1 2018-01-12 Richard Biener <rguenther@suse.de>
3 PR tree-optimization/80846
4 * target.def (split_reduction): New target hook.
5 * targhooks.c (default_split_reduction): New function.
6 * targhooks.h (default_split_reduction): Declare.
7 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
8 target requests first reduce vectors by combining low and high
10 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
11 (get_vectype_for_scalar_type_and_size): Export.
12 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
13 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
14 * doc/tm.texi: Regenerate.
15 * config/i386/i386.c (ix86_split_reduction): Implement
16 TARGET_VECTORIZE_SPLIT_REDUCTION.
18 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
21 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
22 in PIC mode except for TARGET_VXWORKS_RTP.
23 * config/sparc/sparc.c: Include cfgrtl.h.
24 (TARGET_INIT_PIC_REG): Define.
25 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
26 (sparc_pic_register_p): New predicate.
27 (sparc_legitimate_address_p): Use it.
28 (sparc_legitimize_pic_address): Likewise.
29 (sparc_delegitimize_address): Likewise.
30 (sparc_mode_dependent_address_p): Likewise.
31 (gen_load_pcrel_sym): Remove 4th parameter.
32 (load_got_register): Adjust call to above. Remove obsolete stuff.
33 (sparc_expand_prologue): Do not call load_got_register here.
34 (sparc_flat_expand_prologue): Likewise.
35 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
36 (sparc_use_pseudo_pic_reg): New function.
37 (sparc_init_pic_reg): Likewise.
38 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
39 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
41 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
43 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
44 Add item for branch_cost.
46 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
48 PR rtl-optimization/83565
49 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
50 not extend the result to a larger mode for rotate operations.
51 (num_sign_bit_copies1): Likewise.
53 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
56 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
58 Use values-Xc.o for -pedantic.
59 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
61 2018-01-12 Martin Liska <mliska@suse.cz>
64 * ipa-devirt.c (final_warning_record::grow_type_warnings):
66 (possible_polymorphic_call_targets): Use it.
67 (ipa_devirt): Likewise.
69 2018-01-12 Martin Liska <mliska@suse.cz>
71 * profile-count.h (enum profile_quality): Use 0 as invalid
72 enum value of profile_quality.
74 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
76 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
79 2018-01-12 Richard Biener <rguenther@suse.de>
81 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
82 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
83 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
85 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
87 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
89 * configure.ac (--with-long-double-format): Add support for the
90 configuration option to change the default long double format on
92 * config.gcc (powerpc*-linux*-*): Likewise.
93 * configure: Regenerate.
94 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
95 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
96 used without modification.
98 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
100 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
101 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
102 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
103 MISC_BUILTIN_SPEC_BARRIER.
104 (rs6000_init_builtins): Likewise.
105 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
107 (speculation_barrier): New define_insn.
108 * doc/extend.texi: Document __builtin_speculation_barrier.
110 2018-01-11 Jakub Jelinek <jakub@redhat.com>
113 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
114 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
115 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
117 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
118 integral modes instead of "ss" and "sd".
119 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
120 vectors with 32-bit and 64-bit elements.
121 (vecdupssescalarmodesuffix): New mode attribute.
122 (vec_dup<mode>): Use it.
124 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
127 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
128 frame if argument is passed on stack.
130 2018-01-11 Jakub Jelinek <jakub@redhat.com>
133 * ree.c (combine_reaching_defs): Optimize also
134 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
135 reg2=any_extend(exp); reg1=reg2;, formatting fix.
137 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
140 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
142 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
145 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
146 after they are computed.
148 2018-01-11 Bin Cheng <bin.cheng@arm.com>
150 PR tree-optimization/83695
151 * gimple-loop-linterchange.cc
152 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
153 reset cached scev information after interchange.
154 (pass_linterchange::execute): Remove call to scev_reset_htab.
156 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
158 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
159 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
160 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
161 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
162 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
163 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
164 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
165 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
166 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
167 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
168 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
169 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
170 (V_lane_reg): Likewise.
171 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
173 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
174 (vfmal_lane_low<mode>_intrinsic,
175 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
176 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
177 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
178 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
179 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
180 vfmsl_lane_high<mode>_intrinsic): New define_insns.
182 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
184 * config/arm/arm-cpus.in (fp16fml): New feature.
185 (ALL_SIMD): Add fp16fml.
186 (armv8.2-a): Add fp16fml as an option.
187 (armv8.3-a): Likewise.
188 (armv8.4-a): Add fp16fml as part of fp16.
189 * config/arm/arm.h (TARGET_FP16FML): Define.
190 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
192 * config/arm/arm-modes.def (V2HF): Define.
193 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
194 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
195 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
196 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
197 vfmsl_low, vfmsl_high): New set of builtins.
198 * config/arm/iterators.md (PLUSMINUS): New code iterator.
199 (vfml_op): New code attribute.
200 (VFMLHALVES): New int iterator.
201 (VFML, VFMLSEL): New mode attributes.
202 (V_reg): Define mapping for V2HF.
203 (V_hi, V_lo): New mode attributes.
204 (VF_constraint): Likewise.
205 (vfml_half, vfml_half_selector): New int attributes.
206 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
208 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
209 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
211 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
212 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
213 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
214 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
216 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
217 Document new effective target and option set.
219 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
221 * config/arm/arm-cpus.in (armv8_4): New feature.
222 (ARMv8_4a): New fgroup.
223 (armv8.4-a): New arch.
224 * config/arm/arm-tables.opt: Regenerate.
225 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
226 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
227 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
228 Add matching rules for -march=armv8.4-a and extensions.
229 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
231 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
234 * config/rx/rx.md (BW): New mode attribute.
235 (sync_lock_test_and_setsi): Add mode suffix to insn output.
237 2018-01-11 Richard Biener <rguenther@suse.de>
239 PR tree-optimization/83435
240 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
241 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
242 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
244 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
245 Alan Hayward <alan.hayward@arm.com>
246 David Sherwood <david.sherwood@arm.com>
248 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
250 (aarch64_classify_address): Initialize it. Track polynomial offsets.
251 (aarch64_print_address_internal): Use it to check for a zero offset.
253 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
254 Alan Hayward <alan.hayward@arm.com>
255 David Sherwood <david.sherwood@arm.com>
257 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
258 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
259 Return a poly_int64 rather than a HOST_WIDE_INT.
260 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
261 rather than a HOST_WIDE_INT.
262 * config/aarch64/aarch64.h (aarch64_frame): Protect with
263 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
264 hard_fp_offset, frame_size, initial_adjust, callee_offset and
265 final_offset from HOST_WIDE_INT to poly_int64.
266 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
267 to_constant when getting the number of units in an Advanced SIMD
269 (aarch64_builtin_vectorized_function): Check for a constant number
271 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
273 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
274 attribute instead of GET_MODE_NUNITS.
275 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
276 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
277 GET_MODE_SIZE for fixed-size registers.
278 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
279 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
280 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
281 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
282 (aarch64_print_operand, aarch64_print_address_internal)
283 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
284 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
285 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
286 Handle polynomial GET_MODE_SIZE.
287 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
288 wider than SImode without modification.
289 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
290 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
291 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
292 passing and returning SVE modes.
293 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
295 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
296 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
297 (aarch64_allocate_and_probe_stack_space): Likewise.
298 (aarch64_layout_frame): Cope with polynomial offsets.
299 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
300 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
302 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
303 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
304 poly_int64 rather than a HOST_WIDE_INT.
305 (aarch64_get_separate_components, aarch64_process_components)
306 (aarch64_expand_prologue, aarch64_expand_epilogue)
307 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
308 (aarch64_anchor_offset): New function, split out from...
309 (aarch64_legitimize_address): ...here.
310 (aarch64_builtin_vectorization_cost): Handle polynomial
311 TYPE_VECTOR_SUBPARTS.
312 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
314 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
315 number of elements from the PARALLEL rather than the mode.
316 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
317 rather than GET_MODE_BITSIZE.
318 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
319 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
320 (aarch64_expand_vec_perm_const_1): Handle polynomial
321 d->perm.length () and d->perm elements.
322 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
323 Apply to_constant to d->perm elements.
324 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
325 polynomial CONST_VECTOR_NUNITS.
326 (aarch64_move_pointer): Take amount as a poly_int64 rather
328 (aarch64_progress_pointer): Avoid temporary variable.
329 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
330 the mode attribute instead of GET_MODE.
332 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
333 Alan Hayward <alan.hayward@arm.com>
334 David Sherwood <david.sherwood@arm.com>
336 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
337 x exists before using it.
338 (aarch64_add_constant_internal): Rename to...
339 (aarch64_add_offset_1): ...this. Replace regnum with separate
340 src and dest rtxes. Handle the case in which they're different,
341 including when the offset is zero. Replace scratchreg with an rtx.
342 Use 2 additions if there is no spare register into which we can
343 move a 16-bit constant.
344 (aarch64_add_constant): Delete.
345 (aarch64_add_offset): Replace reg with separate src and dest
346 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
347 Use aarch64_add_offset_1.
348 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
349 an rtx rather than an int. Take the delta as a poly_int64
350 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
351 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
352 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
353 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
354 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
356 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
357 aarch64_add_constant.
359 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
361 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
362 Use scalar_float_mode.
364 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
366 * config/aarch64/aarch64-simd.md
367 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
368 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
369 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
370 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
371 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
372 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
373 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
374 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
375 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
376 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
378 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
381 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
382 targ_options->x_arm_arch_string is non NULL.
384 2018-01-11 Tamar Christina <tamar.christina@arm.com>
386 * config/aarch64/aarch64.h
387 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
389 2018-01-11 Sudakshina Das <sudi.das@arm.com>
392 * expmed.c (emit_store_flag_force): Swap if const op0
393 and change VOIDmode to mode of op0.
395 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
397 PR rtl-optimization/83761
398 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
399 than bytes to mode_for_size.
401 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
404 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
405 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
408 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
411 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
413 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
414 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
417 2018-01-10 Michael Collison <michael.collison@arm.com>
419 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
420 * config/aarch64/aarch64-option-extension.def: Add
421 AARCH64_OPT_EXTENSION of 'fp16fml'.
422 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
423 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
424 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
425 * config/aarch64/constraints.md (Ui7): New constraint.
426 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
427 (VFMLA_SEL_W): Ditto.
430 (VFMLA16_LOW): New int iterator.
431 (VFMLA16_HIGH): Ditto.
432 (UNSPEC_FMLAL): New unspec.
433 (UNSPEC_FMLSL): Ditto.
434 (UNSPEC_FMLAL2): Ditto.
435 (UNSPEC_FMLSL2): Ditto.
436 (f16mac): New code attribute.
437 * config/aarch64/aarch64-simd-builtins.def
438 (aarch64_fmlal_lowv2sf): Ditto.
439 (aarch64_fmlsl_lowv2sf): Ditto.
440 (aarch64_fmlalq_lowv4sf): Ditto.
441 (aarch64_fmlslq_lowv4sf): Ditto.
442 (aarch64_fmlal_highv2sf): Ditto.
443 (aarch64_fmlsl_highv2sf): Ditto.
444 (aarch64_fmlalq_highv4sf): Ditto.
445 (aarch64_fmlslq_highv4sf): Ditto.
446 (aarch64_fmlal_lane_lowv2sf): Ditto.
447 (aarch64_fmlsl_lane_lowv2sf): Ditto.
448 (aarch64_fmlal_laneq_lowv2sf): Ditto.
449 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
450 (aarch64_fmlalq_lane_lowv4sf): Ditto.
