aarch64_expand_prologue, aarch64_expand_epilogue)
4379 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4380 (aarch64_anchor_offset): New function, split out from...
4381 (aarch64_legitimize_address): ...here.
4382 (aarch64_builtin_vectorization_cost): Handle polynomial
4383 TYPE_VECTOR_SUBPARTS.
4384 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4385 GET_MODE_NUNITS.
4386 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4387 number of elements from the PARALLEL rather than the mode.
4388 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4389 rather than GET_MODE_BITSIZE.
4390 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4391 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4392 (aarch64_expand_vec_perm_const_1): Handle polynomial
4393 d->perm.length () and d->perm elements.
4394 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4395 Apply to_constant to d->perm elements.
4396 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4397 polynomial CONST_VECTOR_NUNITS.
4398 (aarch64_move_pointer): Take amount as a poly_int64 rather
4399 than an int.
4400 (aarch64_progress_pointer): Avoid temporary variable.
4401 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4402 the mode attribute instead of GET_MODE.
4404 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4405 Alan Hayward <alan.hayward@arm.com>
4406 David Sherwood <david.sherwood@arm.com>
4408 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4409 x exists before using it.
4410 (aarch64_add_constant_internal): Rename to...
4411 (aarch64_add_offset_1): ...this. Replace regnum with separate
4412 src and dest rtxes. Handle the case in which they're different,
4413 including when the offset is zero. Replace scratchreg with an rtx.
4414 Use 2 additions if there is no spare register into which we can
4415 move a 16-bit constant.
4416 (aarch64_add_constant): Delete.
4417 (aarch64_add_offset): Replace reg with separate src and dest
4418 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4419 Use aarch64_add_offset_1.
4420 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4421 an rtx rather than an int. Take the delta as a poly_int64
4422 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4423 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4424 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4425 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4426 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4427 and aarch64_add_sp.
4428 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4429 aarch64_add_constant.
4431 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4433 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4434 Use scalar_float_mode.
4436 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4438 * config/aarch64/aarch64-simd.md
4439 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4440 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4441 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4442 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4443 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4444 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4445 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4446 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4447 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4448 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4450 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4452 PR target/83514
4453 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4454 targ_options->x_arm_arch_string is non NULL.
4456 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4458 * config/aarch64/aarch64.h
4459 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4461 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4463 PR target/82096
4464 * expmed.c (emit_store_flag_force): Swap if const op0
4465 and change VOIDmode to mode of op0.
4467 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4469 PR rtl-optimization/83761
4470 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4471 than bytes to mode_for_size.
4473 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4475 PR middle-end/83189
4476 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4477 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4478 profile.
4480 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4482 PR middle-end/83575
4483 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4484 when in layout mode.
4485 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4486 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4487 partition fixup.
4489 2018-01-10 Michael Collison <michael.collison@arm.com>
4491 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4492 * config/aarch64/aarch64-option-extension.def: Add
4493 AARCH64_OPT_EXTENSION of 'fp16fml'.
4494 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4495 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4496 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4497 * config/aarch64/constraints.md (Ui7): New constraint.
4498 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4499 (VFMLA_SEL_W): Ditto.
4500 (f16quad): Ditto.
4501 (f16mac1): Ditto.
4502 (VFMLA16_LOW): New int iterator.
4503 (VFMLA16_HIGH): Ditto.
4504 (UNSPEC_FMLAL): New unspec.
4505 (UNSPEC_FMLSL): Ditto.
4506 (UNSPEC_FMLAL2): Ditto.
4507 (UNSPEC_FMLSL2): Ditto.
4508 (f16mac): New code attribute.
4509 * config/aarch64/aarch64-simd-builtins.def
4510 (aarch64_fmlal_lowv2sf): Ditto.
4511 (aarch64_fmlsl_lowv2sf): Ditto.
4512 (aarch64_fmlalq_lowv4sf): Ditto.
4513 (aarch64_fmlslq_lowv4sf): Ditto.
4514 (aarch64_fmlal_highv2sf): Ditto.
4515 (aarch64_fmlsl_highv2sf): Ditto.
4516 (aarch64_fmlalq_highv4sf): Ditto.
4517 (aarch64_fmlslq_highv4sf): Ditto.
4518 (aarch64_fmlal_lane_lowv2sf): Ditto.
4519 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4520 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4521 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4522 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4523 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4524 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4525 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4526 (aarch64_fmlal_lane_highv2sf): Ditto.
4527 (aarch64_fmlsl_lane_highv2sf): Ditto.
4528 (aarch64_fmlal_laneq_highv2sf): Ditto.
4529 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4530 (aarch64_fmlalq_lane_highv4sf): Ditto.
4531 (aarch64_fmlsl_lane_highv4sf): Ditto.
4532 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4533 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4534 * config/aarch64/aarch64-simd.md:
4535 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4536 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4537 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4538 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4539 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4540 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4541 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4542 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4543 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4544 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4545 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4546 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4547 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4548 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4549 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4550 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4551 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4552 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4553 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4554 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4555 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4556 (vfmlsl_low_u32): Ditto.
4557 (vfmlalq_low_u32): Ditto.
4558 (vfmlslq_low_u32): Ditto.
4559 (vfmlal_high_u32): Ditto.
4560 (vfmlsl_high_u32): Ditto.
4561 (vfmlalq_high_u32): Ditto.
4562 (vfmlslq_high_u32): Ditto.
4563 (vfmlal_lane_low_u32): Ditto.
4564 (vfmlsl_lane_low_u32): Ditto.
4565 (vfmlal_laneq_low_u32): Ditto.
4566 (vfmlsl_laneq_low_u32): Ditto.
4567 (vfmlalq_lane_low_u32): Ditto.
4568 (vfmlslq_lane_low_u32): Ditto.
4569 (vfmlalq_laneq_low_u32): Ditto.
4570 (vfmlslq_laneq_low_u32): Ditto.
4571 (vfmlal_lane_high_u32): Ditto.
4572 (vfmlsl_lane_high_u32): Ditto.
4573 (vfmlal_laneq_high_u32): Ditto.
4574 (vfmlsl_laneq_high_u32): Ditto.
4575 (vfmlalq_lane_high_u32): Ditto.
4576 (vfmlslq_lane_high_u32): Ditto.
4577 (vfmlalq_laneq_high_u32): Ditto.
4578 (vfmlslq_laneq_high_u32): Ditto.
4579 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4580 (AARCH64_FL_FOR_ARCH8_4): New.
4581 (AARCH64_ISA_F16FML): New ISA flag.
4582 (TARGET_F16FML): New feature flag for fp16fml.
4583 (doc/invoke.texi): Document new fp16fml option.
4585 2018-01-10 Michael Collison <michael.collison@arm.com>
4587 * config/aarch64/aarch64-builtins.c:
4588 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4589 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4590 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4591 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4592 (AARCH64_ISA_SHA3): New ISA flag.
4593 (TARGET_SHA3): New feature flag for sha3.
4594 * config/aarch64/iterators.md (sha512_op): New int attribute.
4595 (CRYPTO_SHA512): New int iterator.
4596 (UNSPEC_SHA512H): New unspec.
4597 (UNSPEC_SHA512H2): Ditto.
4598 (UNSPEC_SHA512SU0): Ditto.
4599 (UNSPEC_SHA512SU1): Ditto.
4600 * config/aarch64/aarch64-simd-builtins.def
4601 (aarch64_crypto_sha512hqv2di): New builtin.
4602 (aarch64_crypto_sha512h2qv2di): Ditto.
4603 (aarch64_crypto_sha512su0qv2di): Ditto.
4604 (aarch64_crypto_sha512su1qv2di): Ditto.
4605 (aarch64_eor3qv8hi): Ditto.
4606 (aarch64_rax1qv2di): Ditto.
4607 (aarch64_xarqv2di): Ditto.
4608 (aarch64_bcaxqv8hi): Ditto.
4609 * config/aarch64/aarch64-simd.md:
4610 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4611 (aarch64_crypto_sha512su0qv2di): Ditto.
4612 (aarch64_crypto_sha512su1qv2di): Ditto.
4613 (aarch64_eor3qv8hi): Ditto.
4614 (aarch64_rax1qv2di): Ditto.
4615 (aarch64_xarqv2di): Ditto.
4616 (aarch64_bcaxqv8hi): Ditto.
4617 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4618 (vsha512h2q_u64): Ditto.
4619 (vsha512su0q_u64): Ditto.
4620 (vsha512su1q_u64): Ditto.
4621 (veor3q_u16): Ditto.
4622 (vrax1q_u64): Ditto.
4623 (vxarq_u64): Ditto.
4624 (vbcaxq_u16): Ditto.
4625 * config/arm/types.md (crypto_sha512): New type attribute.
4626 (crypto_sha3): Ditto.
4627 (doc/invoke.texi): Document new sha3 option.
4629 2018-01-10 Michael Collison <michael.collison@arm.com>
4631 * config/aarch64/aarch64-builtins.c:
4632 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
4633 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4634 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
4635 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
4636 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
4637 (AARCH64_ISA_SM4): New ISA flag.
4638 (TARGET_SM4): New feature flag for sm4.
4639 * config/aarch64/aarch64-simd-builtins.def
4640 (aarch64_sm3ss1qv4si): Ditto.
4641 (aarch64_sm3tt1aq4si): Ditto.
4642 (aarch64_sm3tt1bq4si): Ditto.
4643 (aarch64_sm3tt2aq4si): Ditto.
4644 (aarch64_sm3tt2bq4si): Ditto.
4645 (aarch64_sm3partw1qv4si): Ditto.
4646 (aarch64_sm3partw2qv4si): Ditto.
4647 (aarch64_sm4eqv4si): Ditto.
4648 (aarch64_sm4ekeyqv4si): Ditto.
4649 * config/aarch64/aarch64-simd.md:
4650 (aarch64_sm3ss1qv4si): Ditto.
4651 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
4652 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
4653 (aarch64_sm4eqv4si): Ditto.
4654 (aarch64_sm4ekeyqv4si): Ditto.
4655 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
4656 (sm3part_op): Ditto.
4657 (CRYPTO_SM3TT): Ditto.
4658 (CRYPTO_SM3PART): Ditto.
4659 (UNSPEC_SM3SS1): New unspec.
4660 (UNSPEC_SM3TT1A): Ditto.
4661 (UNSPEC_SM3TT1B): Ditto.
4662 (UNSPEC_SM3TT2A): Ditto.
4663 (UNSPEC_SM3TT2B): Ditto.
4664 (UNSPEC_SM3PARTW1): Ditto.
