1 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2 Alan Hayward <alan.hayward@arm.com>
3 David Sherwood <david.sherwood@arm.com>
5 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
6 (add_offset_to_base): New function, split out from...
7 (create_mem_ref): ...here. When handling a scale other than 1,
8 check first whether the address is valid without the offset.
9 Add it into the base if so, leaving the index and scale as-is.
11 2018-01-12 Jakub Jelinek <jakub@redhat.com>
14 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
15 fold_for_warn before checking if arg2 is INTEGER_CST.
17 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
19 * config/rs6000/predicates.md (load_multiple_operation): Delete.
20 (store_multiple_operation): Delete.
21 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
22 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
23 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
24 guarded by TARGET_STRING.
25 (rs6000_output_load_multiple): Delete.
26 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
27 OPTION_MASK_STRING / TARGET_STRING handling.
28 (print_operand) <'N', 'O'>: Add comment that these are unused now.
29 (const rs6000_opt_masks) <"string">: Change mask to 0.
30 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
31 (MASK_STRING): Delete.
32 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
34 (load_multiple): Delete.
41 (store_multiple): Delete.
48 (movmemsi_8reg): Delete.
49 (corresponding unnamed define_insn): Delete.
50 (movmemsi_6reg): Delete.
51 (corresponding unnamed define_insn): Delete.
52 (movmemsi_4reg): Delete.
53 (corresponding unnamed define_insn): Delete.
54 (movmemsi_2reg): Delete.
55 (corresponding unnamed define_insn): Delete.
56 (movmemsi_1reg): Delete.
57 (corresponding unnamed define_insn): Delete.
58 * config/rs6000/rs6000.opt (mno-string): New.
59 (mstring): Replace by deprecation warning stub.
60 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
62 2018-01-12 Jakub Jelinek <jakub@redhat.com>
64 * regrename.c (regrename_do_replace): If replacing the same
65 reg multiple times, try to reuse last created gen_raw_REG.
68 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
69 main to workaround a bug in GDB.
71 2018-01-12 Tom de Vries <tom@codesourcery.com>
74 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
76 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
78 PR rtl-optimization/80481
79 * ira-color.c (get_cap_member): New function.
80 (allocnos_conflict_by_live_ranges_p): Use it.
81 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
82 (setup_slot_coalesced_allocno_live_ranges): Ditto.
84 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
87 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
92 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
94 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
95 rather than wi::to_widest for DR_INITs.
96 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
97 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
98 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
100 (vect_analyze_group_access_1): Note that here.
102 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
104 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
105 polynomial type sizes.
107 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
109 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
110 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
111 (gimple_add_tmp_var): Likewise.
113 2018-01-12 Martin Liska <mliska@suse.cz>
115 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
116 (gimple_alloc_sizes): Likewise.
117 (dump_gimple_statistics): Use PRIu64 in printf format.
118 * gimple.h: Change uint64_t to int.
120 2018-01-12 Martin Liska <mliska@suse.cz>
122 * tree-core.h: Use uint64_t instead of int.
123 * tree.c (tree_node_counts): Likewise.
124 (tree_node_sizes): Likewise.
125 (dump_tree_statistics): Use PRIu64 in printf format.
127 2018-01-12 Martin Liska <mliska@suse.cz>
129 * Makefile.in: As qsort_chk is implemented in vec.c, add
130 vec.o to linkage of gencfn-macros.
131 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
132 passing the info to record_node_allocation_statistics.
133 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
135 * ggc-common.c (struct ggc_usage): Add operator== and use
136 it in operator< and compare function.
137 * mem-stats.h (struct mem_usage): Likewise.
138 * vec.c (struct vec_usage): Remove operator< and compare
139 function. Can be simply inherited.
141 2018-01-12 Martin Jambor <mjambor@suse.cz>
144 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
145 * tree-ssa-math-opts.c: Include domwalk.h.
146 (convert_mult_to_fma_1): New function.
147 (fma_transformation_info): New type.
148 (fma_deferring_state): Likewise.
149 (cancel_fma_deferring): New function.
150 (result_of_phi): Likewise.
151 (last_fma_candidate_feeds_initial_phi): Likewise.
152 (convert_mult_to_fma): Added deferring logic, split actual
153 transformation to convert_mult_to_fma_1.
154 (math_opts_dom_walker): New type.
155 (math_opts_dom_walker::after_dom_children): New method, body moved
156 here from pass_optimize_widening_mul::execute, added deferring logic
158 (pass_optimize_widening_mul::execute): Moved most of code to
159 math_opts_dom_walker::after_dom_children.
160 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
161 * config/i386/i386.c (ix86_option_override_internal): Added
162 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
164 2018-01-12 Richard Biener <rguenther@suse.de>
167 * dwarf2out.c (gen_variable_die): Do not reset old_die for
168 inline instance vars.