451 (aarch64_fmlsl_lane_lowv4sf): Ditto.
452 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
453 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
454 (aarch64_fmlal_lane_highv2sf): Ditto.
455 (aarch64_fmlsl_lane_highv2sf): Ditto.
456 (aarch64_fmlal_laneq_highv2sf): Ditto.
457 (aarch64_fmlsl_laneq_highv2sf): Ditto.
458 (aarch64_fmlalq_lane_highv4sf): Ditto.
459 (aarch64_fmlsl_lane_highv4sf): Ditto.
460 (aarch64_fmlalq_laneq_highv4sf): Ditto.
461 (aarch64_fmlsl_laneq_highv4sf): Ditto.
462 * config/aarch64/aarch64-simd.md:
463 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
464 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
465 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
466 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
467 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
468 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
469 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
470 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
471 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
472 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
473 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
474 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
475 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
476 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
477 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
478 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
479 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
480 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
481 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
482 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
483 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
484 (vfmlsl_low_u32): Ditto.
485 (vfmlalq_low_u32): Ditto.
486 (vfmlslq_low_u32): Ditto.
487 (vfmlal_high_u32): Ditto.
488 (vfmlsl_high_u32): Ditto.
489 (vfmlalq_high_u32): Ditto.
490 (vfmlslq_high_u32): Ditto.
491 (vfmlal_lane_low_u32): Ditto.
492 (vfmlsl_lane_low_u32): Ditto.
493 (vfmlal_laneq_low_u32): Ditto.
494 (vfmlsl_laneq_low_u32): Ditto.
495 (vfmlalq_lane_low_u32): Ditto.
496 (vfmlslq_lane_low_u32): Ditto.
497 (vfmlalq_laneq_low_u32): Ditto.
498 (vfmlslq_laneq_low_u32): Ditto.
499 (vfmlal_lane_high_u32): Ditto.
500 (vfmlsl_lane_high_u32): Ditto.
501 (vfmlal_laneq_high_u32): Ditto.
502 (vfmlsl_laneq_high_u32): Ditto.
503 (vfmlalq_lane_high_u32): Ditto.
504 (vfmlslq_lane_high_u32): Ditto.
505 (vfmlalq_laneq_high_u32): Ditto.
506 (vfmlslq_laneq_high_u32): Ditto.
507 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
508 (AARCH64_FL_FOR_ARCH8_4): New.
509 (AARCH64_ISA_F16FML): New ISA flag.
510 (TARGET_F16FML): New feature flag for fp16fml.
511 (doc/invoke.texi): Document new fp16fml option.
513 2018-01-10 Michael Collison <michael.collison@arm.com>
515 * config/aarch64/aarch64-builtins.c:
516 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
517 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
518 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
519 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
520 (AARCH64_ISA_SHA3): New ISA flag.
521 (TARGET_SHA3): New feature flag for sha3.
522 * config/aarch64/iterators.md (sha512_op): New int attribute.
523 (CRYPTO_SHA512): New int iterator.
524 (UNSPEC_SHA512H): New unspec.
525 (UNSPEC_SHA512H2): Ditto.
526 (UNSPEC_SHA512SU0): Ditto.
527 (UNSPEC_SHA512SU1): Ditto.
528 * config/aarch64/aarch64-simd-builtins.def
529 (aarch64_crypto_sha512hqv2di): New builtin.
530 (aarch64_crypto_sha512h2qv2di): Ditto.
531 (aarch64_crypto_sha512su0qv2di): Ditto.
532 (aarch64_crypto_sha512su1qv2di): Ditto.
533 (aarch64_eor3qv8hi): Ditto.
534 (aarch64_rax1qv2di): Ditto.
535 (aarch64_xarqv2di): Ditto.
536 (aarch64_bcaxqv8hi): Ditto.
537 * config/aarch64/aarch64-simd.md:
538 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
539 (aarch64_crypto_sha512su0qv2di): Ditto.
540 (aarch64_crypto_sha512su1qv2di): Ditto.
541 (aarch64_eor3qv8hi): Ditto.
542 (aarch64_rax1qv2di): Ditto.
543 (aarch64_xarqv2di): Ditto.
544 (aarch64_bcaxqv8hi): Ditto.
545 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
546 (vsha512h2q_u64): Ditto.
547 (vsha512su0q_u64): Ditto.
548 (vsha512su1q_u64): Ditto.
553 * config/arm/types.md (crypto_sha512): New type attribute.
554 (crypto_sha3): Ditto.
555 (doc/invoke.texi): Document new sha3 option.
557 2018-01-10 Michael Collison <michael.collison@arm.com>
559 * config/aarch64/aarch64-builtins.c:
560 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
561 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
562 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
563 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
564 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
565 (AARCH64_ISA_SM4): New ISA flag.
566 (TARGET_SM4): New feature flag for sm4.
567 * config/aarch64/aarch64-simd-builtins.def
568 (aarch64_sm3ss1qv4si): Ditto.
569 (aarch64_sm3tt1aq4si): Ditto.
570 (aarch64_sm3tt1bq4si): Ditto.
571 (aarch64_sm3tt2aq4si): Ditto.
572 (aarch64_sm3tt2bq4si): Ditto.
573 (aarch64_sm3partw1qv4si): Ditto.
574 (aarch64_sm3partw2qv4si): Ditto.
575 (aarch64_sm4eqv4si): Ditto.
576 (aarch64_sm4ekeyqv4si): Ditto.
577 * config/aarch64/aarch64-simd.md:
578 (aarch64_sm3ss1qv4si): Ditto.
579 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
580 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
581 (aarch64_sm4eqv4si): Ditto.
582 (aarch64_sm4ekeyqv4si): Ditto.
583 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
585 (CRYPTO_SM3TT): Ditto.
586 (CRYPTO_SM3PART): Ditto.
587 (UNSPEC_SM3SS1): New unspec.
588 (UNSPEC_SM3TT1A): Ditto.
589 (UNSPEC_SM3TT1B): Ditto.
590 (UNSPEC_SM3TT2A): Ditto.
591 (UNSPEC_SM3TT2B): Ditto.
592 (UNSPEC_SM3PARTW1): Ditto.
593 (UNSPEC_SM3PARTW2): Ditto.
594 (UNSPEC_SM4E): Ditto.
595 (UNSPEC_SM4EKEY): Ditto.
596 * config/aarch64/constraints.md (Ui2): New constraint.
597 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
598 * config/arm/types.md (crypto_sm3): New type attribute.
600 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
601 (vsm3tt1aq_u32): Ditto.
602 (vsm3tt1bq_u32): Ditto.
603 (vsm3tt2aq_u32): Ditto.
604 (vsm3tt2bq_u32): Ditto.
605 (vsm3partw1q_u32): Ditto.
606 (vsm3partw2q_u32): Ditto.
608 (vsm4ekeyq_u32): Ditto.
609 (doc/invoke.texi): Document new sm4 option.
611 2018-01-10 Michael Collison <michael.collison@arm.com>
613 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
614 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
615 (AARCH64_FL_FOR_ARCH8_4): New.
616 (AARCH64_FL_V8_4): New flag.
617 (doc/invoke.texi): Document new armv8.4-a option.
619 2018-01-10 Michael Collison <michael.collison@arm.com>
621 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
622 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
623 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
624 * config/aarch64/aarch64-option-extension.def: Add
625 AARCH64_OPT_EXTENSION of 'sha2'.
626 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
627 (crypto): Disable sha2 and aes if crypto disabled.
628 (crypto): Enable aes and sha2 if enabled.
629 (simd): Disable sha2 and aes if simd disabled.
630 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
632 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
633 (TARGET_SHA2): New feature flag for sha2.
634 (TARGET_AES): New feature flag for aes.
635 * config/aarch64/aarch64-simd.md:
636 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
637 conditional on TARGET_AES.
638 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
639 (aarch64_crypto_sha1hsi): Make pattern conditional
641 (aarch64_crypto_sha1hv4si): Ditto.
642 (aarch64_be_crypto_sha1hv4si): Ditto.
643 (aarch64_crypto_sha1su1v4si): Ditto.
644 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
645 (aarch64_crypto_sha1su0v4si): Ditto.
646 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
647 (aarch64_crypto_sha256su0v4si): Ditto.
648 (aarch64_crypto_sha256su1v4si): Ditto.
649 (doc/invoke.texi): Document new aes and sha2 options.
651 2018-01-10 Martin Sebor <msebor@redhat.com>
653 PR tree-optimization/83781
654 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
657 2018-01-11 Martin Sebor <msebor@gmail.com>
658 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
660 PR tree-optimization/83501
661 PR tree-optimization/81703
663 * tree-ssa-strlen.c (get_string_cst): Rename...
664 (get_string_len): ...to this. Handle global constants.
665 (handle_char_store): Adjust.
667 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
668 Jim Wilson <jimw@sifive.com>
670 * config/riscv/riscv-protos.h (riscv_output_return): New.
671 * config/riscv/riscv.c (struct machine_function): New naked_p field.
672 (riscv_attribute_table, riscv_output_return),
673 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
674 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
675 (riscv_compute_frame_info): Only compute frame->mask if not a naked
677 (riscv_expand_prologue): Add early return for naked function.
678 (riscv_expand_epilogue): Likewise.
679 (riscv_function_ok_for_sibcall): Return false for naked function.
680 (riscv_set_current_function): New.
681 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
682 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
683 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
684 * doc/extend.texi (RISC-V Function Attributes): New.
686 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
688 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
689 check for 128-bit long double before checking TCmode.
690 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
691 128-bit long doubles before checking TFmode or TCmode.
692 (FLOAT128_IBM_P): Likewise.
694 2018-01-10 Martin Sebor <msebor@redhat.com>
696 PR tree-optimization/83671
697 * builtins.c (c_strlen): Unconditionally return zero for the empty
699 Use -Warray-bounds for warnings.
700 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
701 for non-constant array indices with COMPONENT_REF, arrays of
702 arrays, and pointers to arrays.
703 (gimple_fold_builtin_strlen): Determine and set length range for
704 non-constant character arrays.
706 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
709 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
712 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
714 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
716 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
719 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
720 VECTOR_MEM_ALTIVEC_OR_VSX_P.
721 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
722 indexed_or_indirect_operand predicate.
723 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
724 (*vsx_le_perm_load_v8hi): Likewise.
725 (*vsx_le_perm_load_v16qi): Likewise.
726 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
727 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
728 (*vsx_le_perm_store_v8hi): Likewise.
729 (*vsx_le_perm_store_v16qi): Likewise.
730 (eight unnamed splitters): Likewise.
732 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
734 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
735 * config/rs6000/emmintrin.h: Likewise.
736 * config/rs6000/mmintrin.h: Likewise.
737 * config/rs6000/xmmintrin.h: Likewise.
739 2018-01-10 David Malcolm <dmalcolm@redhat.com>
742 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
744 * tree.c (tree_nop_conversion): Return true for location wrapper
746 (maybe_wrap_with_location): New function.
747 (selftest::check_strip_nops): New function.
748 (selftest::test_location_wrappers): New function.
749 (selftest::tree_c_tests): Call it.
750 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
751 (maybe_wrap_with_location): New decl.
752 (EXPR_LOCATION_WRAPPER_P): New macro.
753 (location_wrapper_p): New inline function.
754 (tree_strip_any_location_wrapper): New inline function.
756 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
759 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
760 stack_realign_offset for the largest alignment of stack slot
762 (ix86_find_max_used_stack_alignment): New function.