4665 (UNSPEC_SM3PARTW2): Ditto.
4666 (UNSPEC_SM4E): Ditto.
4667 (UNSPEC_SM4EKEY): Ditto.
4668 * config/aarch64/constraints.md (Ui2): New constraint.
4669 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
4670 * config/arm/types.md (crypto_sm3): New type attribute.
4671 (crypto_sm4): Ditto.
4672 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
4673 (vsm3tt1aq_u32): Ditto.
4674 (vsm3tt1bq_u32): Ditto.
4675 (vsm3tt2aq_u32): Ditto.
4676 (vsm3tt2bq_u32): Ditto.
4677 (vsm3partw1q_u32): Ditto.
4678 (vsm3partw2q_u32): Ditto.
4679 (vsm4eq_u32): Ditto.
4680 (vsm4ekeyq_u32): Ditto.
4681 (doc/invoke.texi): Document new sm4 option.
4683 2018-01-10 Michael Collison <michael.collison@arm.com>
4685 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
4686 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
4687 (AARCH64_FL_FOR_ARCH8_4): New.
4688 (AARCH64_FL_V8_4): New flag.
4689 (doc/invoke.texi): Document new armv8.4-a option.
4691 2018-01-10 Michael Collison <michael.collison@arm.com>
4693 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4694 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
4695 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
4696 * config/aarch64/aarch64-option-extension.def: Add
4697 AARCH64_OPT_EXTENSION of 'sha2'.
4698 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
4699 (crypto): Disable sha2 and aes if crypto disabled.
4700 (crypto): Enable aes and sha2 if enabled.
4701 (simd): Disable sha2 and aes if simd disabled.
4702 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
4703 New flags.
4704 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
4705 (TARGET_SHA2): New feature flag for sha2.
4706 (TARGET_AES): New feature flag for aes.
4707 * config/aarch64/aarch64-simd.md:
4708 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
4709 conditional on TARGET_AES.
4710 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
4711 (aarch64_crypto_sha1hsi): Make pattern conditional
4712 on TARGET_SHA2.
4713 (aarch64_crypto_sha1hv4si): Ditto.
4714 (aarch64_be_crypto_sha1hv4si): Ditto.
4715 (aarch64_crypto_sha1su1v4si): Ditto.
4716 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
4717 (aarch64_crypto_sha1su0v4si): Ditto.
4718 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
4719 (aarch64_crypto_sha256su0v4si): Ditto.
4720 (aarch64_crypto_sha256su1v4si): Ditto.
4721 (doc/invoke.texi): Document new aes and sha2 options.
4723 2018-01-10 Martin Sebor <msebor@redhat.com>
4725 PR tree-optimization/83781
4726 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
4727 as string arrays.
4729 2018-01-11 Martin Sebor <msebor@gmail.com>
4730 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4732 PR tree-optimization/83501
4733 PR tree-optimization/81703
4735 * tree-ssa-strlen.c (get_string_cst): Rename...
4736 (get_string_len): ...to this. Handle global constants.
4737 (handle_char_store): Adjust.
4739 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
4740 Jim Wilson <jimw@sifive.com>
4742 * config/riscv/riscv-protos.h (riscv_output_return): New.
4743 * config/riscv/riscv.c (struct machine_function): New naked_p field.
4744 (riscv_attribute_table, riscv_output_return),
4745 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
4746 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
4747 (riscv_compute_frame_info): Only compute frame->mask if not a naked
4748 function.
4749 (riscv_expand_prologue): Add early return for naked function.
4750 (riscv_expand_epilogue): Likewise.
4751 (riscv_function_ok_for_sibcall): Return false for naked function.
4752 (riscv_set_current_function): New.
4753 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
4754 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
4755 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
4756 * doc/extend.texi (RISC-V Function Attributes): New.
4758 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
4760 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
4761 check for 128-bit long double before checking TCmode.
4762 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
4763 128-bit long doubles before checking TFmode or TCmode.
4764 (FLOAT128_IBM_P): Likewise.
4766 2018-01-10 Martin Sebor <msebor@redhat.com>
4768 PR tree-optimization/83671
4769 * builtins.c (c_strlen): Unconditionally return zero for the empty
4770 string.
4771 Use -Warray-bounds for warnings.
4772 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
4773 for non-constant array indices with COMPONENT_REF, arrays of
4774 arrays, and pointers to arrays.
4775 (gimple_fold_builtin_strlen): Determine and set length range for
4776 non-constant character arrays.
4778 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
4780 PR middle-end/81897
4781 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
4782 empty blocks.
4784 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
4786 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
4788 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4790 PR target/83399
4791 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
4792 VECTOR_MEM_ALTIVEC_OR_VSX_P.
4793 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
4794 indexed_or_indirect_operand predicate.
4795 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4796 (*vsx_le_perm_load_v8hi): Likewise.
4797 (*vsx_le_perm_load_v16qi): Likewise.
4798 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4799 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4800 (*vsx_le_perm_store_v8hi): Likewise.
4801 (*vsx_le_perm_store_v16qi): Likewise.
4802 (eight unnamed splitters): Likewise.
4804 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4806 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4807 * config/rs6000/emmintrin.h: Likewise.
4808 * config/rs6000/mmintrin.h: Likewise.
4809 * config/rs6000/xmmintrin.h: Likewise.
4811 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4813 PR c++/43486
4814 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4815 "public_flag".
4816 * tree.c (tree_nop_conversion): Return true for location wrapper
4817 nodes.
4818 (maybe_wrap_with_location): New function.
4819 (selftest::check_strip_nops): New function.
4820 (selftest::test_location_wrappers): New function.
4821 (selftest::tree_c_tests): Call it.
4822 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4823 (maybe_wrap_with_location): New decl.
4824 (EXPR_LOCATION_WRAPPER_P): New macro.
4825 (location_wrapper_p): New inline function.
4826 (tree_strip_any_location_wrapper): New inline function.
4828 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4830 PR target/83735
4831 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4832 stack_realign_offset for the largest alignment of stack slot
4833 actually used.
4834 (ix86_find_max_used_stack_alignment): New function.
4835 (ix86_finalize_stack_frame_flags): Use it. Set
4836 max_used_stack_alignment if we don't realign stack.
4837 * config/i386/i386.h (machine_function): Add
4838 max_used_stack_alignment.
4840 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4842 * config/arm/arm.opt (-mbranch-cost): New option.
4843 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4844 account.
4846 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4848 PR target/83629
4849 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4850 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4852 2018-01-10 Richard Biener <rguenther@suse.de>
4854 PR debug/83765
4855 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4856 early out so it also covers the case where we have a non-NULL
4857 origin.
4859 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4861 PR tree-optimization/83753
4862 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4863 for non-strided grouped accesses if the number of elements is 1.
4865 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4867 PR target/81616
4868 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4869 * i386.h (TARGET_USE_GATHER): Define.
4870 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4872 2018-01-10 Martin Liska <mliska@suse.cz>
4874 PR bootstrap/82831
4875 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4876 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4877 partitioning.
4878 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4879 CLEANUP_NO_PARTITIONING is not set.
4881 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4883 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4884 for vectors, as a partial revert of r254296.
4885 * rtl.h (const_vec_p): Delete.
4886 (const_vec_duplicate_p): Don't test for vector CONSTs.
4887 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4888 * expmed.c (make_tree): Likewise.
4890 Revert:
4891 * common.md (E, F): Use CONSTANT_P instead of checking for
4892 CONST_VECTOR.
4893 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4894 checking for CONST_VECTOR.
4896 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4898 PR middle-end/83575
4899 * predict.c (force_edge_cold): Handle in more sane way edges
4900 with no prediction.
4902 2018-01-09 Carl Love <cel@us.ibm.com>
4904 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4905 V4SI, V4SF types.
4906 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4907 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4908 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4909 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4910 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4911 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4912 * config/rs6000/rs6000-protos.h: Add extern defition for
4913 rs6000_generate_float2_double_code.
4914 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4915 function.
4916 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4917 (float2_v2df): Add define_expand.
4919 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4921 PR target/83628
4922 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4923 op_mode in the force_to_mode call.
4925 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4927 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4928 instead of checking each element individually.
4929 (aarch64_evpc_uzp): Likewise.
4930 (aarch64_evpc_zip): Likewise.
4931 (aarch64_evpc_ext): Likewise.
4932 (aarch64_evpc_rev): Likewise.
4933 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4934 instead of checking each element individually. Return true without
4935 generating rtl if
4936 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4937 whether all selected elements come from the same input, instead of
4938 checking each element individually. Remove calls to gen_rtx_REG,
4939 start_sequence and end_sequence and instead assert that no rtl is
4940 generated.
4942 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4944 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4945 order of HIGH and CONST checks.
4947 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4949 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4950 if the destination isn't an SSA_NAME.
4952 2018-01-09 Richard Biener <rguenther@suse.de>
4954 PR tree-optimization/83668
4955 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4956 move prologue...
4957 (canonicalize_loop_form): ... here, renamed from ...
4958 (canonicalize_loop_closed_ssa_form): ... this and amended to
4959 swap successor edges for loop exit blocks to make us use
4960 the RPO order we need for initial schedule generation.
4962 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4964 PR tree-optimization/64811
4965 * match.pd: When optimizing comparisons with Inf, avoid
4966 introducing or losing exceptions from comparisons with NaN.
4968 2018-01-09 Martin Liska <mliska@suse.cz>
4970 PR sanitizer/82517
4971 * asan.c (shadow_mem_size): Add gcc_assert.
4973 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4975 Don't save registers in main().
4977 PR target/83738
4978 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4979 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4980 * config/avr/avr.c (avr_set_current_function): Don't error if
4981 naked, OS_task or OS_main are specified at the same time.
4982 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4983 OS_main.
4984 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4985 attribute.
4986 * common/config/avr/avr-common.c (avr_option_optimization_table):
4987 Switch on -mmain-is-OS_task for optimizing compilations.
4989 2018-01-09 Richard Biener <rguenther@suse.de>
4991 PR tree-optimization/83572
4992 * graphite.c: Include cfganal.h.
4993 (graphite_transform_loops): Connect infinite loops to exit
4994 and remove fake edges at the end.
4996 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4998 * ipa-inline.c (edge_badness): Revert accidental checkin.
5000 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5002 PR ipa/80763
5003 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5004 symbols; not inline clones.
5006 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5008 PR target/83507
5009 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5010 hard registers. Formatting fixes.
5012 PR preprocessor/83722
5013 * gcc.c (try_generate_repro): Pass
5014 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5015 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5016 do_report_bug.