170 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
173 * config/rx/rx.c (rx_is_restricted_memory_address):
176 2018-01-12 Richard Biener <rguenther@suse.de>
178 PR tree-optimization/80846
179 * target.def (split_reduction): New target hook.
180 * targhooks.c (default_split_reduction): New function.
181 * targhooks.h (default_split_reduction): Declare.
182 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
183 target requests first reduce vectors by combining low and high
185 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
186 (get_vectype_for_scalar_type_and_size): Export.
187 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
188 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
189 * doc/tm.texi: Regenerate.
190 * config/i386/i386.c (ix86_split_reduction): Implement
191 TARGET_VECTORIZE_SPLIT_REDUCTION.
193 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
196 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
197 in PIC mode except for TARGET_VXWORKS_RTP.
198 * config/sparc/sparc.c: Include cfgrtl.h.
199 (TARGET_INIT_PIC_REG): Define.
200 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
201 (sparc_pic_register_p): New predicate.
202 (sparc_legitimate_address_p): Use it.
203 (sparc_legitimize_pic_address): Likewise.
204 (sparc_delegitimize_address): Likewise.
205 (sparc_mode_dependent_address_p): Likewise.
206 (gen_load_pcrel_sym): Remove 4th parameter.
207 (load_got_register): Adjust call to above. Remove obsolete stuff.
208 (sparc_expand_prologue): Do not call load_got_register here.
209 (sparc_flat_expand_prologue): Likewise.
210 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
211 (sparc_use_pseudo_pic_reg): New function.
212 (sparc_init_pic_reg): Likewise.
213 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
214 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
216 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
218 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
219 Add item for branch_cost.
221 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
223 PR rtl-optimization/83565
224 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
225 not extend the result to a larger mode for rotate operations.
226 (num_sign_bit_copies1): Likewise.
228 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
231 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
233 Use values-Xc.o for -pedantic.
234 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
236 2018-01-12 Martin Liska <mliska@suse.cz>
239 * ipa-devirt.c (final_warning_record::grow_type_warnings):
241 (possible_polymorphic_call_targets): Use it.
242 (ipa_devirt): Likewise.
244 2018-01-12 Martin Liska <mliska@suse.cz>
246 * profile-count.h (enum profile_quality): Use 0 as invalid
247 enum value of profile_quality.
249 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
251 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
252 -mext-string options.
254 2018-01-12 Richard Biener <rguenther@suse.de>
256 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
257 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
258 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
260 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
262 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
264 * configure.ac (--with-long-double-format): Add support for the
265 configuration option to change the default long double format on
267 * config.gcc (powerpc*-linux*-*): Likewise.
268 * configure: Regenerate.
269 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
270 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
271 used without modification.
273 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
275 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
276 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
277 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
278 MISC_BUILTIN_SPEC_BARRIER.
279 (rs6000_init_builtins): Likewise.
280 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
282 (speculation_barrier): New define_insn.
283 * doc/extend.texi: Document __builtin_speculation_barrier.
285 2018-01-11 Jakub Jelinek <jakub@redhat.com>
288 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
289 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
290 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
292 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
293 integral modes instead of "ss" and "sd".
294 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
295 vectors with 32-bit and 64-bit elements.
296 (vecdupssescalarmodesuffix): New mode attribute.
297 (vec_dup<mode>): Use it.
299 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
302 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
303 frame if argument is passed on stack.
305 2018-01-11 Jakub Jelinek <jakub@redhat.com>
308 * ree.c (combine_reaching_defs): Optimize also
309 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
310 reg2=any_extend(exp); reg1=reg2;, formatting fix.
312 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
315 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
317 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
320 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
321 after they are computed.
323 2018-01-11 Bin Cheng <bin.cheng@arm.com>
325 PR tree-optimization/83695
326 * gimple-loop-linterchange.cc
327 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
328 reset cached scev information after interchange.
329 (pass_linterchange::execute): Remove call to scev_reset_htab.
331 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
333 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
334 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
335 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
336 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
337 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
338 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
339 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
340 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
341 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
342 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
343 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
344 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
345 (V_lane_reg): Likewise.
346 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
348 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
349 (vfmal_lane_low<mode>_intrinsic,
350 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
351 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
352 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
353 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
354 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
355 vfmsl_lane_high<mode>_intrinsic): New define_insns.
357 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
359 * config/arm/arm-cpus.in (fp16fml): New feature.
360 (ALL_SIMD): Add fp16fml.
361 (armv8.2-a): Add fp16fml as an option.
362 (armv8.3-a): Likewise.
363 (armv8.4-a): Add fp16fml as part of fp16.
364 * config/arm/arm.h (TARGET_FP16FML): Define.
365 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
367 * config/arm/arm-modes.def (V2HF): Define.