763 (ix86_finalize_stack_frame_flags): Use it. Set
764 max_used_stack_alignment if we don't realign stack.
765 * config/i386/i386.h (machine_function): Add
766 max_used_stack_alignment.
768 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
770 * config/arm/arm.opt (-mbranch-cost): New option.
771 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
774 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
777 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
778 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
780 2018-01-10 Richard Biener <rguenther@suse.de>
783 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
784 early out so it also covers the case where we have a non-NULL
787 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
789 PR tree-optimization/83753
790 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
791 for non-strided grouped accesses if the number of elements is 1.
793 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
796 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
797 * i386.h (TARGET_USE_GATHER): Define.
798 * x86-tune.def (X86_TUNE_USE_GATHER): New.
800 2018-01-10 Martin Liska <mliska@suse.cz>
803 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
804 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
806 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
807 CLEANUP_NO_PARTITIONING is not set.
809 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
811 * doc/rtl.texi: Remove documentation of (const ...) wrappers
812 for vectors, as a partial revert of r254296.
813 * rtl.h (const_vec_p): Delete.
814 (const_vec_duplicate_p): Don't test for vector CONSTs.
815 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
816 * expmed.c (make_tree): Likewise.
819 * common.md (E, F): Use CONSTANT_P instead of checking for
821 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
822 checking for CONST_VECTOR.
824 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
827 * predict.c (force_edge_cold): Handle in more sane way edges
830 2018-01-09 Carl Love <cel@us.ibm.com>
832 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
834 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
835 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
836 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
837 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
838 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
839 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
840 * config/rs6000/rs6000-protos.h: Add extern defition for
841 rs6000_generate_float2_double_code.
842 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
844 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
845 (float2_v2df): Add define_expand.
847 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
850 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
851 op_mode in the force_to_mode call.
853 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
855 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
856 instead of checking each element individually.
857 (aarch64_evpc_uzp): Likewise.
858 (aarch64_evpc_zip): Likewise.
859 (aarch64_evpc_ext): Likewise.
860 (aarch64_evpc_rev): Likewise.
861 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
862 instead of checking each element individually. Return true without
864 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
865 whether all selected elements come from the same input, instead of
866 checking each element individually. Remove calls to gen_rtx_REG,
867 start_sequence and end_sequence and instead assert that no rtl is
870 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
872 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
873 order of HIGH and CONST checks.
875 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
877 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
878 if the destination isn't an SSA_NAME.
880 2018-01-09 Richard Biener <rguenther@suse.de>
882 PR tree-optimization/83668
883 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
885 (canonicalize_loop_form): ... here, renamed from ...
886 (canonicalize_loop_closed_ssa_form): ... this and amended to
887 swap successor edges for loop exit blocks to make us use
888 the RPO order we need for initial schedule generation.
890 2018-01-09 Joseph Myers <joseph@codesourcery.com>
892 PR tree-optimization/64811
893 * match.pd: When optimizing comparisons with Inf, avoid
894 introducing or losing exceptions from comparisons with NaN.
896 2018-01-09 Martin Liska <mliska@suse.cz>
899 * asan.c (shadow_mem_size): Add gcc_assert.
901 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
903 Don't save registers in main().
906 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
907 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
908 * config/avr/avr.c (avr_set_current_function): Don't error if
909 naked, OS_task or OS_main are specified at the same time.
910 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
912 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
914 * common/config/avr/avr-common.c (avr_option_optimization_table):
915 Switch on -mmain-is-OS_task for optimizing compilations.
917 2018-01-09 Richard Biener <rguenther@suse.de>
919 PR tree-optimization/83572
920 * graphite.c: Include cfganal.h.
921 (graphite_transform_loops): Connect infinite loops to exit
922 and remove fake edges at the end.
924 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
926 * ipa-inline.c (edge_badness): Revert accidental checkin.
928 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
931 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
932 symbols; not inline clones.
934 2018-01-09 Jakub Jelinek <jakub@redhat.com>
937 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
938 hard registers. Formatting fixes.
940 PR preprocessor/83722
941 * gcc.c (try_generate_repro): Pass
942 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
943 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
946 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
947 Kito Cheng <kito.cheng@gmail.com>
949 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
950 (riscv_leaf_function_p): Delete.
951 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
953 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
955 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
957 (do_ifelse): New function.
958 (do_isel): New function.
959 (do_sub3): New function.
960 (do_add3): New function.
961 (do_load_mask_compare): New function.
962 (do_overlap_load_compare): New function.
963 (expand_compare_loop): New function.
964 (expand_block_compare): Call expand_compare_loop() when appropriate.
965 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
967 (-mblock-compare-inline-loop-limit): New option.
969 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
972 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
973 Reverse order of second and third operands in first alternative.
974 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
975 of first and second elements in UNSPEC_VPERMR vector.
976 (altivec_expand_vec_perm_le): Likewise.
978 2017-01-08 Jeff Law <law@redhat.com>
980 PR rtl-optimizatin/81308
981 * tree-switch-conversion.c (cfg_altered): New file scoped static.
982 (process_switch): If group_case_labels makes a change, then set
984 (pass_convert_switch::execute): If a switch is converted, then
985 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
987 PR rtl-optimization/81308
988 * recog.c (split_all_insns): Conditionally cleanup the CFG after
991 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
993 PR target/83663 - Revert r255946
994 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
995 generation for cases where splatting a value is not useful.
996 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
997 across a vec_duplicate and a paradoxical subreg forming a vector
998 mode to a vec_concat.
1000 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1002 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
1003 -march=armv8.3-a variants.
1004 * config/arm/t-multilib: Likewise.
1005 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
1007 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1009 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
1011 (cceq_ior_compare_complement): Give it a name so I can use it, and
1012 change boolean_or_operator predicate to boolean_operator so it can
1013 be used to generate a crand.
1014 (eqne): New code iterator.
1015 (bd/bd_neg): New code_attrs.
1016 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
1017 a single define_insn.
1018 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
1019 decrement (bdnzt/bdnzf/bdzt/bdzf).
1020 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
1021 with the new names of the branch decrement patterns, and added the
1022 names of the branch decrement conditional patterns.
1024 2018-01-08 Richard Biener <rguenther@suse.de>
1026 PR tree-optimization/83563
1027 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
1030 2018-01-08 Richard Biener <rguenther@suse.de>
1033 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
1035 2018-01-08 Richard Biener <rguenther@suse.de>
1037 PR tree-optimization/83685
1038 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
1039 references to abnormals.
1041 2018-01-08 Richard Biener <rguenther@suse.de>
1044 * dwarf2out.c (output_indirect_strings): Handle empty
1045 skeleton_debug_str_hash.
1046 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
1048 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
1050 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
1051 (emit_store_direct): Likewise.
1052 (arc_trampoline_adjust_address): Likewise.
1053 (arc_asm_trampoline_template): New function.
1054 (arc_initialize_trampoline): Use asm_trampoline_template.
1055 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
1056 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
1057 * config/arc/arc.md (flush_icache): Delete pattern.
1059 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
1061 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
1062 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
1065 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
1068 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
1069 by not USED_FOR_TARGET.
1070 (make_pass_resolve_sw_modes): Likewise.
1072 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
1074 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
1077 2018-01-08 Richard Biener <rguenther@suse.de>
1080 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
1082 2018-01-08 Richard Biener <rguenther@suse.de>
1085 * match.pd ((t * 2) / 2) -> t): Add missing :c.
1087 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
1090 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
1091 basic blocks with a small number of successors.
1092 (convert_control_dep_chain_into_preds): Improve handling of
1094 (dump_predicates): Split apart into...
1095 (dump_pred_chain): ...here...
1096 (dump_pred_info): ...and here.
1097 (can_one_predicate_be_invalidated_p): Add debugging printfs.
1098 (can_chain_union_be_invalidated_p): Improve check for invalidation
1100 (uninit_uses_cannot_happen): Avoid unnecessary if
1101 convert_control_dep_chain_into_preds yielded nothing.
1103 2018-01-06 Martin Sebor <msebor@redhat.com>
1105 PR tree-optimization/83640
1106 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
1107 subtracting negative offset from size.
1108 (builtin_access::overlap): Adjust offset bounds of the access to fall
1109 within the size of the object if possible.
1111 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
1113 PR rtl-optimization/83699
1114 * expmed.c (extract_bit_field_1): Restrict the vector usage of
1115 extract_bit_field_as_subreg to cases in which the extracted
1116 value is also a vector.
1118 * lra-constraints.c (process_alt_operands): Test for the equivalence
1119 substitutions when detecting a possible reload cycle.
1121 2018-01-06 Jakub Jelinek <jakub@redhat.com>
1124 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
1125 by default if flag_selective_schedling{,2}. Formatting fixes.
1127 PR rtl-optimization/83682
1128 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
1129 if it has non-VECTOR_MODE element mode.
1130 (vec_duplicate_p): Likewise.
1133 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
1134 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
1136 2018-01-05 Jakub Jelinek <jakub@redhat.com>
1139 * config/i386/i386-builtin.def
1140 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
1141 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
1142 Require also OPTION_MASK_ISA_AVX512F in addition to
1143 OPTION_MASK_ISA_GFNI.
1144 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
1145 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
1146 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
1147 to OPTION_MASK_ISA_GFNI.
1148 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
1149 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
1150 OPTION_MASK_ISA_AVX512BW.
1151 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
1152 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
1153 addition to OPTION_MASK_ISA_GFNI.
1154 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
1155 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
1156 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
1157 to OPTION_MASK_ISA_GFNI.
1158 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
1159 a requirement for all ISAs rather than any of them with a few
1161 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
1163 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
1164 bitmasks to be enabled with 3 exceptions, instead of requiring any
1165 enabled ISA with lots of exceptions.
1166 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
1167 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
1168 Change avx512bw in isa attribute to avx512f.
1169 * config/i386/sgxintrin.h: Add license boilerplate.
1170 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
1171 to __AVX512F__ and __AVX512VL to __AVX512VL__.
1172 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
1173 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
1175 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
1176 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
1177 temporarily sse2 rather than sse if not enabled already.
1180 * config/i386/sse.md (VI248_VLBW): Rename to ...
1181 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
1182 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
1183 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
1184 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
1185 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
1186 mode iterator instead of VI248_VLBW.
1188 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
1190 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
1191 (record_modified): Skip clobbers; add debug output.
1192 (param_change_prob): Use sreal frequencies.
1194 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
1196 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
1197 punt for user-aligned variables.
1199 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
1201 * tree-chrec.c (chrec_contains_symbols): Return true for
1204 2018-01-05 Sudakshina Das <sudi.das@arm.com>
1207 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
1208 of (x|y) == x for BICS pattern.
1210 2018-01-05 Jakub Jelinek <jakub@redhat.com>
1212 PR tree-optimization/83605
1213 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
1214 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
1217 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
1219 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
1220 * config/epiphany/rtems.h: New file.
1222 2018-01-04 Jakub Jelinek <jakub@redhat.com>
1223 Uros Bizjak <ubizjak@gmail.com>
1226 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
1227 QIreg_operand instead of register_operand predicate.
1228 * config/i386/i386.c (ix86_rop_should_change_byte_p,
1229 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
1230 comments instead of -fmitigate[-_]rop.
1232 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1235 * cgraphunit.c (symbol_table::compile): Switch to text_section
1236 before calling assembly_start debug hook.
1237 * run-rtl-passes.c (run_rtl_passes): Likewise.
1240 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
1242 * tree-vrp.c (extract_range_from_binary_expr_1): Check
1243 range_int_cst_p rather than !symbolic_range_p before calling
1244 extract_range_from_multiplicative_op_1.