5018 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5019 Kito Cheng <kito.cheng@gmail.com>
5021 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5022 (riscv_leaf_function_p): Delete.
5023 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5025 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5027 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5028 function.
5029 (do_ifelse): New function.
5030 (do_isel): New function.
5031 (do_sub3): New function.
5032 (do_add3): New function.
5033 (do_load_mask_compare): New function.
5034 (do_overlap_load_compare): New function.
5035 (expand_compare_loop): New function.
5036 (expand_block_compare): Call expand_compare_loop() when appropriate.
5037 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5038 option description.
5039 (-mblock-compare-inline-loop-limit): New option.
5041 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5043 PR target/83677
5044 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5045 Reverse order of second and third operands in first alternative.
5046 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5047 of first and second elements in UNSPEC_VPERMR vector.
5048 (altivec_expand_vec_perm_le): Likewise.
5050 2018-01-08 Jeff Law <law@redhat.com>
5052 PR rtl-optimizatin/81308
5053 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5054 (process_switch): If group_case_labels makes a change, then set
5055 cfg_altered.
5056 (pass_convert_switch::execute): If a switch is converted, then
5057 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5059 PR rtl-optimization/81308
5060 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5061 splitting insns.
5063 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5065 PR target/83663 - Revert r255946
5066 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5067 generation for cases where splatting a value is not useful.
5068 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5069 across a vec_duplicate and a paradoxical subreg forming a vector
5070 mode to a vec_concat.
5072 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5074 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5075 -march=armv8.3-a variants.
5076 * config/arm/t-multilib: Likewise.
5077 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5079 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5081 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5082 to generate rtl.
5083 (cceq_ior_compare_complement): Give it a name so I can use it, and
5084 change boolean_or_operator predicate to boolean_operator so it can
5085 be used to generate a crand.
5086 (eqne): New code iterator.
5087 (bd/bd_neg): New code_attrs.
5088 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5089 a single define_insn.
5090 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5091 decrement (bdnzt/bdnzf/bdzt/bdzf).
5092 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5093 with the new names of the branch decrement patterns, and added the
5094 names of the branch decrement conditional patterns.
5096 2018-01-08 Richard Biener <rguenther@suse.de>
5098 PR tree-optimization/83563
5099 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5100 cache.
5102 2018-01-08 Richard Biener <rguenther@suse.de>
5104 PR middle-end/83713
5105 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5107 2018-01-08 Richard Biener <rguenther@suse.de>
5109 PR tree-optimization/83685
5110 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5111 references to abnormals.
5113 2018-01-08 Richard Biener <rguenther@suse.de>
5115 PR lto/83719
5116 * dwarf2out.c (output_indirect_strings): Handle empty
5117 skeleton_debug_str_hash.
5118 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5120 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5122 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5123 (emit_store_direct): Likewise.
5124 (arc_trampoline_adjust_address): Likewise.
5125 (arc_asm_trampoline_template): New function.
5126 (arc_initialize_trampoline): Use asm_trampoline_template.
5127 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5128 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5129 * config/arc/arc.md (flush_icache): Delete pattern.
5131 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5133 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5134 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5135 munaligned-access.
5137 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5139 PR target/83681
5140 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5141 by not USED_FOR_TARGET.
5142 (make_pass_resolve_sw_modes): Likewise.
5144 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5146 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5147 USED_FOR_TARGET.
5149 2018-01-08 Richard Biener <rguenther@suse.de>
5151 PR middle-end/83580
5152 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5154 2018-01-08 Richard Biener <rguenther@suse.de>
5156 PR middle-end/83517
5157 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5159 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5161 PR middle-end/81897
5162 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5163 basic blocks with a small number of successors.
5164 (convert_control_dep_chain_into_preds): Improve handling of
5165 forwarder blocks.
5166 (dump_predicates): Split apart into...
5167 (dump_pred_chain): ...here...
5168 (dump_pred_info): ...and here.
5169 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5170 (can_chain_union_be_invalidated_p): Improve check for invalidation
5171 of paths.
5172 (uninit_uses_cannot_happen): Avoid unnecessary if
5173 convert_control_dep_chain_into_preds yielded nothing.
5175 2018-01-06 Martin Sebor <msebor@redhat.com>
5177 PR tree-optimization/83640
5178 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5179 subtracting negative offset from size.
5180 (builtin_access::overlap): Adjust offset bounds of the access to fall
5181 within the size of the object if possible.
5183 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5185 PR rtl-optimization/83699
5186 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5187 extract_bit_field_as_subreg to cases in which the extracted
5188 value is also a vector.
5190 * lra-constraints.c (process_alt_operands): Test for the equivalence
5191 substitutions when detecting a possible reload cycle.
5193 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5195 PR debug/83480
5196 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5197 by default if flag_selective_schedling{,2}. Formatting fixes.
5199 PR rtl-optimization/83682
5200 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5201 if it has non-VECTOR_MODE element mode.
5202 (vec_duplicate_p): Likewise.
5204 PR middle-end/83694
5205 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5206 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5208 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5210 PR target/83604
5211 * config/i386/i386-builtin.def
5212 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5213 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5214 Require also OPTION_MASK_ISA_AVX512F in addition to
5215 OPTION_MASK_ISA_GFNI.
5216 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5217 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5218 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5219 to OPTION_MASK_ISA_GFNI.
5220 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5221 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5222 OPTION_MASK_ISA_AVX512BW.
5223 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5224 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5225 addition to OPTION_MASK_ISA_GFNI.
5226 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5227 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5228 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5229 to OPTION_MASK_ISA_GFNI.
5230 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5231 a requirement for all ISAs rather than any of them with a few
5232 exceptions.
5233 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5234 processing.
5235 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5236 bitmasks to be enabled with 3 exceptions, instead of requiring any
5237 enabled ISA with lots of exceptions.
5238 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5239 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5240 Change avx512bw in isa attribute to avx512f.
5241 * config/i386/sgxintrin.h: Add license boilerplate.
5242 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5243 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5244 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5245 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5246 defined.
5247 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5248 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5249 temporarily sse2 rather than sse if not enabled already.
5251 PR target/83604
5252 * config/i386/sse.md (VI248_VLBW): Rename to ...
5253 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5254 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5255 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5256 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5257 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5258 mode iterator instead of VI248_VLBW.
5260 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5262 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5263 (record_modified): Skip clobbers; add debug output.
5264 (param_change_prob): Use sreal frequencies.
5266 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5268 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5269 punt for user-aligned variables.
5271 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5273 * tree-chrec.c (chrec_contains_symbols): Return true for
5274 POLY_INT_CST.
5276 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5278 PR target/82439
5279 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5280 of (x|y) == x for BICS pattern.
5282 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5284 PR tree-optimization/83605
5285 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5286 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5287 can throw.
5289 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5291 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5292 * config/epiphany/rtems.h: New file.
5294 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5295 Uros Bizjak <ubizjak@gmail.com>
5297 PR target/83554
5298 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5299 QIreg_operand instead of register_operand predicate.
5300 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5301 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5302 comments instead of -fmitigate[-_]rop.
5304 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5306 PR bootstrap/81926
5307 * cgraphunit.c (symbol_table::compile): Switch to text_section
5308 before calling assembly_start debug hook.
5309 * run-rtl-passes.c (run_rtl_passes): Likewise.
5310 Include output.h.
5312 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5314 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5315 range_int_cst_p rather than !symbolic_range_p before calling
5316 extract_range_from_multiplicative_op_1.
5318 2018-01-04 Jeff Law <law@redhat.com>
5320 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5321 redundant test in assertion.
5323 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5325 * doc/rtl.texi: Document machine_mode wrapper classes.
5327 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5329 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5330 using tree_to_uhwi.
5332 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5334 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5335 the VEC_PERM_EXPR fold to fail.
5337 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5339 PR debug/83585
5340 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5341 to switched_sections.
5343 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5345 PR target/83680
5346 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5347 test for d.testing.
5349 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5351 PR target/83387
5352 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5353 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5355 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5357 PR debug/83666
5358 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5359 is BLKmode and bitpos not zero or mode change is needed.
5361 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5363 PR target/83675
5364 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5365 TARGET_VIS2.
5367 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5369 PR target/83628
5370 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5371 instead of MULT rtx. Update all corresponding splitters.
5372 (*saddl_se): Ditto.
5373 (*ssub<modesuffix>): Ditto.
5374 (*ssubl_se): Ditto.
5375 (*cmp_sadd_di): Update split patterns.
5376 (*cmp_sadd_si): Ditto.
5377 (*cmp_sadd_sidi): Ditto.
5378 (*cmp_ssub_di): Ditto.
5379 (*cmp_ssub_si): Ditto.
5380 (*cmp_ssub_sidi): Ditto.
5381 * config/alpha/predicates.md (const23_operand): New predicate.
5382 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5383 Look for ASHIFT, not MULT inner operand.
5384 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5386 2018-01-04 Martin Liska <mliska@suse.cz>
5388 PR gcov-profile/83669
5389 * gcov.c (output_intermediate_file): Add version to intermediate
5390 gcov file.
5391 * doc/gcov.texi: Document new field 'version' in intermediate
5392 file format. Fix location of '-k' option of gcov command.
5394 2018-01-04 Martin Liska <mliska@suse.cz>
5396 PR ipa/82352
5397 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5399 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5401 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5403 2018-01-03 Martin Sebor <msebor@redhat.com>
5405 PR tree-optimization/83655
5406 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5407 checking calls with invalid arguments.
5409 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5411 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5412 (vectorizable_mask_load_store): Delete.
5413 (vectorizable_call): Return false for masked loads and stores.
5414 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5415 instead of gimple_assign_rhs1.
5416 (vectorizable_load): Handle IFN_MASK_LOAD.
5417 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5419 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5421 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5422 split out from..,
5423 (vectorizable_mask_load_store): ...here.
5424 (vectorizable_load): ...and here.
5426 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5428 * tree-vect-stmts.c (vect_build_all_ones_mask)
5429 (vect_build_zero_merge_argument): New functions, split out from...
5430 (vectorizable_load): ...here.
5432 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5434 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5435 split out from...
5436 (vectorizable_mask_load_store): ...here.
5437 (vectorizable_store): ...and here.
5439 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5441 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5442 split out from...
5443 (vectorizable_mask_load_store): ...here.
5445 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5447 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5448 (vect_model_store_cost): Take a vec_load_store_type instead of a
5449 vect_def_type.
5450 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5451 (vect_model_store_cost): Take a vec_load_store_type instead of a
5452 vect_def_type.
5453 (vectorizable_mask_load_store): Update accordingly.
5454 (vectorizable_store): Likewise.
5455 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5457 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5459 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5460 IFN_MASK_LOAD calls here rather than...
5461 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5463 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5464 Alan Hayward <alan.hayward@arm.com>
5465 David Sherwood <david.sherwood@arm.com>
5467 * expmed.c (extract_bit_field_1): For vector extracts,
5468 fall back to extract_bit_field_as_subreg if vec_extract
5469 isn't available.
5471 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5472 Alan Hayward <alan.hayward@arm.com>
5473 David Sherwood <david.sherwood@arm.com>
5475 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5476 they are variable or constant sized.
5477 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5478 slots for constant-sized data.
5480 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5481 Alan Hayward <alan.hayward@arm.com>
5482 David Sherwood <david.sherwood@arm.com>
5484 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5485 handling COND_EXPRs with boolean comparisons, try to find a better
5486 basis for the mask type than the boolean itself.
5488 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5490 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5491 is calculated and how it can be overridden.
5492 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5493 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5494 if defined.
5495 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5496 if nonzero.
5498 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5499 Alan Hayward <alan.hayward@arm.com>
5500 David Sherwood <david.sherwood@arm.com>
5502 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5503 Remove the mode argument.
5504 (aarch64_simd_valid_immediate): Remove the mode and inverse
5505 arguments.
5506 * config/aarch64/iterators.md (bitsize): New iterator.
5507 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5508 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5509 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5510 aarch64_simd_valid_immediate.
5511 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5512 (aarch64_reg_or_bic_imm): Likewise.
5513 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5514 with an insn_type enum and msl with a modifier_type enum.
5515 Replace element_width with a scalar_mode. Change the shift
5516 to unsigned int. Add constructors for scalar_float_mode and
5517 scalar_int_mode elements.
5518 (aarch64_vect_float_const_representable_p): Delete.
5519 (aarch64_can_const_movi_rtx_p)
5520 (aarch64_simd_scalar_immediate_valid_for_move)
5521 (aarch64_simd_make_constant): Update call to
5522 aarch64_simd_valid_immediate.
5523 (aarch64_advsimd_valid_immediate_hs): New function.
5524 (aarch64_advsimd_valid_immediate): Likewise.
5525 (aarch64_simd_valid_immediate): Remove mode and inverse
5526 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5527 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5528 and aarch64_float_const_representable_p on the result.
5529 (aarch64_output_simd_mov_immediate): Remove mode argument.
5530 Update call to aarch64_simd_valid_immediate and use of
5531 simd_immediate_info.
5532 (aarch64_output_scalar_simd_mov_immediate): Update call
5533 accordingly.
5535 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5536 Alan Hayward <alan.hayward@arm.com>
5537 David Sherwood <david.sherwood@arm.com>
5539 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5540 (mode_nunits): Likewise CONST_MODE_NUNITS.
5541 * machmode.def (ADJUST_NUNITS): Document.
5542 * genmodes.c (mode_data::need_nunits_adj): New field.
5543 (blank_mode): Update accordingly.
5544 (adj_nunits): New variable.
5545 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5546 parameter.
5547 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5548 listed in adj_nunits.
5549 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5550 listed in adj_nunits. Don't emit case statements for such modes.
5551 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5552 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5553 nothing if adj_nunits is nonnull.
5554 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5555 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5556 (emit_mode_fbit): Update use of print_maybe_const_decl.
5557 (emit_move_size): Likewise. Treat the array as non-const
5558 if adj_nunits.
5559 (emit_mode_adjustments): Handle adj_nunits.
5561 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5563 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5564 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5565 (VECTOR_MODES): Use it.
5566 (make_vector_modes): Take the prefix as an argument.
5568 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5569 Alan Hayward <alan.hayward@arm.com>
5570 David Sherwood <david.sherwood@arm.com>
5572 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5573 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5574 for MODE_VECTOR_BOOL.
5575 * machmode.def (VECTOR_BOOL_MODE): Document.
5576 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5577 (make_vector_bool_mode): New function.
5578 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5579 MODE_VECTOR_BOOL.
5580 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5581 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5582 Likewise.
5583 * stor-layout.c (int_mode_for_mode): Likewise.
5584 * tree.c (build_vector_type_for_mode): Likewise.
5585 * varasm.c (output_constant_pool_2): Likewise.
5586 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5587 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5588 for MODE_VECTOR_BOOL.
5589 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5590 of mode class checks.
5591 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5592 instead of a list of mode class checks.
5593 (expand_vector_scalar_condition): Likewise.
5594 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5596 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5597 Alan Hayward <alan.hayward@arm.com>
5598 David Sherwood <david.sherwood@arm.com>
5600 * machmode.h (mode_size): Change from unsigned short to
5601 poly_uint16_pod.
5602 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5603 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5604 or if measurement_type is not polynomial.
5605 (fixed_size_mode::includes_p): Check for constant-sized modes.
5606 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5607 return a poly_uint16 rather than an unsigned short.
5608 (emit_mode_size): Change the type of mode_size from unsigned short
5609 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5610 (emit_mode_adjustments): Cope with polynomial vector sizes.
5611 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5612 for GET_MODE_SIZE.
5613 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5614 for GET_MODE_SIZE.
5615 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5616 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5617 * caller-save.c (setup_save_areas): Likewise.
5618 (replace_reg_with_saved_mem): Likewise.
5619 * calls.c (emit_library_call_value_1): Likewise.
5620 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5621 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5622 (gen_lowpart_for_combine): Likewise.
5623 * convert.c (convert_to_integer_1): Likewise.
5624 * cse.c (equiv_constant, cse_insn): Likewise.
5625 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5626 (cselib_subst_to_values): Likewise.
5627 * dce.c (word_dce_process_block): Likewise.
5628 * df-problems.c (df_word_lr_mark_ref): Likewise.
5629 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5630 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
5631 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
5632 (rtl_for_decl_location): Likewise.
5633 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
5634 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
5635 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
5636 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
5637 (expand_expr_real_1): Likewise.
5638 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
5639 (pad_below): Likewise.
5640 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5641 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
5642 * ira.c (get_subreg_tracking_sizes): Likewise.
5643 * ira-build.c (ira_create_allocno_objects): Likewise.
5644 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
5645 (ira_sort_regnos_for_alter_reg): Likewise.
5646 * ira-costs.c (record_operand_costs): Likewise.
5647 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
5648 (resolve_simple_move): Likewise.
5649 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
5650 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
5651 (lra_constraints): Likewise.
5652 (CONST_POOL_OK_P): Reject variable-sized modes.
5653 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
5654 (add_pseudo_to_slot, lra_spill): Likewise.
5655 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5656 * optabs-query.c (get_best_extraction_insn): Likewise.
5657 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5658 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
5659 (expand_mult_highpart, valid_multiword_target_p): Likewise.
5660 * recog.c (offsettable_address_addr_space_p): Likewise.
5661 * regcprop.c (maybe_mode_change): Likewise.
5662 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
5663 * regrename.c (build_def_use): Likewise.
5664 * regstat.c (dump_reg_info): Likewise.
5665 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
5666 (find_reloads, find_reloads_subreg_address): Likewise.
5667 * reload1.c (eliminate_regs_1): Likewise.
5668 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
5669 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
5670 (simplify_binary_operation_1, simplify_subreg): Likewise.
5671 * targhooks.c (default_function_arg_padding): Likewise.
5672 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
5673 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
5674 (verify_gimple_assign_ternary): Likewise.
5675 * tree-inline.c (estimate_move_cost): Likewise.
5676 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5677 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
5678 (get_address_cost_ainc): Likewise.
5679 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
5680 (vect_supportable_dr_alignment): Likewise.
5681 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5682 (vectorizable_reduction): Likewise.
5683 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
5684 (vectorizable_operation, vectorizable_load): Likewise.
5685 * tree.c (build_same_sized_truth_vector_type): Likewise.
5686 * valtrack.c (cleanup_auto_inc_dec): Likewise.
5687 * var-tracking.c (emit_note_insn_var_location): Likewise.
5688 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
5689 (ADDR_VEC_ALIGN): Likewise.
5691 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5692 Alan Hayward <alan.hayward@arm.com>
5693 David Sherwood <david.sherwood@arm.com>
5695 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
5696 unsigned short.
5697 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5698 or if measurement_type is polynomial.
5699 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
5700 * combine.c (make_extraction): Likewise.
5701 * dse.c (find_shift_sequence): Likewise.
5702 * dwarf2out.c (mem_loc_descriptor): Likewise.
5703 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
5704 (extract_bit_field, extract_low_bits): Likewise.
5705 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
5706 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
5707 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
5708 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
5709 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
5710 * reload.c (find_reloads): Likewise.
5711 * reload1.c (alter_reg): Likewise.
5712 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
5713 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
5714 * tree-if-conv.c (predicate_mem_writes): Likewise.
5715 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
5716 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
5717 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
5718 * valtrack.c (dead_debug_insert_temp): Likewise.
5719 * varasm.c (mergeable_constant_section): Likewise.
5720 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
5722 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5723 Alan Hayward <alan.hayward@arm.com>
5724 David Sherwood <david.sherwood@arm.com>
5726 * expr.c (expand_assignment): Cope with polynomial mode sizes
5727 when assigning to a CONCAT.
5729 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5730 Alan Hayward <alan.hayward@arm.com>
5731 David Sherwood <david.sherwood@arm.com>
5733 * machmode.h (mode_precision): Change from unsigned short to
5734 poly_uint16_pod.
5735 (mode_to_precision): Return a poly_uint16 rather than an unsigned
5736 short.
5737 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
5738 or if measurement_type is not polynomial.
5739 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
5740 in which the mode is already known to be a scalar_int_mode.
5741 * genmodes.c (emit_mode_precision): Change the type of mode_precision
5742 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
5743 initializer.
5744 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5745 for GET_MODE_PRECISION.
5746 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5747 for GET_MODE_PRECISION.
5748 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
5749 as polynomial.
5750 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
5751 (expand_field_assignment, make_extraction): Likewise.
5752 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
5753 (get_last_value): Likewise.
5754 * convert.c (convert_to_integer_1): Likewise.
5755 * cse.c (cse_insn): Likewise.
5756 * expr.c (expand_expr_real_1): Likewise.
5757 * lra-constraints.c (simplify_operand_subreg): Likewise.
5758 * optabs-query.c (can_atomic_load_p): Likewise.
5759 * optabs.c (expand_atomic_load): Likewise.
5760 (expand_atomic_store): Likewise.