1246 2017-01-04 Jeff Law <law@redhat.com>
1248 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
1249 redundant test in assertion.
1251 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
1253 * doc/rtl.texi: Document machine_mode wrapper classes.
1255 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
1257 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
1260 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
1262 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
1263 the VEC_PERM_EXPR fold to fail.
1265 2018-01-04 Jakub Jelinek <jakub@redhat.com>
1268 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
1269 to switched_sections.
1271 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
1274 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
1277 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
1280 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
1281 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
1283 2018-01-04 Jakub Jelinek <jakub@redhat.com>
1286 * cfgexpand.c (expand_dbeug_expr) <case BIT_FIELD_REF>: Punt if mode
1287 is BLKmode and bitpos not zero or mode change is needed.
1289 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
1292 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
1295 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
1298 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
1299 instead of MULT rtx. Update all corresponding splitters.
1301 (*ssub<modesuffix>): Ditto.
1303 (*cmp_sadd_di): Update split patterns.
1304 (*cmp_sadd_si): Ditto.
1305 (*cmp_sadd_sidi): Ditto.
1306 (*cmp_ssub_di): Ditto.
1307 (*cmp_ssub_si): Ditto.
1308 (*cmp_ssub_sidi): Ditto.
1309 * config/alpha/predicates.md (const23_operand): New predicate.
1310 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
1311 Look for ASHIFT, not MULT inner operand.
1312 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
1314 2018-01-04 Martin Liska <mliska@suse.cz>
1316 PR gcov-profile/83669
1317 * gcov.c (output_intermediate_file): Add version to intermediate
1319 * doc/gcov.texi: Document new field 'version' in intermediate
1320 file format. Fix location of '-k' option of gcov command.
1322 2018-01-04 Martin Liska <mliska@suse.cz>
1325 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
1327 2018-01-04 Jakub Jelinek <jakub@redhat.com>
1329 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
1331 2018-01-03 Martin Sebor <msebor@redhat.com>
1333 PR tree-optimization/83655
1334 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
1335 checking calls with invalid arguments.
1337 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1339 * tree-vect-stmts.c (vect_get_store_rhs): New function.
1340 (vectorizable_mask_load_store): Delete.
1341 (vectorizable_call): Return false for masked loads and stores.
1342 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
1343 instead of gimple_assign_rhs1.
1344 (vectorizable_load): Handle IFN_MASK_LOAD.
1345 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
1347 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1349 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
1351 (vectorizable_mask_load_store): ...here.
1352 (vectorizable_load): ...and here.
1354 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1356 * tree-vect-stmts.c (vect_build_all_ones_mask)
1357 (vect_build_zero_merge_argument): New functions, split out from...
1358 (vectorizable_load): ...here.
1360 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1362 * tree-vect-stmts.c (vect_check_store_rhs): New function,
1364 (vectorizable_mask_load_store): ...here.
1365 (vectorizable_store): ...and here.
1367 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1369 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
1371 (vectorizable_mask_load_store): ...here.
1373 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1375 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
1376 (vect_model_store_cost): Take a vec_load_store_type instead of a
1378 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
1379 (vect_model_store_cost): Take a vec_load_store_type instead of a
1381 (vectorizable_mask_load_store): Update accordingly.
1382 (vectorizable_store): Likewise.
1383 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
1385 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1387 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
1388 IFN_MASK_LOAD calls here rather than...
1389 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
1391 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1392 Alan Hayward <alan.hayward@arm.com>
1393 David Sherwood <david.sherwood@arm.com>
1395 * expmed.c (extract_bit_field_1): For vector extracts,
1396 fall back to extract_bit_field_as_subreg if vec_extract
1399 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1400 Alan Hayward <alan.hayward@arm.com>
1401 David Sherwood <david.sherwood@arm.com>
1403 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
1404 they are variable or constant sized.
1405 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
1406 slots for constant-sized data.
1408 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1409 Alan Hayward <alan.hayward@arm.com>
1410 David Sherwood <david.sherwood@arm.com>
1412 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
1413 handling COND_EXPRs with boolean comparisons, try to find a better
1414 basis for the mask type than the boolean itself.
1416 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1418 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
1419 is calculated and how it can be overridden.
1420 * genmodes.c (max_bitsize_mode_any_mode): New variable.
1421 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
1423 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
1426 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1427 Alan Hayward <alan.hayward@arm.com>
1428 David Sherwood <david.sherwood@arm.com>
1430 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
1431 Remove the mode argument.
1432 (aarch64_simd_valid_immediate): Remove the mode and inverse
1434 * config/aarch64/iterators.md (bitsize): New iterator.
1435 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
1436 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
1437 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
1438 aarch64_simd_valid_immediate.
1439 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
1440 (aarch64_reg_or_bic_imm): Likewise.
1441 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
1442 with an insn_type enum and msl with a modifier_type enum.
1443 Replace element_width with a scalar_mode. Change the shift
1444 to unsigned int. Add constructors for scalar_float_mode and
1445 scalar_int_mode elements.
1446 (aarch64_vect_float_const_representable_p): Delete.
1447 (aarch64_can_const_movi_rtx_p)
1448 (aarch64_simd_scalar_immediate_valid_for_move)
1449 (aarch64_simd_make_constant): Update call to
1450 aarch64_simd_valid_immediate.
1451 (aarch64_advsimd_valid_immediate_hs): New function.
1452 (aarch64_advsimd_valid_immediate): Likewise.
1453 (aarch64_simd_valid_immediate): Remove mode and inverse
1454 arguments. Rewrite to use the above. Use const_vec_duplicate_p
1455 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
1456 and aarch64_float_const_representable_p on the result.
1457 (aarch64_output_simd_mov_immediate): Remove mode argument.
1458 Update call to aarch64_simd_valid_immediate and use of
1459 simd_immediate_info.
1460 (aarch64_output_scalar_simd_mov_immediate): Update call
1463 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1464 Alan Hayward <alan.hayward@arm.com>
1465 David Sherwood <david.sherwood@arm.com>
1467 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
1468 (mode_nunits): Likewise CONST_MODE_NUNITS.
1469 * machmode.def (ADJUST_NUNITS): Document.
1470 * genmodes.c (mode_data::need_nunits_adj): New field.
1471 (blank_mode): Update accordingly.
1472 (adj_nunits): New variable.
1473 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
1475 (emit_mode_size_inline): Set need_bytesize_adj for all modes
1476 listed in adj_nunits.
1477 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
1478 listed in adj_nunits. Don't emit case statements for such modes.
1479 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
1480 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
1481 nothing if adj_nunits is nonnull.
1482 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
1483 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
1484 (emit_mode_fbit): Update use of print_maybe_const_decl.
1485 (emit_move_size): Likewise. Treat the array as non-const
1487 (emit_mode_adjustments): Handle adj_nunits.
1489 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1491 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
1492 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
1493 (VECTOR_MODES): Use it.
1494 (make_vector_modes): Take the prefix as an argument.
1496 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1497 Alan Hayward <alan.hayward@arm.com>
1498 David Sherwood <david.sherwood@arm.com>
1500 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
1501 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
1502 for MODE_VECTOR_BOOL.
1503 * machmode.def (VECTOR_BOOL_MODE): Document.
1504 * genmodes.c (VECTOR_BOOL_MODE): New macro.
1505 (make_vector_bool_mode): New function.
1506 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
1508 * lto-streamer-in.c (lto_input_mode_table): Likewise.
1509 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
1511 * stor-layout.c (int_mode_for_mode): Likewise.
1512 * tree.c (build_vector_type_for_mode): Likewise.
1513 * varasm.c (output_constant_pool_2): Likewise.
1514 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
1515 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
1516 for MODE_VECTOR_BOOL.
1517 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
1518 of mode class checks.
1519 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
1520 instead of a list of mode class checks.
1521 (expand_vector_scalar_condition): Likewise.
1522 (type_for_widest_vector_mode): Handle BImode as an inner mode.
1524 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1525 Alan Hayward <alan.hayward@arm.com>
1526 David Sherwood <david.sherwood@arm.com>
1528 * machmode.h (mode_size): Change from unsigned short to
1530 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
1531 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
1532 or if measurement_type is not polynomial.
1533 (fixed_size_mode::includes_p): Check for constant-sized modes.
1534 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
1535 return a poly_uint16 rather than an unsigned short.
1536 (emit_mode_size): Change the type of mode_size from unsigned short
1537 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
1538 (emit_mode_adjustments): Cope with polynomial vector sizes.
1539 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
1541 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
1543 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
1544 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
1545 * caller-save.c (setup_save_areas): Likewise.
1546 (replace_reg_with_saved_mem): Likewise.
1547 * calls.c (emit_library_call_value_1): Likewise.
1548 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
1549 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
1550 (gen_lowpart_for_combine): Likewise.
1551 * convert.c (convert_to_integer_1): Likewise.
1552 * cse.c (equiv_constant, cse_insn): Likewise.
1553 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
1554 (cselib_subst_to_values): Likewise.
1555 * dce.c (word_dce_process_block): Likewise.
1556 * df-problems.c (df_word_lr_mark_ref): Likewise.
1557 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
1558 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
1559 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
1560 (rtl_for_decl_location): Likewise.
1561 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
1562 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
1563 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
1564 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
1565 (expand_expr_real_1): Likewise.
1566 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
1567 (pad_below): Likewise.
1568 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
1569 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
1570 * ira.c (get_subreg_tracking_sizes): Likewise.
1571 * ira-build.c (ira_create_allocno_objects): Likewise.
1572 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
1573 (ira_sort_regnos_for_alter_reg): Likewise.
1574 * ira-costs.c (record_operand_costs): Likewise.
1575 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
1576 (resolve_simple_move): Likewise.
1577 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
1578 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
1579 (lra_constraints): Likewise.
1580 (CONST_POOL_OK_P): Reject variable-sized modes.
1581 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
1582 (add_pseudo_to_slot, lra_spill): Likewise.
1583 * omp-low.c (omp_clause_aligned_alignment): Likewise.
1584 * optabs-query.c (get_best_extraction_insn): Likewise.
1585 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
1586 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
1587 (expand_mult_highpart, valid_multiword_target_p): Likewise.
1588 * recog.c (offsettable_address_addr_space_p): Likewise.
1589 * regcprop.c (maybe_mode_change): Likewise.
1590 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
1591 * regrename.c (build_def_use): Likewise.
1592 * regstat.c (dump_reg_info): Likewise.
1593 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
1594 (find_reloads, find_reloads_subreg_address): Likewise.
1595 * reload1.c (eliminate_regs_1): Likewise.
1596 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
1597 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
1598 (simplify_binary_operation_1, simplify_subreg): Likewise.
1599 * targhooks.c (default_function_arg_padding): Likewise.
1600 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
1601 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
1602 (verify_gimple_assign_ternary): Likewise.
1603 * tree-inline.c (estimate_move_cost): Likewise.
1604 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1605 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
1606 (get_address_cost_ainc): Likewise.
1607 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
1608 (vect_supportable_dr_alignment): Likewise.
1609 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
1610 (vectorizable_reduction): Likewise.
1611 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
1612 (vectorizable_operation, vectorizable_load): Likewise.
1613 * tree.c (build_same_sized_truth_vector_type): Likewise.
1614 * valtrack.c (cleanup_auto_inc_dec): Likewise.
1615 * var-tracking.c (emit_note_insn_var_location): Likewise.
1616 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
1617 (ADDR_VEC_ALIGN): Likewise.