5761 * ree.c (combine_reaching_defs): Likewise.
5762 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
5763 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
5764 * tree.h (type_has_mode_precision_p): Likewise.
5765 * ubsan.c (instrument_si_overflow): Likewise.
5767 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5768 Alan Hayward <alan.hayward@arm.com>
5769 David Sherwood <david.sherwood@arm.com>
5771 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
5772 polynomial numbers of units.
5773 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
5774 (valid_vector_subparts_p): New function.
5775 (build_vector_type): Remove temporary shim and take the number
5776 of units as a poly_uint64 rather than an int.
5777 (build_opaque_vector_type): Take the number of units as a
5778 poly_uint64 rather than an int.
5779 * tree.c (build_vector_from_ctor): Handle polynomial
5780 TYPE_VECTOR_SUBPARTS.
5781 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
5782 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
5783 (build_vector_from_val): If the number of units is variable,
5784 use build_vec_duplicate_cst for constant operands and
5785 VEC_DUPLICATE_EXPR otherwise.
5786 (make_vector_type): Remove temporary is_constant ().
5787 (build_vector_type, build_opaque_vector_type): Take the number of
5788 units as a poly_uint64 rather than an int.
5789 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
5790 VECTOR_CST_NELTS.
5791 * cfgexpand.c (expand_debug_expr): Likewise.
5792 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
5793 (store_constructor, expand_expr_real_1): Likewise.
5794 (const_scalar_mask_from_tree): Likewise.
5795 * fold-const-call.c (fold_const_reduction): Likewise.
5796 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5797 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5798 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5799 (fold_relational_const): Likewise.
5800 (native_interpret_vector): Likewise. Change the size from an
5801 int to an unsigned int.
5802 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5803 TYPE_VECTOR_SUBPARTS.
5804 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5805 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5806 duplicating a non-constant operand into a variable-length vector.
5807 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5808 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5809 * ipa-icf.c (sem_variable::equals): Likewise.
5810 * match.pd: Likewise.
5811 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5812 * print-tree.c (print_node): Likewise.
5813 * stor-layout.c (layout_type): Likewise.
5814 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5815 * tree-cfg.c (verify_gimple_comparison): Likewise.
5816 (verify_gimple_assign_binary): Likewise.
5817 (verify_gimple_assign_ternary): Likewise.
5818 (verify_gimple_assign_single): Likewise.
5819 * tree-pretty-print.c (dump_generic_node): Likewise.
5820 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5821 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5822 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5823 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5824 (vect_shift_permute_load_chain): Likewise.
5825 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5826 (expand_vector_condition, optimize_vector_constructor): Likewise.
5827 (lower_vec_perm, get_compute_type): Likewise.
5828 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5829 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5830 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5831 (vect_recog_mask_conversion_pattern): Likewise.
5832 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5833 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5834 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5835 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5836 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5837 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5838 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5839 (supportable_widening_operation): Likewise.
5840 (supportable_narrowing_operation): Likewise.
5841 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5842 Likewise.
5843 * varasm.c (output_constant): Likewise.
5845 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5846 Alan Hayward <alan.hayward@arm.com>
5847 David Sherwood <david.sherwood@arm.com>
5849 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5850 so that both the length == 3 and length != 3 cases set up their
5851 own permute vectors. Add comments explaining why we know the
5852 number of elements is constant.
5853 (vect_permute_load_chain): Likewise.
5855 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5856 Alan Hayward <alan.hayward@arm.com>
5857 David Sherwood <david.sherwood@arm.com>
5859 * machmode.h (mode_nunits): Change from unsigned char to
5860 poly_uint16_pod.
5861 (ONLY_FIXED_SIZE_MODES): New macro.
5862 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5863 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5864 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5865 New typedefs.
5866 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5867 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5868 or if measurement_type is not polynomial.
5869 * genmodes.c (ZERO_COEFFS): New macro.
5870 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5871 poly_uint16.
5872 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5873 Use ZERO_COEFFS when emitting initializers.
5874 * data-streamer.h (bp_pack_poly_value): New function.
5875 (bp_unpack_poly_value): Likewise.
5876 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5877 for GET_MODE_NUNITS.
5878 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5879 for GET_MODE_NUNITS.
5880 * tree.c (make_vector_type): Remove temporary shim and make
5881 the real function take the number of units as a poly_uint64
5882 rather than an int.
5883 (build_vector_type_for_mode): Handle polynomial nunits.
5884 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5885 * emit-rtl.c (const_vec_series_p_1): Likewise.
5886 (gen_rtx_CONST_VECTOR): Likewise.
5887 * fold-const.c (test_vec_duplicate_folding): Likewise.
5888 * genrecog.c (validate_pattern): Likewise.
5889 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5890 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5891 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5892 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5893 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5894 * rtlanal.c (subreg_get_info): Likewise.
5895 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5896 (vect_grouped_load_supported): Likewise.
5897 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5898 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5899 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5900 (simplify_const_unary_operation, simplify_binary_operation_1)
5901 (simplify_const_binary_operation, simplify_ternary_operation)
5902 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5903 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5904 instead of CONST_VECTOR_NUNITS.
5905 * varasm.c (output_constant_pool_2): Likewise.
5906 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5907 explicit-encoded elements in the XVEC for variable-length vectors.
5909 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5911 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5913 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5914 Alan Hayward <alan.hayward@arm.com>
5915 David Sherwood <david.sherwood@arm.com>
5917 * coretypes.h (fixed_size_mode): Declare.
5918 (fixed_size_mode_pod): New typedef.
5919 * builtins.h (target_builtins::x_apply_args_mode)
5920 (target_builtins::x_apply_result_mode): Change type to
5921 fixed_size_mode_pod.
5922 * builtins.c (apply_args_size, apply_result_size, result_vector)
5923 (expand_builtin_apply_args_1, expand_builtin_apply)
5924 (expand_builtin_return): Update accordingly.
5926 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5928 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5929 * cselib.c (cselib_hash_rtx): Likewise.
5930 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5931 CONST_VECTOR encoding.
5933 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5934 Jeff Law <law@redhat.com>
5936 PR target/83641
5937 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5938 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5939 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5940 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5942 PR target/83641
5943 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5944 explicitly probe *sp in a noreturn function if there were any callee
5945 register saves or frame pointer is needed.
5947 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5949 PR debug/83621
5950 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5951 BLKmode for ternary, binary or unary expressions.
5953 PR debug/83645
5954 * var-tracking.c (delete_vta_debug_insn): New inline function.
5955 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5956 insns from get_insns () to NULL instead of each bb separately.
5957 Use delete_vta_debug_insn. No longer static.
5958 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5959 delete_vta_debug_insns callers.
5960 * rtl.h (delete_vta_debug_insns): Declare.
5961 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5962 instead of variable_tracking_main.
5964 2018-01-03 Martin Sebor <msebor@redhat.com>
5966 PR tree-optimization/83603
5967 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5968 arguments past the endof the argument list in functions declared
5969 without a prototype.
5970 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5971 Avoid checking when arguments are null.
5973 2018-01-03 Martin Sebor <msebor@redhat.com>
5975 PR c/83559
5976 * doc/extend.texi (attribute const): Fix a typo.
5977 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5978 issuing -Wsuggest-attribute for void functions.
5980 2018-01-03 Martin Sebor <msebor@redhat.com>
5982 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5983 offset_int::from instead of wide_int::to_shwi.
5984 (maybe_diag_overlap): Remove assertion.
5985 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5986 * gimple-ssa-sprintf.c (format_directive): Same.
5987 (parse_directive): Same.
5988 (sprintf_dom_walker::compute_format_length): Same.
5989 (try_substitute_return_value): Same.
5991 2018-01-03 Jeff Law <law@redhat.com>
5993 PR middle-end/83654
5994 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5995 non-constant residual for zero at runtime and avoid probing in
5996 that case. Reorganize code for trailing problem to mirror handling
5997 of the residual.
5999 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6001 PR tree-optimization/83501
6002 * tree-ssa-strlen.c (get_string_cst): New.
6003 (handle_char_store): Call get_string_cst.
6005 2018-01-03 Martin Liska <mliska@suse.cz>
6007 PR tree-optimization/83593
6008 * tree-ssa-strlen.c: Include tree-cfg.h.
6009 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6010 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6011 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6012 to false.
6013 (strlen_dom_walker::before_dom_children): Call
6014 gimple_purge_dead_eh_edges. Dump tranformation with details
6015 dump flags.
6016 (strlen_dom_walker::before_dom_children): Update call by adding
6017 new argument cleanup_eh.
6018 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6020 2018-01-03 Martin Liska <mliska@suse.cz>
6022 PR ipa/83549
6023 * cif-code.def (VARIADIC_THUNK): New enum value.
6024 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6025 thunks.
6027 2018-01-03 Jan Beulich <jbeulich@suse.com>
6029 * sse.md (mov<mode>_internal): Tighten condition for when to use
6030 vmovdqu<ssescalarsize> for TI and OI modes.
6032 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6034 Update copyright years.
6036 2018-01-03 Martin Liska <mliska@suse.cz>
6038 PR ipa/83594
6039 * ipa-visibility.c (function_and_variable_visibility): Skip
6040 functions with noipa attribure.
6042 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6044 * gcc.c (process_command): Update copyright notice dates.
6045 * gcov-dump.c (print_version): Ditto.
6046 * gcov.c (print_version): Ditto.
6047 * gcov-tool.c (print_version): Ditto.
6048 * gengtype.c (create_file): Ditto.
6049 * doc/cpp.texi: Bump @copying's copyright year.
6050 * doc/cppinternals.texi: Ditto.
6051 * doc/gcc.texi: Ditto.
6052 * doc/gccint.texi: Ditto.
6053 * doc/gcov.texi: Ditto.
6054 * doc/install.texi: Ditto.
6055 * doc/invoke.texi: Ditto.
6057 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6059 * vector-builder.h (vector_builder::m_full_nelts): Change from
6060 unsigned int to poly_uint64.
6061 (vector_builder::full_nelts): Update prototype accordingly.
6062 (vector_builder::new_vector): Likewise.
6063 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6064 (vector_builder::operator ==): Likewise.
6065 (vector_builder::finalize): Likewise.
6066 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6067 Take the number of elements as a poly_uint64 rather than an
6068 unsigned int.
6069 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6070 from unsigned int to poly_uint64.
6071 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6072 (vec_perm_indices::new_vector): Likewise.
6073 (vec_perm_indices::length): Likewise.
6074 (vec_perm_indices::nelts_per_input): Likewise.
6075 (vec_perm_indices::input_nelts): Likewise.
6076 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6077 number of elements per input as a poly_uint64 rather than an
6078 unsigned int. Use the original encoding for variable-length
6079 vectors, rather than clamping each individual element.
6080 For the second and subsequent elements in each pattern,
6081 clamp the step and base before clamping their sum.
6082 (vec_perm_indices::series_p): Handle polynomial element counts.
6083 (vec_perm_indices::all_in_range_p): Likewise.
6084 (vec_perm_indices_to_tree): Likewise.
6085 (vec_perm_indices_to_rtx): Likewise.
6086 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6087 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6088 (tree_vector_builder::new_binary_operation): Handle polynomial
6089 element counts. Return false if we need to know the number
6090 of elements at compile time.
6091 * fold-const.c (fold_vec_perm): Punt if the number of elements
6092 isn't known at compile time.
6094 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6096 * vec-perm-indices.h (vec_perm_builder): Change element type
6097 from HOST_WIDE_INT to poly_int64.
6098 (vec_perm_indices::element_type): Update accordingly.
6099 (vec_perm_indices::clamp): Handle polynomial element_types.
6100 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6101 (vec_perm_indices::all_in_range_p): Likewise.
6102 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6103 than shwi trees.
6104 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6105 polynomial vec_perm_indices element types.
6106 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6107 * fold-const.c (fold_vec_perm): Likewise.
6108 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6109 * tree-vect-generic.c (lower_vec_perm): Likewise.
6110 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6111 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6112 element type to HOST_WIDE_INT.
6114 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6115 Alan Hayward <alan.hayward@arm.com>
6116 David Sherwood <david.sherwood@arm.com>
6118 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6119 rather than an int. Use plus_constant.
6120 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6121 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6123 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6124 Alan Hayward <alan.hayward@arm.com>
6125 David Sherwood <david.sherwood@arm.com>
6127 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6128 a HOST_WIDE_INT to a poly_int64.
6130 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6131 Alan Hayward <alan.hayward@arm.com>
6132 David Sherwood <david.sherwood@arm.com>
6134 * calls.c (load_register_parameters): Cope with polynomial
6135 mode sizes. Require a constant size for BLKmode parameters
6136 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6137 forces a parameter to be padded at the lsb end in order to
6138 fill a complete number of words, require the parameter size
6139 to be ordered wrt UNITS_PER_WORD.
6141 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6142 Alan Hayward <alan.hayward@arm.com>
6143 David Sherwood <david.sherwood@arm.com>
6145 * reload1.c (spill_stack_slot_width): Change element type
6146 from unsigned int to poly_uint64_pod.
6147 (alter_reg): Treat mode sizes as polynomial.
6149 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6150 Alan Hayward <alan.hayward@arm.com>
6151 David Sherwood <david.sherwood@arm.com>
6153 * reload.c (complex_word_subreg_p): New function.
6154 (reload_inner_reg_of_subreg, push_reload): Use it.
6156 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6157 Alan Hayward <alan.hayward@arm.com>
6158 David Sherwood <david.sherwood@arm.com>
6160 * lra-constraints.c (process_alt_operands): Reject matched
6161 operands whose sizes aren't ordered.
6162 (match_reload): Refer to this check here.
6164 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6165 Alan Hayward <alan.hayward@arm.com>
6166 David Sherwood <david.sherwood@arm.com>
6168 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6169 that the mode size is in the set {1, 2, 4, 8, 16}.
6171 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6172 Alan Hayward <alan.hayward@arm.com>
6173 David Sherwood <david.sherwood@arm.com>
6175 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6176 Use plus_constant instead of gen_rtx_PLUS.
6178 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6179 Alan Hayward <alan.hayward@arm.com>
6180 David Sherwood <david.sherwood@arm.com>
6182 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6183 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6184 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6185 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6186 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6187 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6188 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6189 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6190 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6191 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6192 a poly_int64.
6193 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6194 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6195 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6196 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6197 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6198 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6199 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6200 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6201 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6202 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6203 function.
6204 * expr.c (emit_move_resolve_push): Treat the input and result
6205 of PUSH_ROUNDING as a poly_int64.
6206 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6207 (emit_push_insn): Likewise.
6208 * lra-eliminations.c (mark_not_eliminable): Likewise.
6209 * recog.c (push_operand): Likewise.
6210 * reload1.c (elimination_effects): Likewise.
6211 * rtlanal.c (nonzero_bits1): Likewise.
6212 * calls.c (store_one_arg): Likewise. Require the padding to be
6213 known at compile time.
6215 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6216 Alan Hayward <alan.hayward@arm.com>
6217 David Sherwood <david.sherwood@arm.com>
6219 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6220 Use plus_constant instead of gen_rtx_PLUS.
6222 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6223 Alan Hayward <alan.hayward@arm.com>
6224 David Sherwood <david.sherwood@arm.com>
6226 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6227 rather than an int.
6229 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6230 Alan Hayward <alan.hayward@arm.com>
6231 David Sherwood <david.sherwood@arm.com>
6233 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6234 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6235 via stack temporaries. Treat the mode size as polynomial too.
6237 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6238 Alan Hayward <alan.hayward@arm.com>
6239 David Sherwood <david.sherwood@arm.com>
6241 * expr.c (expand_expr_real_2): When handling conversions involving
6242 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6243 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6244 as a poly_uint64 too.
6246 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6247 Alan Hayward <alan.hayward@arm.com>
6248 David Sherwood <david.sherwood@arm.com>
6250 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6252 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6253 Alan Hayward <alan.hayward@arm.com>
6254 David Sherwood <david.sherwood@arm.com>
6256 * combine.c (can_change_dest_mode): Handle polynomial
6257 REGMODE_NATURAL_SIZE.
6258 * expmed.c (store_bit_field_1): Likewise.
6259 * expr.c (store_constructor): Likewise.
6260 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6261 and polynomial REGMODE_NATURAL_SIZE.
6262 (gen_lowpart_common): Likewise.
6263 * reginfo.c (record_subregs_of_mode): Likewise.
6264 * rtlanal.c (read_modify_subreg_p): Likewise.
6266 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6267 Alan Hayward <alan.hayward@arm.com>
6268 David Sherwood <david.sherwood@arm.com>
6270 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6271 numbers of elements.
6273 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6274 Alan Hayward <alan.hayward@arm.com>
6275 David Sherwood <david.sherwood@arm.com>
6277 * match.pd: Cope with polynomial numbers of vector elements.
6279 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6280 Alan Hayward <alan.hayward@arm.com>
6281 David Sherwood <david.sherwood@arm.com>
6283 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6284 in a POINTER_PLUS_EXPR.
6286 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6287 Alan Hayward <alan.hayward@arm.com>
6288 David Sherwood <david.sherwood@arm.com>
6290 * omp-simd-clone.c (simd_clone_subparts): New function.
6291 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6292 (ipa_simd_modify_function_body): Likewise.
6294 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6295 Alan Hayward <alan.hayward@arm.com>
6296 David Sherwood <david.sherwood@arm.com>
6298 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6299 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6300 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6301 (expand_vector_condition, vector_element): Likewise.
6302 (subparts_gt): New function.
6303 (get_compute_type): Use subparts_gt.
6304 (count_type_subparts): Delete.
6305 (expand_vector_operations_1): Use subparts_gt instead of
6306 count_type_subparts.
6308 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6309 Alan Hayward <alan.hayward@arm.com>
6310 David Sherwood <david.sherwood@arm.com>
6312 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6313 (vect_compile_time_alias): ...this new function. Do the calculation
6314 on poly_ints rather than trees.
6315 (vect_prune_runtime_alias_test_list): Update call accordingly.
6317 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6318 Alan Hayward <alan.hayward@arm.com>
6319 David Sherwood <david.sherwood@arm.com>
6321 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6322 numbers of units.
6323 (vect_schedule_slp_instance): Likewise.
6325 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6326 Alan Hayward <alan.hayward@arm.com>
6327 David Sherwood <david.sherwood@arm.com>
6329 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6330 constant and extern definitions for variable-length vectors.
6331 (vect_get_constant_vectors): Note that the number of units
6332 is known to be constant.
6334 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6335 Alan Hayward <alan.hayward@arm.com>
6336 David Sherwood <david.sherwood@arm.com>
6338 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6339 of units as polynomial. Choose between WIDE and NARROW based
6340 on multiple_p.
6342 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6343 Alan Hayward <alan.hayward@arm.com>
6344 David Sherwood <david.sherwood@arm.com>
6346 * tree-vect-stmts.c (simd_clone_subparts): New function.
6347 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6349 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6350 Alan Hayward <alan.hayward@arm.com>
6351 David Sherwood <david.sherwood@arm.com>
6353 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6354 vectors as polynomial. Use build_index_vector for
6355 IFN_GOMP_SIMD_LANE.
6357 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6358 Alan Hayward <alan.hayward@arm.com>
6359 David Sherwood <david.sherwood@arm.com>
6361 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6362 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6363 for variable-length vectors.
6364 (vectorizable_mask_load_store): Treat the number of units as
6365 polynomial, asserting that it is constant if the condition has
6366 already been enforced.
6367 (vectorizable_store, vectorizable_load): Likewise.
6369 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6370 Alan Hayward <alan.hayward@arm.com>
6371 David Sherwood <david.sherwood@arm.com>
6373 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6374 of units as polynomial. Punt if we can't tell at compile time
6375 which vector contains the final result.
6377 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6378 Alan Hayward <alan.hayward@arm.com>
6379 David Sherwood <david.sherwood@arm.com>
6381 * tree-vect-loop.c (vectorizable_induction): Treat the number
6382 of units as polynomial. Punt on SLP inductions. Use an integer
6383 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6384 cast of such a series for variable-length floating-point
6385 reductions.
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 * tree.h (build_index_vector): Declare.
6392 * tree.c (build_index_vector): New function.
6393 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6394 of units as polynomial, forcibly converting it to a constant if
6395 vectorizable_reduction has already enforced the condition.
6396 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6397 to create a {1,2,3,...} vector.
6398 (vectorizable_reduction): Treat the number of units as polynomial.
6399 Choose vectype_in based on the largest scalar element size rather
6400 than the smallest number of units. Enforce the restrictions
6401 relied on above.
6403 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6404 Alan Hayward <alan.hayward@arm.com>
6405 David Sherwood <david.sherwood@arm.com>
6407 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6408 number of units as polynomial.
6410 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6411 Alan Hayward <alan.hayward@arm.com>
6412 David Sherwood <david.sherwood@arm.com>
6414 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6415 * target.def (autovectorize_vector_sizes): Return the vector sizes
6416 by pointer, using vector_sizes rather than a bitmask.