1619 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1620 Alan Hayward <alan.hayward@arm.com>
1621 David Sherwood <david.sherwood@arm.com>
1623 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
1625 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
1626 or if measurement_type is polynomial.
1627 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
1628 * combine.c (make_extraction): Likewise.
1629 * dse.c (find_shift_sequence): Likewise.
1630 * dwarf2out.c (mem_loc_descriptor): Likewise.
1631 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
1632 (extract_bit_field, extract_low_bits): Likewise.
1633 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
1634 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
1635 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
1636 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
1637 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
1638 * reload.c (find_reloads): Likewise.
1639 * reload1.c (alter_reg): Likewise.
1640 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
1641 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
1642 * tree-if-conv.c (predicate_mem_writes): Likewise.
1643 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
1644 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
1645 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
1646 * valtrack.c (dead_debug_insert_temp): Likewise.
1647 * varasm.c (mergeable_constant_section): Likewise.
1648 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
1650 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1651 Alan Hayward <alan.hayward@arm.com>
1652 David Sherwood <david.sherwood@arm.com>
1654 * expr.c (expand_assignment): Cope with polynomial mode sizes
1655 when assigning to a CONCAT.
1657 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1658 Alan Hayward <alan.hayward@arm.com>
1659 David Sherwood <david.sherwood@arm.com>
1661 * machmode.h (mode_precision): Change from unsigned short to
1663 (mode_to_precision): Return a poly_uint16 rather than an unsigned
1665 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
1666 or if measurement_type is not polynomial.
1667 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
1668 in which the mode is already known to be a scalar_int_mode.
1669 * genmodes.c (emit_mode_precision): Change the type of mode_precision
1670 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
1672 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
1673 for GET_MODE_PRECISION.
1674 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
1675 for GET_MODE_PRECISION.
1676 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
1678 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
1679 (expand_field_assignment, make_extraction): Likewise.
1680 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
1681 (get_last_value): Likewise.
1682 * convert.c (convert_to_integer_1): Likewise.
1683 * cse.c (cse_insn): Likewise.
1684 * expr.c (expand_expr_real_1): Likewise.
1685 * lra-constraints.c (simplify_operand_subreg): Likewise.
1686 * optabs-query.c (can_atomic_load_p): Likewise.
1687 * optabs.c (expand_atomic_load): Likewise.
1688 (expand_atomic_store): Likewise.
1689 * ree.c (combine_reaching_defs): Likewise.
1690 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
1691 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
1692 * tree.h (type_has_mode_precision_p): Likewise.
1693 * ubsan.c (instrument_si_overflow): Likewise.
1695 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1696 Alan Hayward <alan.hayward@arm.com>
1697 David Sherwood <david.sherwood@arm.com>
1699 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
1700 polynomial numbers of units.
1701 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
1702 (valid_vector_subparts_p): New function.
1703 (build_vector_type): Remove temporary shim and take the number
1704 of units as a poly_uint64 rather than an int.
1705 (build_opaque_vector_type): Take the number of units as a
1706 poly_uint64 rather than an int.
1707 * tree.c (build_vector_from_ctor): Handle polynomial
1708 TYPE_VECTOR_SUBPARTS.
1709 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
1710 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
1711 (build_vector_from_val): If the number of units is variable,
1712 use build_vec_duplicate_cst for constant operands and
1713 VEC_DUPLICATE_EXPR otherwise.
1714 (make_vector_type): Remove temporary is_constant ().
1715 (build_vector_type, build_opaque_vector_type): Take the number of
1716 units as a poly_uint64 rather than an int.
1717 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
1719 * cfgexpand.c (expand_debug_expr): Likewise.
1720 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
1721 (store_constructor, expand_expr_real_1): Likewise.
1722 (const_scalar_mask_from_tree): Likewise.
1723 * fold-const-call.c (fold_const_reduction): Likewise.
1724 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
1725 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
1726 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
1727 (fold_relational_const): Likewise.
1728 (native_interpret_vector): Likewise. Change the size from an
1729 int to an unsigned int.
1730 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
1731 TYPE_VECTOR_SUBPARTS.
1732 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
1733 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
1734 duplicating a non-constant operand into a variable-length vector.
1735 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
1736 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
1737 * ipa-icf.c (sem_variable::equals): Likewise.
1738 * match.pd: Likewise.
1739 * omp-simd-clone.c (simd_clone_subparts): Likewise.
1740 * print-tree.c (print_node): Likewise.
1741 * stor-layout.c (layout_type): Likewise.
1742 * targhooks.c (default_builtin_vectorization_cost): Likewise.
1743 * tree-cfg.c (verify_gimple_comparison): Likewise.
1744 (verify_gimple_assign_binary): Likewise.
1745 (verify_gimple_assign_ternary): Likewise.
1746 (verify_gimple_assign_single): Likewise.
1747 * tree-pretty-print.c (dump_generic_node): Likewise.
1748 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1749 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
1750 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
1751 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
1752 (vect_shift_permute_load_chain): Likewise.
1753 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
1754 (expand_vector_condition, optimize_vector_constructor): Likewise.
1755 (lower_vec_perm, get_compute_type): Likewise.
1756 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
1757 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
1758 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
1759 (vect_recog_mask_conversion_pattern): Likewise.
1760 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
1761 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
1762 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1763 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
1764 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
1765 (vectorizable_shift, vectorizable_operation, vectorizable_store)
1766 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
1767 (supportable_widening_operation): Likewise.
1768 (supportable_narrowing_operation): Likewise.
1769 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
1771 * varasm.c (output_constant): Likewise.
1773 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1774 Alan Hayward <alan.hayward@arm.com>
1775 David Sherwood <david.sherwood@arm.com>
1777 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
1778 so that both the length == 3 and length != 3 cases set up their
1779 own permute vectors. Add comments explaining why we know the
1780 number of elements is constant.
1781 (vect_permute_load_chain): Likewise.
1783 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1784 Alan Hayward <alan.hayward@arm.com>
1785 David Sherwood <david.sherwood@arm.com>
1787 * machmode.h (mode_nunits): Change from unsigned char to
1789 (ONLY_FIXED_SIZE_MODES): New macro.
1790 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
1791 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
1792 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
1794 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
1795 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
1796 or if measurement_type is not polynomial.
1797 * genmodes.c (ZERO_COEFFS): New macro.
1798 (emit_mode_nunits_inline): Make mode_nunits_inline return a
1800 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
1801 Use ZERO_COEFFS when emitting initializers.
1802 * data-streamer.h (bp_pack_poly_value): New function.
1803 (bp_unpack_poly_value): Likewise.
1804 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
1805 for GET_MODE_NUNITS.
1806 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
1807 for GET_MODE_NUNITS.
1808 * tree.c (make_vector_type): Remove temporary shim and make
1809 the real function take the number of units as a poly_uint64
1811 (build_vector_type_for_mode): Handle polynomial nunits.
1812 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
1813 * emit-rtl.c (const_vec_series_p_1): Likewise.
1814 (gen_rtx_CONST_VECTOR): Likewise.
1815 * fold-const.c (test_vec_duplicate_folding): Likewise.
1816 * genrecog.c (validate_pattern): Likewise.
1817 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
1818 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
1819 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
1820 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
1821 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
1822 * rtlanal.c (subreg_get_info): Likewise.
1823 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1824 (vect_grouped_load_supported): Likewise.
1825 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
1826 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
1827 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
1828 (simplify_const_unary_operation, simplify_binary_operation_1)
1829 (simplify_const_binary_operation, simplify_ternary_operation)
1830 (test_vector_ops_duplicate, test_vector_ops): Likewise.
1831 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
1832 instead of CONST_VECTOR_NUNITS.
1833 * varasm.c (output_constant_pool_2): Likewise.
1834 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
1835 explicit-encoded elements in the XVEC for variable-length vectors.
1837 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1839 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
1841 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1842 Alan Hayward <alan.hayward@arm.com>
1843 David Sherwood <david.sherwood@arm.com>
1845 * coretypes.h (fixed_size_mode): Declare.
1846 (fixed_size_mode_pod): New typedef.
1847 * builtins.h (target_builtins::x_apply_args_mode)
1848 (target_builtins::x_apply_result_mode): Change type to
1849 fixed_size_mode_pod.
1850 * builtins.c (apply_args_size, apply_result_size, result_vector)
1851 (expand_builtin_apply_args_1, expand_builtin_apply)
1852 (expand_builtin_return): Update accordingly.
1854 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1856 * cse.c (hash_rtx_cb): Hash only the encoded elements.
1857 * cselib.c (cselib_hash_rtx): Likewise.
1858 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
1859 CONST_VECTOR encoding.
1861 2017-01-03 Jakub Jelinek <jakub@redhat.com>
1862 Jeff Law <law@redhat.com>
1865 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
1866 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
1867 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
1868 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
1871 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
1872 explicitly probe *sp in a noreturn function if there were any callee
1873 register saves or frame pointer is needed.
1875 2018-01-03 Jakub Jelinek <jakub@redhat.com>
1878 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
1879 BLKmode for ternary, binary or unary expressions.
1882 * var-tracking.c (delete_vta_debug_insn): New inline function.
1883 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
1884 insns from get_insns () to NULL instead of each bb separately.
1885 Use delete_vta_debug_insn. No longer static.
1886 (vt_debug_insns_local, variable_tracking_main_1): Adjust
1887 delete_vta_debug_insns callers.
1888 * rtl.h (delete_vta_debug_insns): Declare.
1889 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
1890 instead of variable_tracking_main.
1892 2018-01-03 Martin Sebor <msebor@redhat.com>
1894 PR tree-optimization/83603
1895 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
1896 arguments past the endof the argument list in functions declared
1897 without a prototype.
1898 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
1899 Avoid checking when arguments are null.
1901 2018-01-03 Martin Sebor <msebor@redhat.com>
1904 * doc/extend.texi (attribute const): Fix a typo.
1905 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
1906 issuing -Wsuggest-attribute for void functions.
1908 2018-01-03 Martin Sebor <msebor@redhat.com>
1910 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
1911 offset_int::from instead of wide_int::to_shwi.
1912 (maybe_diag_overlap): Remove assertion.
1913 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
1914 * gimple-ssa-sprintf.c (format_directive): Same.
1915 (parse_directive): Same.
1916 (sprintf_dom_walker::compute_format_length): Same.
1917 (try_substitute_return_value): Same.
1919 2017-01-03 Jeff Law <law@redhat.com>
1922 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
1923 non-constant residual for zero at runtime and avoid probing in
1924 that case. Reorganize code for trailing problem to mirror handling
1927 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
1929 PR tree-optimization/83501
1930 * tree-ssa-strlen.c (get_string_cst): New.
1931 (handle_char_store): Call get_string_cst.
1933 2018-01-03 Martin Liska <mliska@suse.cz>
1935 PR tree-optimization/83593
1936 * tree-ssa-strlen.c: Include tree-cfg.h.
1937 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
1938 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
1939 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
1941 (strlen_dom_walker::before_dom_children): Call
1942 gimple_purge_dead_eh_edges. Dump tranformation with details
1944 (strlen_dom_walker::before_dom_children): Update call by adding
1945 new argument cleanup_eh.
1946 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
1948 2018-01-03 Martin Liska <mliska@suse.cz>
1951 * cif-code.def (VARIADIC_THUNK): New enum value.
1952 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
1955 2018-01-03 Jan Beulich <jbeulich@suse.com>
1957 * sse.md (mov<mode>_internal): Tighten condition for when to use
1958 vmovdqu<ssescalarsize> for TI and OI modes.