6417 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6418 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6419 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6420 Likewise.
6421 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6422 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6423 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6424 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6425 * omp-general.c (omp_max_vf): Likewise.
6426 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6427 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6428 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6429 * tree-vect-slp.c (vect_slp_bb): Likewise.
6430 * doc/tm.texi: Regenerate.
6431 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6432 to a poly_uint64.
6433 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6434 the vector size as a poly_uint64 rather than an unsigned int.
6435 (current_vector_size): Change from an unsigned int to a poly_uint64.
6436 (get_vectype_for_scalar_type): Update accordingly.
6437 * tree.h (build_truth_vector_type): Take the size and number of
6438 units as a poly_uint64 rather than an unsigned int.
6439 (build_vector_type): Add a temporary overload that takes
6440 the number of units as a poly_uint64 rather than an unsigned int.
6441 * tree.c (make_vector_type): Likewise.
6442 (build_truth_vector_type): Take the number of units as a poly_uint64
6443 rather than an unsigned int.
6445 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6446 Alan Hayward <alan.hayward@arm.com>
6447 David Sherwood <david.sherwood@arm.com>
6449 * target.def (get_mask_mode): Take the number of units and length
6450 as poly_uint64s rather than unsigned ints.
6451 * targhooks.h (default_get_mask_mode): Update accordingly.
6452 * targhooks.c (default_get_mask_mode): Likewise.
6453 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6454 * doc/tm.texi: Regenerate.
6456 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6457 Alan Hayward <alan.hayward@arm.com>
6458 David Sherwood <david.sherwood@arm.com>
6460 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6461 * omp-general.c (omp_max_vf): Likewise.
6462 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6463 (expand_omp_simd): Handle polynomial safelen.
6464 * omp-low.c (omplow_simd_context): Add a default constructor.
6465 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6466 (lower_rec_simd_input_clauses): Update accordingly.
6467 (lower_rec_input_clauses): Likewise.
6469 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6470 Alan Hayward <alan.hayward@arm.com>
6471 David Sherwood <david.sherwood@arm.com>
6473 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6474 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6475 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6476 (vect_analyze_slp_cost): Likewise.
6477 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6478 (vect_model_load_cost): Likewise.
6480 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6481 Alan Hayward <alan.hayward@arm.com>
6482 David Sherwood <david.sherwood@arm.com>
6484 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6485 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6486 from an unsigned int * to a poly_uint64_pod *.
6487 (calculate_unrolling_factor): New function.
6488 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
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-vectorizer.h (_slp_instance::unrolling_factor): Change
6495 from an unsigned int to a poly_uint64.
6496 (_loop_vec_info::slp_unrolling_factor): Likewise.
6497 (_loop_vec_info::vectorization_factor): Change from an int
6498 to a poly_uint64.
6499 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6500 (vect_get_num_vectors): New function.
6501 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6502 (vect_get_num_copies): Use vect_get_num_vectors.
6503 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6504 to an unsigned int *.
6505 (vect_analyze_data_refs): Change min_vf from an int * to a
6506 poly_uint64 *.
6507 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6508 than an unsigned HOST_WIDE_INT.
6509 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6510 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6511 to an unsigned int *.
6512 (vect_analyze_data_ref_dependences): Likewise.
6513 (vect_compute_data_ref_alignment): Handle polynomial vf.
6514 (vect_enhance_data_refs_alignment): Likewise.
6515 (vect_prune_runtime_alias_test_list): Likewise.
6516 (vect_shift_permute_load_chain): Likewise.
6517 (vect_supportable_dr_alignment): Likewise.
6518 (dependence_distance_ge_vf): Take the vectorization factor as a
6519 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6520 (vect_analyze_data_refs): Change min_vf from an int * to a
6521 poly_uint64 *.
6522 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6523 vfm1 as a poly_uint64 rather than an int. Make the same change
6524 for the returned bound_scalar.
6525 (vect_gen_vector_loop_niters): Handle polynomial vf.
6526 (vect_do_peeling): Likewise. Update call to
6527 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6528 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6529 be constant.
6530 * tree-vect-loop.c (vect_determine_vectorization_factor)
6531 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6532 (vect_get_known_peeling_cost): Likewise.
6533 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6534 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6535 (vect_transform_loop): Likewise. Use the lowest possible VF when
6536 updating the upper bounds of the loop.
6537 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6538 rather than an int.
6539 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6540 polynomial unroll factors.
6541 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6542 (vect_make_slp_decision): Likewise.
6543 (vect_supported_load_permutation_p): Likewise, and polynomial
6544 vf too.
6545 (vect_analyze_slp_cost): Handle polynomial vf.
6546 (vect_slp_analyze_node_operations): Likewise.
6547 (vect_slp_analyze_bb_1): Likewise.
6548 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6549 than an unsigned HOST_WIDE_INT.
6550 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6551 (vectorizable_load): Handle polynomial vf.
6552 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6553 a poly_uint64.
6554 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6556 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6557 Alan Hayward <alan.hayward@arm.com>
6558 David Sherwood <david.sherwood@arm.com>
6560 * match.pd: Handle bit operations involving three constants
6561 and try to fold one pair.
6563 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6565 * tree-vect-loop-manip.c: Include gimple-fold.h.
6566 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6567 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6568 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6569 Add a path that uses a step of VF instead of 1, but disable it
6570 for now.
6571 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6572 and niters_no_overflow parameters. Update calls to
6573 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6574 Create a new SSA name if the latter choses to use a ste other
6575 than zero, and return it via niters_vector_mult_vf_var.
6576 * tree-vect-loop.c (vect_transform_loop): Update calls to
6577 vect_do_peeling, vect_gen_vector_loop_niters and
6578 slpeel_make_loop_iterate_ntimes.
6579 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6580 (vect_gen_vector_loop_niters): Update declarations after above changes.
6582 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6584 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6585 128-bit round to integer instructions.
6586 (ceil<mode>2): Likewise.
6587 (btrunc<mode>2): Likewise.
6588 (round<mode>2): Likewise.
6590 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6592 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6593 unaligned VSX load/store on P8/P9.
6594 (expand_block_clear): Allow the use of unaligned VSX
6595 load/store on P8/P9.
6597 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6599 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6600 New function.
6601 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6602 swap associated with both a load and a store.
6604 2018-01-02 Andrew Waterman <andrew@sifive.com>
6606 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6607 * config/riscv/riscv.md (clear_cache): Use it.
6609 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6611 * web.c: Remove out-of-date comment.
6613 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6615 * expr.c (fixup_args_size_notes): Check that any existing
6616 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6617 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6618 (emit_single_push_insn): ...here.
6620 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6622 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6623 (const_vector_encoded_nelts): New function.
6624 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6625 (const_vector_int_elt, const_vector_elt): Declare.
6626 * emit-rtl.c (const_vector_int_elt_1): New function.
6627 (const_vector_elt): Likewise.
6628 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6629 of CONST_VECTOR_ELT.
6631 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6633 * expr.c: Include rtx-vector-builder.h.
6634 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
6635 directly on the tree encoding.
6636 (const_vector_from_tree): Likewise.
6637 * optabs.c: Include rtx-vector-builder.h.
6638 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
6639 sequence of "u" values.
6640 * vec-perm-indices.c: Include rtx-vector-builder.h.
6641 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
6642 directly on the vec_perm_indices encoding.
6644 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6646 * doc/rtl.texi (const_vector): Describe new encoding scheme.
6647 * Makefile.in (OBJS): Add rtx-vector-builder.o.
6648 * rtx-vector-builder.h: New file.
6649 * rtx-vector-builder.c: Likewise.
6650 * rtl.h (rtx_def::u2): Add a const_vector field.
6651 (CONST_VECTOR_NPATTERNS): New macro.
6652 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
6653 (CONST_VECTOR_DUPLICATE_P): Likewise.
6654 (CONST_VECTOR_STEPPED_P): Likewise.
6655 (CONST_VECTOR_ENCODED_ELT): Likewise.
6656 (const_vec_duplicate_p): Check for a duplicated vector encoding.
6657 (unwrap_const_vec_duplicate): Likewise.
6658 (const_vec_series_p): Check for a non-duplicated vector encoding.
6659 Say that the function only returns true for integer vectors.
6660 * emit-rtl.c: Include rtx-vector-builder.h.
6661 (gen_const_vec_duplicate_1): Delete.
6662 (gen_const_vector): Call gen_const_vec_duplicate instead of
6663 gen_const_vec_duplicate_1.
6664 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
6665 (gen_const_vec_duplicate): Use rtx_vector_builder.
6666 (gen_const_vec_series): Likewise.
6667 (gen_rtx_CONST_VECTOR): Likewise.
6668 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
6669 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6670 Build a new vector rather than modifying a CONST_VECTOR in-place.
6671 (handle_special_swappables): Update call accordingly.
6672 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
6673 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
6674 Build a new vector rather than modifying a CONST_VECTOR in-place.
6675 (handle_special_swappables): Update call accordingly.
6677 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6679 * simplify-rtx.c (simplify_const_binary_operation): Use
6680 CONST_VECTOR_ELT instead of XVECEXP.
6682 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6684 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
6685 the selector elements to be different from the data elements
6686 if the selector is a VECTOR_CST.
6687 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
6688 ssizetype for the selector.
6690 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6692 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
6693 before testing each element individually.
6694 * tree-vect-generic.c (lower_vec_perm): Likewise.
6696 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6698 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
6699 * selftest-run-tests.c (selftest::run_tests): Call it.
6700 * vector-builder.h (vector_builder::operator ==): New function.
6701 (vector_builder::operator !=): Likewise.
6702 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
6703 (vec_perm_indices::all_from_input_p): New function.
6704 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6705 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
6706 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
6707 instead of reading the VECTOR_CST directly. Detect whether both
6708 vector inputs are the same before constructing the vec_perm_indices,
6709 and update the number of inputs argument accordingly. Use the
6710 utility functions added above. Only construct sel2 if we need to.
6712 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6714 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
6715 the broadcast of the low byte.
6716 (expand_mult_highpart): Use an explicit encoding for the permutes.
6717 * optabs-query.c (can_mult_highpart_p): Likewise.
6718 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
6719 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6720 (vectorizable_bswap): Likewise.
6721 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
6722 explicit encoding for the power-of-2 permutes.
6723 (vect_permute_store_chain): Likewise.
6724 (vect_grouped_load_supported): Likewise.
6725 (vect_permute_load_chain): Likewise.