1960 2018-01-03 Jakub Jelinek <jakub@redhat.com>
1962 Update copyright years.
1964 2018-01-03 Martin Liska <mliska@suse.cz>
1967 * ipa-visibility.c (function_and_variable_visibility): Skip
1968 functions with noipa attribure.
1970 2018-01-03 Jakub Jelinek <jakub@redhat.com>
1972 * gcc.c (process_command): Update copyright notice dates.
1973 * gcov-dump.c (print_version): Ditto.
1974 * gcov.c (print_version): Ditto.
1975 * gcov-tool.c (print_version): Ditto.
1976 * gengtype.c (create_file): Ditto.
1977 * doc/cpp.texi: Bump @copying's copyright year.
1978 * doc/cppinternals.texi: Ditto.
1979 * doc/gcc.texi: Ditto.
1980 * doc/gccint.texi: Ditto.
1981 * doc/gcov.texi: Ditto.
1982 * doc/install.texi: Ditto.
1983 * doc/invoke.texi: Ditto.
1985 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1987 * vector-builder.h (vector_builder::m_full_nelts): Change from
1988 unsigned int to poly_uint64.
1989 (vector_builder::full_nelts): Update prototype accordingly.
1990 (vector_builder::new_vector): Likewise.
1991 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
1992 (vector_builder::operator ==): Likewise.
1993 (vector_builder::finalize): Likewise.
1994 * int-vector-builder.h (int_vector_builder::int_vector_builder):
1995 Take the number of elements as a poly_uint64 rather than an
1997 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
1998 from unsigned int to poly_uint64.
1999 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
2000 (vec_perm_indices::new_vector): Likewise.
2001 (vec_perm_indices::length): Likewise.
2002 (vec_perm_indices::nelts_per_input): Likewise.
2003 (vec_perm_indices::input_nelts): Likewise.
2004 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
2005 number of elements per input as a poly_uint64 rather than an
2006 unsigned int. Use the original encoding for variable-length
2007 vectors, rather than clamping each individual element.
2008 For the second and subsequent elements in each pattern,
2009 clamp the step and base before clamping their sum.
2010 (vec_perm_indices::series_p): Handle polynomial element counts.
2011 (vec_perm_indices::all_in_range_p): Likewise.
2012 (vec_perm_indices_to_tree): Likewise.
2013 (vec_perm_indices_to_rtx): Likewise.
2014 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
2015 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
2016 (tree_vector_builder::new_binary_operation): Handle polynomial
2017 element counts. Return false if we need to know the number
2018 of elements at compile time.
2019 * fold-const.c (fold_vec_perm): Punt if the number of elements
2020 isn't known at compile time.
2022 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2024 * vec-perm-indices.h (vec_perm_builder): Change element type
2025 from HOST_WIDE_INT to poly_int64.
2026 (vec_perm_indices::element_type): Update accordingly.
2027 (vec_perm_indices::clamp): Handle polynomial element_types.
2028 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
2029 (vec_perm_indices::all_in_range_p): Likewise.
2030 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
2032 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
2033 polynomial vec_perm_indices element types.
2034 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
2035 * fold-const.c (fold_vec_perm): Likewise.
2036 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
2037 * tree-vect-generic.c (lower_vec_perm): Likewise.
2038 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
2039 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
2040 element type to HOST_WIDE_INT.
2042 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2043 Alan Hayward <alan.hayward@arm.com>
2044 David Sherwood <david.sherwood@arm.com>
2046 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
2047 rather than an int. Use plus_constant.
2048 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
2049 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
2051 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2052 Alan Hayward <alan.hayward@arm.com>
2053 David Sherwood <david.sherwood@arm.com>
2055 * calls.c (emit_call_1, expand_call): Change struct_value_size from
2056 a HOST_WIDE_INT to a poly_int64.
2058 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2059 Alan Hayward <alan.hayward@arm.com>
2060 David Sherwood <david.sherwood@arm.com>
2062 * calls.c (load_register_parameters): Cope with polynomial
2063 mode sizes. Require a constant size for BLKmode parameters
2064 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
2065 forces a parameter to be padded at the lsb end in order to
2066 fill a complete number of words, require the parameter size
2067 to be ordered wrt UNITS_PER_WORD.
2069 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2070 Alan Hayward <alan.hayward@arm.com>
2071 David Sherwood <david.sherwood@arm.com>
2073 * reload1.c (spill_stack_slot_width): Change element type
2074 from unsigned int to poly_uint64_pod.
2075 (alter_reg): Treat mode sizes as polynomial.
2077 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2078 Alan Hayward <alan.hayward@arm.com>
2079 David Sherwood <david.sherwood@arm.com>
2081 * reload.c (complex_word_subreg_p): New function.
2082 (reload_inner_reg_of_subreg, push_reload): Use it.
2084 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2085 Alan Hayward <alan.hayward@arm.com>
2086 David Sherwood <david.sherwood@arm.com>
2088 * lra-constraints.c (process_alt_operands): Reject matched
2089 operands whose sizes aren't ordered.
2090 (match_reload): Refer to this check here.
2092 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2093 Alan Hayward <alan.hayward@arm.com>
2094 David Sherwood <david.sherwood@arm.com>
2096 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
2097 that the mode size is in the set {1, 2, 4, 8, 16}.
2099 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2100 Alan Hayward <alan.hayward@arm.com>
2101 David Sherwood <david.sherwood@arm.com>
2103 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
2104 Use plus_constant instead of gen_rtx_PLUS.
2106 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2107 Alan Hayward <alan.hayward@arm.com>
2108 David Sherwood <david.sherwood@arm.com>
2110 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
2111 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
2112 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
2113 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
2114 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
2115 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
2116 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
2117 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
2118 * config/i386/i386.c (ix86_push_rounding): ...this new function.
2119 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
2121 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
2122 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
2123 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
2124 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
2125 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
2126 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
2127 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
2128 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
2129 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
2130 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
2132 * expr.c (emit_move_resolve_push): Treat the input and result
2133 of PUSH_ROUNDING as a poly_int64.
2134 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
2135 (emit_push_insn): Likewise.
2136 * lra-eliminations.c (mark_not_eliminable): Likewise.
2137 * recog.c (push_operand): Likewise.
2138 * reload1.c (elimination_effects): Likewise.
2139 * rtlanal.c (nonzero_bits1): Likewise.
2140 * calls.c (store_one_arg): Likewise. Require the padding to be
2141 known at compile time.
2143 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2144 Alan Hayward <alan.hayward@arm.com>
2145 David Sherwood <david.sherwood@arm.com>
2147 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
2148 Use plus_constant instead of gen_rtx_PLUS.
2150 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2151 Alan Hayward <alan.hayward@arm.com>
2152 David Sherwood <david.sherwood@arm.com>
2154 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
2157 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2158 Alan Hayward <alan.hayward@arm.com>
2159 David Sherwood <david.sherwood@arm.com>
2161 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
2162 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
2163 via stack temporaries. Treat the mode size as polynomial too.
2165 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2166 Alan Hayward <alan.hayward@arm.com>
2167 David Sherwood <david.sherwood@arm.com>
2169 * expr.c (expand_expr_real_2): When handling conversions involving
2170 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
2171 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
2172 as a poly_uint64 too.
2174 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2175 Alan Hayward <alan.hayward@arm.com>
2176 David Sherwood <david.sherwood@arm.com>
2178 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
2180 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2181 Alan Hayward <alan.hayward@arm.com>
2182 David Sherwood <david.sherwood@arm.com>
2184 * combine.c (can_change_dest_mode): Handle polynomial
2185 REGMODE_NATURAL_SIZE.
2186 * expmed.c (store_bit_field_1): Likewise.
2187 * expr.c (store_constructor): Likewise.
2188 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
2189 and polynomial REGMODE_NATURAL_SIZE.
2190 (gen_lowpart_common): Likewise.
2191 * reginfo.c (record_subregs_of_mode): Likewise.
2192 * rtlanal.c (read_modify_subreg_p): Likewise.
2194 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2195 Alan Hayward <alan.hayward@arm.com>
2196 David Sherwood <david.sherwood@arm.com>
2198 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
2199 numbers of elements.
2201 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2202 Alan Hayward <alan.hayward@arm.com>
2203 David Sherwood <david.sherwood@arm.com>
2205 * match.pd: Cope with polynomial numbers of vector elements.
2207 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2208 Alan Hayward <alan.hayward@arm.com>
2209 David Sherwood <david.sherwood@arm.com>
2211 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
2212 in a POINTER_PLUS_EXPR.
2214 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2215 Alan Hayward <alan.hayward@arm.com>
2216 David Sherwood <david.sherwood@arm.com>
2218 * omp-simd-clone.c (simd_clone_subparts): New function.
2219 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
2220 (ipa_simd_modify_function_body): Likewise.
2222 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2223 Alan Hayward <alan.hayward@arm.com>
2224 David Sherwood <david.sherwood@arm.com>
2226 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
2227 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
2228 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
2229 (expand_vector_condition, vector_element): Likewise.
2230 (subparts_gt): New function.
2231 (get_compute_type): Use subparts_gt.
2232 (count_type_subparts): Delete.
2233 (expand_vector_operations_1): Use subparts_gt instead of
2234 count_type_subparts.
2236 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2237 Alan Hayward <alan.hayward@arm.com>
2238 David Sherwood <david.sherwood@arm.com>
2240 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
2241 (vect_compile_time_alias): ...this new function. Do the calculation
2242 on poly_ints rather than trees.
2243 (vect_prune_runtime_alias_test_list): Update call accordingly.
2245 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2246 Alan Hayward <alan.hayward@arm.com>
2247 David Sherwood <david.sherwood@arm.com>
2249 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
2251 (vect_schedule_slp_instance): Likewise.
2253 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2254 Alan Hayward <alan.hayward@arm.com>
2255 David Sherwood <david.sherwood@arm.com>
2257 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
2258 constant and extern definitions for variable-length vectors.
2259 (vect_get_constant_vectors): Note that the number of units
2260 is known to be constant.
2262 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2263 Alan Hayward <alan.hayward@arm.com>
2264 David Sherwood <david.sherwood@arm.com>
2266 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
2267 of units as polynomial. Choose between WIDE and NARROW based
2270 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2271 Alan Hayward <alan.hayward@arm.com>
2272 David Sherwood <david.sherwood@arm.com>
2274 * tree-vect-stmts.c (simd_clone_subparts): New function.
2275 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
2277 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2278 Alan Hayward <alan.hayward@arm.com>
2279 David Sherwood <david.sherwood@arm.com>
2281 * tree-vect-stmts.c (vectorizable_call): Treat the number of
2282 vectors as polynomial. Use build_index_vector for
2285 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2286 Alan Hayward <alan.hayward@arm.com>
2287 David Sherwood <david.sherwood@arm.com>
2289 * tree-vect-stmts.c (get_load_store_type): Treat the number of
2290 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
2291 for variable-length vectors.
2292 (vectorizable_mask_load_store): Treat the number of units as
2293 polynomial, asserting that it is constant if the condition has
2294 already been enforced.
2295 (vectorizable_store, vectorizable_load): Likewise.
2297 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2298 Alan Hayward <alan.hayward@arm.com>
2299 David Sherwood <david.sherwood@arm.com>
2301 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
2302 of units as polynomial. Punt if we can't tell at compile time
2303 which vector contains the final result.
2305 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2306 Alan Hayward <alan.hayward@arm.com>
2307 David Sherwood <david.sherwood@arm.com>
2309 * tree-vect-loop.c (vectorizable_induction): Treat the number
2310 of units as polynomial. Punt on SLP inductions. Use an integer
2311 VEC_SERIES_EXPR for variable-length integer reductions. Use a
2312 cast of such a series for variable-length floating-point
2315 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2316 Alan Hayward <alan.hayward@arm.com>
2317 David Sherwood <david.sherwood@arm.com>
2319 * tree.h (build_index_vector): Declare.
2320 * tree.c (build_index_vector): New function.
2321 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
2322 of units as polynomial, forcibly converting it to a constant if
2323 vectorizable_reduction has already enforced the condition.
2324 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
2325 to create a {1,2,3,...} vector.
2326 (vectorizable_reduction): Treat the number of units as polynomial.
2327 Choose vectype_in based on the largest scalar element size rather
2328 than the smallest number of units. Enforce the restrictions
2331 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2332 Alan Hayward <alan.hayward@arm.com>
2333 David Sherwood <david.sherwood@arm.com>
2335 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
2336 number of units as polynomial.
2338 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2339 Alan Hayward <alan.hayward@arm.com>
2340 David Sherwood <david.sherwood@arm.com>
2342 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
2343 * target.def (autovectorize_vector_sizes): Return the vector sizes
2344 by pointer, using vector_sizes rather than a bitmask.
2345 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
2346 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
2347 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
2349 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
2350 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
2351 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
2352 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
2353 * omp-general.c (omp_max_vf): Likewise.
2354 * omp-low.c (omp_clause_aligned_alignment): Likewise.
2355 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
2356 * tree-vect-loop.c (vect_analyze_loop): Likewise.
2357 * tree-vect-slp.c (vect_slp_bb): Likewise.
2358 * doc/tm.texi: Regenerate.
2359 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
2361 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
2362 the vector size as a poly_uint64 rather than an unsigned int.
2363 (current_vector_size): Change from an unsigned int to a poly_uint64.
2364 (get_vectype_for_scalar_type): Update accordingly.
2365 * tree.h (build_truth_vector_type): Take the size and number of
2366 units as a poly_uint64 rather than an unsigned int.
2367 (build_vector_type): Add a temporary overload that takes
2368 the number of units as a poly_uint64 rather than an unsigned int.
2369 * tree.c (make_vector_type): Likewise.
2370 (build_truth_vector_type): Take the number of units as a poly_uint64
2371 rather than an unsigned int.
2373 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2374 Alan Hayward <alan.hayward@arm.com>
2375 David Sherwood <david.sherwood@arm.com>
2377 * target.def (get_mask_mode): Take the number of units and length
2378 as poly_uint64s rather than unsigned ints.
2379 * targhooks.h (default_get_mask_mode): Update accordingly.
2380 * targhooks.c (default_get_mask_mode): Likewise.
2381 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
2382 * doc/tm.texi: Regenerate.
2384 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2385 Alan Hayward <alan.hayward@arm.com>
2386 David Sherwood <david.sherwood@arm.com>
2388 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
2389 * omp-general.c (omp_max_vf): Likewise.
2390 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
2391 (expand_omp_simd): Handle polynomial safelen.
2392 * omp-low.c (omplow_simd_context): Add a default constructor.
2393 (omplow_simd_context::max_vf): Change from int to poly_uint64.
2394 (lower_rec_simd_input_clauses): Update accordingly.
2395 (lower_rec_input_clauses): Likewise.
2397 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2398 Alan Hayward <alan.hayward@arm.com>
2399 David Sherwood <david.sherwood@arm.com>
2401 * tree-vectorizer.h (vect_nunits_for_cost): New function.
2402 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
2403 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
2404 (vect_analyze_slp_cost): Likewise.
2405 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
2406 (vect_model_load_cost): Likewise.
2408 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2409 Alan Hayward <alan.hayward@arm.com>
2410 David Sherwood <david.sherwood@arm.com>
2412 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
2413 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
2414 from an unsigned int * to a poly_uint64_pod *.
2415 (calculate_unrolling_factor): New function.
2416 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
2418 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2419 Alan Hayward <alan.hayward@arm.com>
2420 David Sherwood <david.sherwood@arm.com>
2422 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
2423 from an unsigned int to a poly_uint64.
2424 (_loop_vec_info::slp_unrolling_factor): Likewise.
2425 (_loop_vec_info::vectorization_factor): Change from an int
2427 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
2428 (vect_get_num_vectors): New function.
2429 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
2430 (vect_get_num_copies): Use vect_get_num_vectors.
2431 (vect_analyze_data_ref_dependences): Change max_vf from an int *
2432 to an unsigned int *.
2433 (vect_analyze_data_refs): Change min_vf from an int * to a
2435 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
2436 than an unsigned HOST_WIDE_INT.
2437 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
2438 (vect_analyze_data_ref_dependence): Change max_vf from an int *
2439 to an unsigned int *.
2440 (vect_analyze_data_ref_dependences): Likewise.
2441 (vect_compute_data_ref_alignment): Handle polynomial vf.
2442 (vect_enhance_data_refs_alignment): Likewise.
2443 (vect_prune_runtime_alias_test_list): Likewise.
2444 (vect_shift_permute_load_chain): Likewise.
2445 (vect_supportable_dr_alignment): Likewise.
2446 (dependence_distance_ge_vf): Take the vectorization factor as a
2447 poly_uint64 rather than an unsigned HOST_WIDE_INT.
2448 (vect_analyze_data_refs): Change min_vf from an int * to a
2450 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
2451 vfm1 as a poly_uint64 rather than an int. Make the same change
2452 for the returned bound_scalar.
2453 (vect_gen_vector_loop_niters): Handle polynomial vf.
2454 (vect_do_peeling): Likewise. Update call to
2455 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
2456 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
2458 * tree-vect-loop.c (vect_determine_vectorization_factor)
2459 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
2460 (vect_get_known_peeling_cost): Likewise.
2461 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
2462 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
2463 (vect_transform_loop): Likewise. Use the lowest possible VF when
2464 updating the upper bounds of the loop.
2465 (vect_min_worthwhile_factor): Make static. Return an unsigned int
2467 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
2468 polynomial unroll factors.
2469 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
2470 (vect_make_slp_decision): Likewise.
2471 (vect_supported_load_permutation_p): Likewise, and polynomial
2473 (vect_analyze_slp_cost): Handle polynomial vf.
2474 (vect_slp_analyze_node_operations): Likewise.
2475 (vect_slp_analyze_bb_1): Likewise.
2476 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
2477 than an unsigned HOST_WIDE_INT.
2478 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
2479 (vectorizable_load): Handle polynomial vf.
2480 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
2482 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
2484 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2485 Alan Hayward <alan.hayward@arm.com>
2486 David Sherwood <david.sherwood@arm.com>
2488 * match.pd: Handle bit operations involving three constants
2489 and try to fold one pair.
2491 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2493 * tree-vect-loop-manip.c: Include gimple-fold.h.
2494 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
2495 niters_maybe_zero parameters. Handle other cases besides a step of 1.
2496 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
2497 Add a path that uses a step of VF instead of 1, but disable it
2499 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
2500 and niters_no_overflow parameters. Update calls to
2501 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
2502 Create a new SSA name if the latter choses to use a ste other
2503 than zero, and return it via niters_vector_mult_vf_var.
2504 * tree-vect-loop.c (vect_transform_loop): Update calls to
2505 vect_do_peeling, vect_gen_vector_loop_niters and
2506 slpeel_make_loop_iterate_ntimes.
2507 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
2508 (vect_gen_vector_loop_niters): Update declarations after above changes.
2510 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
2512 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
2513 128-bit round to integer instructions.
2514 (ceil<mode>2): Likewise.
2515 (btrunc<mode>2): Likewise.
2516 (round<mode>2): Likewise.
2518 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2520 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
2521 unaligned VSX load/store on P8/P9.
2522 (expand_block_clear): Allow the use of unaligned VSX
2523 load/store on P8/P9.
2525 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2527 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
2529 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
2530 swap associated with both a load and a store.
2532 2018-01-02 Andrew Waterman <andrew@sifive.com>
2534 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
2535 * config/riscv/riscv.md (clear_cache): Use it.
2537 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
2539 * web.c: Remove out-of-date comment.
2541 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2543 * expr.c (fixup_args_size_notes): Check that any existing
2544 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
2545 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
2546 (emit_single_push_insn): ...here.
2548 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2550 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
2551 (const_vector_encoded_nelts): New function.
2552 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
2553 (const_vector_int_elt, const_vector_elt): Declare.
2554 * emit-rtl.c (const_vector_int_elt_1): New function.
2555 (const_vector_elt): Likewise.
2556 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
2557 of CONST_VECTOR_ELT.
2559 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2561 * expr.c: Include rtx-vector-builder.h.
2562 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
2563 directly on the tree encoding.
2564 (const_vector_from_tree): Likewise.
2565 * optabs.c: Include rtx-vector-builder.h.
2566 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
2567 sequence of "u" values.
2568 * vec-perm-indices.c: Include rtx-vector-builder.h.
2569 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
2570 directly on the vec_perm_indices encoding.
2572 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2574 * doc/rtl.texi (const_vector): Describe new encoding scheme.
2575 * Makefile.in (OBJS): Add rtx-vector-builder.o.
2576 * rtx-vector-builder.h: New file.
2577 * rtx-vector-builder.c: Likewise.
2578 * rtl.h (rtx_def::u2): Add a const_vector field.
2579 (CONST_VECTOR_NPATTERNS): New macro.
2580 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
2581 (CONST_VECTOR_DUPLICATE_P): Likewise.
2582 (CONST_VECTOR_STEPPED_P): Likewise.
2583 (CONST_VECTOR_ENCODED_ELT): Likewise.
2584 (const_vec_duplicate_p): Check for a duplicated vector encoding.
2585 (unwrap_const_vec_duplicate): Likewise.
2586 (const_vec_series_p): Check for a non-duplicated vector encoding.
2587 Say that the function only returns true for integer vectors.
2588 * emit-rtl.c: Include rtx-vector-builder.h.
2589 (gen_const_vec_duplicate_1): Delete.
2590 (gen_const_vector): Call gen_const_vec_duplicate instead of
2591 gen_const_vec_duplicate_1.
2592 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
2593 (gen_const_vec_duplicate): Use rtx_vector_builder.
2594 (gen_const_vec_series): Likewise.
2595 (gen_rtx_CONST_VECTOR): Likewise.
2596 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
2597 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
2598 Build a new vector rather than modifying a CONST_VECTOR in-place.
2599 (handle_special_swappables): Update call accordingly.
2600 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
2601 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
2602 Build a new vector rather than modifying a CONST_VECTOR in-place.
2603 (handle_special_swappables): Update call accordingly.
2605 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2607 * simplify-rtx.c (simplify_const_binary_operation): Use
2608 CONST_VECTOR_ELT instead of XVECEXP.
2610 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2612 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
2613 the selector elements to be different from the data elements
2614 if the selector is a VECTOR_CST.
2615 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
2616 ssizetype for the selector.
2618 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2620 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
2621 before testing each element individually.
2622 * tree-vect-generic.c (lower_vec_perm): Likewise.
2624 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2626 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
2627 * selftest-run-tests.c (selftest::run_tests): Call it.
2628 * vector-builder.h (vector_builder::operator ==): New function.
2629 (vector_builder::operator !=): Likewise.
2630 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
2631 (vec_perm_indices::all_from_input_p): New function.