6727 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6729 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
6730 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
6731 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
6732 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6733 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
6734 (vect_gen_perm_mask_any): Likewise.
6736 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6738 * int-vector-builder.h: New file.
6739 * vec-perm-indices.h: Include int-vector-builder.h.
6740 (vec_perm_indices): Redefine as an int_vector_builder.
6741 (auto_vec_perm_indices): Delete.
6742 (vec_perm_builder): Redefine as a stand-alone class.
6743 (vec_perm_indices::vec_perm_indices): New function.
6744 (vec_perm_indices::clamp): Likewise.
6745 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
6746 (vec_perm_indices::new_vector): New function.
6747 (vec_perm_indices::new_expanded_vector): Update for new
6748 vec_perm_indices class.
6749 (vec_perm_indices::rotate_inputs): New function.
6750 (vec_perm_indices::all_in_range_p): Operate directly on the
6751 encoded form, without computing elided elements.
6752 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
6753 encoding. Update for new vec_perm_indices class.
6754 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
6755 the given vec_perm_builder.
6756 (expand_vec_perm_var): Update vec_perm_builder constructor.
6757 (expand_mult_highpart): Use vec_perm_builder instead of
6758 auto_vec_perm_indices.
6759 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
6760 vec_perm_indices instead of auto_vec_perm_indices. Use a single
6761 or double series encoding as appropriate.
6762 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
6763 vec_perm_indices instead of auto_vec_perm_indices.
6764 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6765 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6766 (vect_permute_store_chain): Likewise.
6767 (vect_grouped_load_supported): Likewise.
6768 (vect_permute_load_chain): Likewise.
6769 (vect_shift_permute_load_chain): Likewise.
6770 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6771 (vect_transform_slp_perm_load): Likewise.
6772 (vect_schedule_slp_instance): Likewise.
6773 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6774 (vectorizable_mask_load_store): Likewise.
6775 (vectorizable_bswap): Likewise.
6776 (vectorizable_store): Likewise.
6777 (vectorizable_load): Likewise.
6778 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
6779 vec_perm_indices instead of auto_vec_perm_indices. Use
6780 tree_to_vec_perm_builder to read the vector from a tree.
6781 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
6782 vec_perm_builder instead of a vec_perm_indices.
6783 (have_whole_vector_shift): Use vec_perm_builder and
6784 vec_perm_indices instead of auto_vec_perm_indices. Leave the
6785 truncation to calc_vec_perm_mask_for_shift.
6786 (vect_create_epilog_for_reduction): Likewise.
6787 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
6788 from auto_vec_perm_indices to vec_perm_indices.
6789 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6790 instead of changing individual elements.
6791 (aarch64_vectorize_vec_perm_const): Use new_vector to install
6792 the vector in d.perm.
6793 * config/arm/arm.c (expand_vec_perm_d::perm): Change
6794 from auto_vec_perm_indices to vec_perm_indices.
6795 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6796 instead of changing individual elements.
6797 (arm_vectorize_vec_perm_const): Use new_vector to install
6798 the vector in d.perm.
6799 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6800 Update vec_perm_builder constructor.
6801 (rs6000_expand_interleave): Likewise.
6802 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6803 (rs6000_expand_interleave): Likewise.
6805 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6807 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6808 to qimode could truncate the indices.
6809 * optabs.c (expand_vec_perm_var): Likewise.
6811 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6813 * Makefile.in (OBJS): Add vec-perm-indices.o.
6814 * vec-perm-indices.h: New file.
6815 * vec-perm-indices.c: Likewise.
6816 * target.h (vec_perm_indices): Replace with a forward class
6817 declaration.
6818 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6819 * optabs.h: Include vec-perm-indices.h.
6820 (expand_vec_perm): Delete.
6821 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6822 (expand_vec_perm_const): Declare.
6823 * target.def (vec_perm_const_ok): Replace with...
6824 (vec_perm_const): ...this new hook.
6825 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6826 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6827 * doc/tm.texi: Regenerate.
6828 * optabs.def (vec_perm_const): Delete.
6829 * doc/md.texi (vec_perm_const): Likewise.
6830 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6831 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6832 expand_vec_perm for constant permutation vectors. Assert that
6833 the mode of variable permutation vectors is the integer equivalent
6834 of the mode that is being permuted.
6835 * optabs-query.h (selector_fits_mode_p): Declare.
6836 * optabs-query.c: Include vec-perm-indices.h.
6837 (selector_fits_mode_p): New function.
6838 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6839 is defined, instead of checking whether the vec_perm_const_optab
6840 exists. Use targetm.vectorize.vec_perm_const instead of
6841 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6842 fit in the vector mode before using a variable permute.
6843 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6844 vec_perm_indices instead of an rtx.
6845 (expand_vec_perm): Replace with...
6846 (expand_vec_perm_const): ...this new function. Take the selector
6847 as a vec_perm_indices rather than an rtx. Also take the mode of
6848 the selector. Update call to shift_amt_for_vec_perm_mask.
6849 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6850 Use vec_perm_indices::new_expanded_vector to expand the original
6851 selector into bytes. Check whether the indices fit in the vector
6852 mode before using a variable permute.
6853 (expand_vec_perm_var): Make global.
6854 (expand_mult_highpart): Use expand_vec_perm_const.
6855 * fold-const.c: Includes vec-perm-indices.h.
6856 * tree-ssa-forwprop.c: Likewise.
6857 * tree-vect-data-refs.c: Likewise.
6858 * tree-vect-generic.c: Likewise.
6859 * tree-vect-loop.c: Likewise.
6860 * tree-vect-slp.c: Likewise.
6861 * tree-vect-stmts.c: Likewise.
6862 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6863 Delete.
6864 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6865 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6866 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6867 (aarch64_vectorize_vec_perm_const): ...this new function.
6868 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6869 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6870 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6871 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6872 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6873 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6874 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6875 into...
6876 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6877 check for NEON modes.
6878 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6879 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6880 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6881 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6882 into...
6883 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6884 the old VEC_PERM_CONST conditions.
6885 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6886 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6887 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6888 (ia64_vectorize_vec_perm_const_ok): Merge into...
6889 (ia64_vectorize_vec_perm_const): ...this new function.
6890 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6891 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6892 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6893 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6894 * config/mips/mips.c (mips_expand_vec_perm_const)
6895 (mips_vectorize_vec_perm_const_ok): Merge into...
6896 (mips_vectorize_vec_perm_const): ...this new function.
6897 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6898 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6899 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6900 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6901 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6902 (rs6000_expand_vec_perm_const): Delete.
6903 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6904 Delete.
6905 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6906 (altivec_expand_vec_perm_const_le): Take each operand individually.
6907 Operate on constant selectors rather than rtxes.
6908 (altivec_expand_vec_perm_const): Likewise. Update call to
6909 altivec_expand_vec_perm_const_le.
6910 (rs6000_expand_vec_perm_const): Delete.
6911 (rs6000_vectorize_vec_perm_const_ok): Delete.
6912 (rs6000_vectorize_vec_perm_const): New function.
6913 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6914 an element count and rtx array.
6915 (rs6000_expand_extract_even): Update call accordingly.
6916 (rs6000_expand_interleave): Likewise.
6917 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6918 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6919 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6920 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6921 (rs6000_expand_vec_perm_const): Delete.
6922 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6923 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6924 (altivec_expand_vec_perm_const_le): Take each operand individually.
6925 Operate on constant selectors rather than rtxes.
6926 (altivec_expand_vec_perm_const): Likewise. Update call to
6927 altivec_expand_vec_perm_const_le.
6928 (rs6000_expand_vec_perm_const): Delete.
6929 (rs6000_vectorize_vec_perm_const_ok): Delete.
6930 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6931 reference to the SPE evmerge intructions.
6932 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6933 an element count and rtx array.
6934 (rs6000_expand_extract_even): Update call accordingly.
6935 (rs6000_expand_interleave): Likewise.
6936 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6937 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6938 new function.
6939 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6941 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6943 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6944 vector mode and that that mode matches the mode of the data
6945 being permuted.
6946 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6947 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6948 directly using expand_vec_perm_1 when forcing selectors into
6949 registers.
6950 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6952 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6954 * optabs-query.h (can_vec_perm_p): Delete.
6955 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6956 * optabs-query.c (can_vec_perm_p): Split into...
6957 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6958 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6959 particular selector is valid.
6960 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6961 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6962 (vect_grouped_load_supported): Likewise.
6963 (vect_shift_permute_load_chain): Likewise.
6964 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6965 (vect_transform_slp_perm_load): Likewise.
6966 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6967 (vectorizable_bswap): Likewise.
6968 (vect_gen_perm_mask_checked): Likewise.
6969 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6970 implementations of variable permutation vectors into account
6971 when deciding which selector to use.
6972 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6973 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6974 with a false third argument.
6975 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6976 to test whether the constant selector is valid and can_vec_perm_var_p
6977 to test whether a variable selector is valid.
6979 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6981 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6982 * optabs-query.c (can_vec_perm_p): Likewise.
6983 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6984 instead of vec_perm_indices.
6985 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6986 (vect_gen_perm_mask_checked): Likewise,
6987 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6988 (vect_gen_perm_mask_checked): Likewise,
6990 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6992 * optabs-query.h (qimode_for_vec_perm): Declare.
6993 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6994 (qimode_for_vec_perm): ...this new function.
6995 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6997 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6999 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7000 does not have a conditional at the top.
7002 2018-01-02 Richard Biener <rguenther@suse.de>
7004 * ipa-inline.c (big_speedup_p): Fix expression.
7006 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7008 PR target/81616
7009 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7010 for generic 4->6.
7012 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7014 PR target/81616
7015 Generic tuning.
7016 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7017 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7018 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7019 cond_taken_branch_cost 3->4.
7021 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7023 PR tree-optimization/83581
7024 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7025 TODO_cleanup_cfg if any changes have been made.
7027 PR middle-end/83608
7028 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7029 convert_modes if target mode has the right side, but different mode
7030 class.
7032 PR middle-end/83609
7033 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7034 last argument when extracting from CONCAT. If either from_real or
7035 from_imag is NULL, use expansion through memory. If result is not
7036 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7037 the parts directly to inner mode, if even that fails, use expansion
7038 through memory.
7040 PR middle-end/83623
7041 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7042 check for bswap in mode rather than HImode and use that in expand_unop
7043 too.
7044 \f
7045 Copyright (C) 2018 Free Software Foundation, Inc.
7047 Copying and distribution of this file, with or without modification,
7048 are permitted in any medium without royalty provided the copyright
7049 notice and this notice are preserved.