2632 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
2633 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
2634 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
2635 instead of reading the VECTOR_CST directly. Detect whether both
2636 vector inputs are the same before constructing the vec_perm_indices,
2637 and update the number of inputs argument accordingly. Use the
2638 utility functions added above. Only construct sel2 if we need to.
2640 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2642 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
2643 the broadcast of the low byte.
2644 (expand_mult_highpart): Use an explicit encoding for the permutes.
2645 * optabs-query.c (can_mult_highpart_p): Likewise.
2646 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
2647 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
2648 (vectorizable_bswap): Likewise.
2649 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
2650 explicit encoding for the power-of-2 permutes.
2651 (vect_permute_store_chain): Likewise.
2652 (vect_grouped_load_supported): Likewise.
2653 (vect_permute_load_chain): Likewise.
2655 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2657 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
2658 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
2659 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
2660 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
2661 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
2662 (vect_gen_perm_mask_any): Likewise.
2664 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2666 * int-vector-builder.h: New file.
2667 * vec-perm-indices.h: Include int-vector-builder.h.
2668 (vec_perm_indices): Redefine as an int_vector_builder.
2669 (auto_vec_perm_indices): Delete.
2670 (vec_perm_builder): Redefine as a stand-alone class.
2671 (vec_perm_indices::vec_perm_indices): New function.
2672 (vec_perm_indices::clamp): Likewise.
2673 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
2674 (vec_perm_indices::new_vector): New function.
2675 (vec_perm_indices::new_expanded_vector): Update for new
2676 vec_perm_indices class.
2677 (vec_perm_indices::rotate_inputs): New function.
2678 (vec_perm_indices::all_in_range_p): Operate directly on the
2679 encoded form, without computing elided elements.
2680 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
2681 encoding. Update for new vec_perm_indices class.
2682 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
2683 the given vec_perm_builder.
2684 (expand_vec_perm_var): Update vec_perm_builder constructor.
2685 (expand_mult_highpart): Use vec_perm_builder instead of
2686 auto_vec_perm_indices.
2687 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
2688 vec_perm_indices instead of auto_vec_perm_indices. Use a single
2689 or double series encoding as appropriate.
2690 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
2691 vec_perm_indices instead of auto_vec_perm_indices.
2692 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
2693 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
2694 (vect_permute_store_chain): Likewise.
2695 (vect_grouped_load_supported): Likewise.
2696 (vect_permute_load_chain): Likewise.
2697 (vect_shift_permute_load_chain): Likewise.
2698 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
2699 (vect_transform_slp_perm_load): Likewise.
2700 (vect_schedule_slp_instance): Likewise.
2701 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
2702 (vectorizable_mask_load_store): Likewise.
2703 (vectorizable_bswap): Likewise.
2704 (vectorizable_store): Likewise.
2705 (vectorizable_load): Likewise.
2706 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
2707 vec_perm_indices instead of auto_vec_perm_indices. Use
2708 tree_to_vec_perm_builder to read the vector from a tree.
2709 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
2710 vec_perm_builder instead of a vec_perm_indices.
2711 (have_whole_vector_shift): Use vec_perm_builder and
2712 vec_perm_indices instead of auto_vec_perm_indices. Leave the
2713 truncation to calc_vec_perm_mask_for_shift.
2714 (vect_create_epilog_for_reduction): Likewise.
2715 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
2716 from auto_vec_perm_indices to vec_perm_indices.
2717 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
2718 instead of changing individual elements.
2719 (aarch64_vectorize_vec_perm_const): Use new_vector to install
2720 the vector in d.perm.
2721 * config/arm/arm.c (expand_vec_perm_d::perm): Change
2722 from auto_vec_perm_indices to vec_perm_indices.
2723 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
2724 instead of changing individual elements.
2725 (arm_vectorize_vec_perm_const): Use new_vector to install
2726 the vector in d.perm.
2727 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
2728 Update vec_perm_builder constructor.
2729 (rs6000_expand_interleave): Likewise.
2730 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
2731 (rs6000_expand_interleave): Likewise.
2733 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2735 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
2736 to qimode could truncate the indices.
2737 * optabs.c (expand_vec_perm_var): Likewise.
2739 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2741 * Makefile.in (OBJS): Add vec-perm-indices.o.
2742 * vec-perm-indices.h: New file.
2743 * vec-perm-indices.c: Likewise.
2744 * target.h (vec_perm_indices): Replace with a forward class
2746 (auto_vec_perm_indices): Move to vec-perm-indices.h.
2747 * optabs.h: Include vec-perm-indices.h.
2748 (expand_vec_perm): Delete.
2749 (selector_fits_mode_p, expand_vec_perm_var): Declare.
2750 (expand_vec_perm_const): Declare.
2751 * target.def (vec_perm_const_ok): Replace with...
2752 (vec_perm_const): ...this new hook.
2753 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
2754 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
2755 * doc/tm.texi: Regenerate.
2756 * optabs.def (vec_perm_const): Delete.
2757 * doc/md.texi (vec_perm_const): Likewise.
2758 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
2759 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
2760 expand_vec_perm for constant permutation vectors. Assert that
2761 the mode of variable permutation vectors is the integer equivalent
2762 of the mode that is being permuted.
2763 * optabs-query.h (selector_fits_mode_p): Declare.
2764 * optabs-query.c: Include vec-perm-indices.h.
2765 (selector_fits_mode_p): New function.
2766 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
2767 is defined, instead of checking whether the vec_perm_const_optab
2768 exists. Use targetm.vectorize.vec_perm_const instead of
2769 targetm.vectorize.vec_perm_const_ok. Check whether the indices
2770 fit in the vector mode before using a variable permute.
2771 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
2772 vec_perm_indices instead of an rtx.
2773 (expand_vec_perm): Replace with...
2774 (expand_vec_perm_const): ...this new function. Take the selector
2775 as a vec_perm_indices rather than an rtx. Also take the mode of
2776 the selector. Update call to shift_amt_for_vec_perm_mask.
2777 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
2778 Use vec_perm_indices::new_expanded_vector to expand the original
2779 selector into bytes. Check whether the indices fit in the vector
2780 mode before using a variable permute.
2781 (expand_vec_perm_var): Make global.
2782 (expand_mult_highpart): Use expand_vec_perm_const.
2783 * fold-const.c: Includes vec-perm-indices.h.
2784 * tree-ssa-forwprop.c: Likewise.
2785 * tree-vect-data-refs.c: Likewise.
2786 * tree-vect-generic.c: Likewise.
2787 * tree-vect-loop.c: Likewise.
2788 * tree-vect-slp.c: Likewise.
2789 * tree-vect-stmts.c: Likewise.
2790 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
2792 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
2793 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
2794 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
2795 (aarch64_vectorize_vec_perm_const): ...this new function.
2796 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
2797 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2798 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
2799 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
2800 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
2801 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2802 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
2804 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
2805 check for NEON modes.
2806 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
2807 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
2808 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
2809 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
2811 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
2812 the old VEC_PERM_CONST conditions.
2813 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
2814 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
2815 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
2816 (ia64_vectorize_vec_perm_const_ok): Merge into...
2817 (ia64_vectorize_vec_perm_const): ...this new function.
2818 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
2819 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
2820 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
2821 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
2822 * config/mips/mips.c (mips_expand_vec_perm_const)
2823 (mips_vectorize_vec_perm_const_ok): Merge into...
2824 (mips_vectorize_vec_perm_const): ...this new function.
2825 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
2826 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
2827 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
2828 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
2829 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
2830 (rs6000_expand_vec_perm_const): Delete.
2831 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
2833 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2834 (altivec_expand_vec_perm_const_le): Take each operand individually.
2835 Operate on constant selectors rather than rtxes.
2836 (altivec_expand_vec_perm_const): Likewise. Update call to
2837 altivec_expand_vec_perm_const_le.
2838 (rs6000_expand_vec_perm_const): Delete.
2839 (rs6000_vectorize_vec_perm_const_ok): Delete.
2840 (rs6000_vectorize_vec_perm_const): New function.
2841 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
2842 an element count and rtx array.
2843 (rs6000_expand_extract_even): Update call accordingly.
2844 (rs6000_expand_interleave): Likewise.
2845 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
2846 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
2847 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
2848 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
2849 (rs6000_expand_vec_perm_const): Delete.
2850 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
2851 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2852 (altivec_expand_vec_perm_const_le): Take each operand individually.
2853 Operate on constant selectors rather than rtxes.
2854 (altivec_expand_vec_perm_const): Likewise. Update call to
2855 altivec_expand_vec_perm_const_le.
2856 (rs6000_expand_vec_perm_const): Delete.
2857 (rs6000_vectorize_vec_perm_const_ok): Delete.
2858 (rs6000_vectorize_vec_perm_const): New function. Remove stray
2859 reference to the SPE evmerge intructions.
2860 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
2861 an element count and rtx array.
2862 (rs6000_expand_extract_even): Update call accordingly.
2863 (rs6000_expand_interleave): Likewise.
2864 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
2865 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
2867 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2869 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2871 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
2872 vector mode and that that mode matches the mode of the data
2874 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
2875 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
2876 directly using expand_vec_perm_1 when forcing selectors into
2878 (expand_vec_perm_var): New function, split out from expand_vec_perm.
2880 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2882 * optabs-query.h (can_vec_perm_p): Delete.
2883 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
2884 * optabs-query.c (can_vec_perm_p): Split into...
2885 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
2886 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
2887 particular selector is valid.
2888 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
2889 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
2890 (vect_grouped_load_supported): Likewise.
2891 (vect_shift_permute_load_chain): Likewise.
2892 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
2893 (vect_transform_slp_perm_load): Likewise.
2894 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
2895 (vectorizable_bswap): Likewise.
2896 (vect_gen_perm_mask_checked): Likewise.
2897 * fold-const.c (fold_ternary_loc): Likewise. Don't take
2898 implementations of variable permutation vectors into account
2899 when deciding which selector to use.
2900 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
2901 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
2902 with a false third argument.
2903 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
2904 to test whether the constant selector is valid and can_vec_perm_var_p
2905 to test whether a variable selector is valid.
2907 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2909 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
2910 * optabs-query.c (can_vec_perm_p): Likewise.
2911 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
2912 instead of vec_perm_indices.
2913 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
2914 (vect_gen_perm_mask_checked): Likewise,
2915 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
2916 (vect_gen_perm_mask_checked): Likewise,
2918 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2920 * optabs-query.h (qimode_for_vec_perm): Declare.
2921 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
2922 (qimode_for_vec_perm): ...this new function.
2923 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
2925 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2927 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
2928 does not have a conditional at the top.
2930 2018-01-02 Richard Biener <rguenther@suse.de>
2932 * ipa-inline.c (big_speedup_p): Fix expression.
2934 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
2937 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
2940 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
2944 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
2945 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
2946 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
2947 cond_taken_branch_cost 3->4.
2949 2018-01-01 Jakub Jelinek <jakub@redhat.com>
2951 PR tree-optimization/83581
2952 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
2953 TODO_cleanup_cfg if any changes have been made.
2956 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
2957 convert_modes if target mode has the right side, but different mode
2961 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
2962 last argument when extracting from CONCAT. If either from_real or
2963 from_imag is NULL, use expansion through memory. If result is not
2964 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
2965 the parts directly to inner mode, if even that fails, use expansion
2969 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
2970 check for bswap in mode rather than HImode and use that in expand_unop
2973 Copyright (C) 2018 Free Software Foundation, Inc.
2975 Copying and distribution of this file, with or without modification,
2976 are permitted in any medium without royalty provided the copyright
2977 notice and this notice are preserved.