368 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
369 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
370 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
371 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
372 vfmsl_low, vfmsl_high): New set of builtins.
373 * config/arm/iterators.md (PLUSMINUS): New code iterator.
374 (vfml_op): New code attribute.
375 (VFMLHALVES): New int iterator.
376 (VFML, VFMLSEL): New mode attributes.
377 (V_reg): Define mapping for V2HF.
378 (V_hi, V_lo): New mode attributes.
379 (VF_constraint): Likewise.
380 (vfml_half, vfml_half_selector): New int attributes.
381 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
383 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
384 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
386 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
387 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
388 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
389 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
391 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
392 Document new effective target and option set.
394 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
396 * config/arm/arm-cpus.in (armv8_4): New feature.
397 (ARMv8_4a): New fgroup.
398 (armv8.4-a): New arch.
399 * config/arm/arm-tables.opt: Regenerate.
400 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
401 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
402 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
403 Add matching rules for -march=armv8.4-a and extensions.
404 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
406 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
409 * config/rx/rx.md (BW): New mode attribute.
410 (sync_lock_test_and_setsi): Add mode suffix to insn output.
412 2018-01-11 Richard Biener <rguenther@suse.de>
414 PR tree-optimization/83435
415 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
416 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
417 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
419 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
420 Alan Hayward <alan.hayward@arm.com>
421 David Sherwood <david.sherwood@arm.com>
423 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
425 (aarch64_classify_address): Initialize it. Track polynomial offsets.
426 (aarch64_print_address_internal): Use it to check for a zero offset.
428 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
429 Alan Hayward <alan.hayward@arm.com>
430 David Sherwood <david.sherwood@arm.com>
432 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
433 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
434 Return a poly_int64 rather than a HOST_WIDE_INT.
435 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
436 rather than a HOST_WIDE_INT.
437 * config/aarch64/aarch64.h (aarch64_frame): Protect with
438 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
439 hard_fp_offset, frame_size, initial_adjust, callee_offset and
440 final_offset from HOST_WIDE_INT to poly_int64.
441 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
442 to_constant when getting the number of units in an Advanced SIMD
444 (aarch64_builtin_vectorized_function): Check for a constant number
446 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
448 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
449 attribute instead of GET_MODE_NUNITS.
450 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
451 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
452 GET_MODE_SIZE for fixed-size registers.
453 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
454 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
455 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
456 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
457 (aarch64_print_operand, aarch64_print_address_internal)
458 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
459 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
460 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
461 Handle polynomial GET_MODE_SIZE.
462 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
463 wider than SImode without modification.
464 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
465 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
466 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
467 passing and returning SVE modes.
468 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
470 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
471 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
472 (aarch64_allocate_and_probe_stack_space): Likewise.
473 (aarch64_layout_frame): Cope with polynomial offsets.
474 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
475 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
477 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
478 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
479 poly_int64 rather than a HOST_WIDE_INT.
480 (aarch64_get_separate_components, aarch64_process_components)
481 (aarch64_expand_prologue, aarch64_expand_epilogue)
482 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
483 (aarch64_anchor_offset): New function, split out from...
484 (aarch64_legitimize_address): ...here.
485 (aarch64_builtin_vectorization_cost): Handle polynomial
486 TYPE_VECTOR_SUBPARTS.
487 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
489 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
490 number of elements from the PARALLEL rather than the mode.
491 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
492 rather than GET_MODE_BITSIZE.
493 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
494 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
495 (aarch64_expand_vec_perm_const_1): Handle polynomial
496 d->perm.length () and d->perm elements.
497 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
498 Apply to_constant to d->perm elements.
499 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
500 polynomial CONST_VECTOR_NUNITS.
501 (aarch64_move_pointer): Take amount as a poly_int64 rather
503 (aarch64_progress_pointer): Avoid temporary variable.
504 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
505 the mode attribute instead of GET_MODE.
507 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
508 Alan Hayward <alan.hayward@arm.com>
509 David Sherwood <david.sherwood@arm.com>
511 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
512 x exists before using it.
513 (aarch64_add_constant_internal): Rename to...
514 (aarch64_add_offset_1): ...this. Replace regnum with separate
515 src and dest rtxes. Handle the case in which they're different,
516 including when the offset is zero. Replace scratchreg with an rtx.
517 Use 2 additions if there is no spare register into which we can
518 move a 16-bit constant.
519 (aarch64_add_constant): Delete.
520 (aarch64_add_offset): Replace reg with separate src and dest
521 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
522 Use aarch64_add_offset_1.
523 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
524 an rtx rather than an int. Take the delta as a poly_int64
525 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
526 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
527 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
528 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
529 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
531 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
532 aarch64_add_constant.
534 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
536 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
537 Use scalar_float_mode.
539 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
541 * config/aarch64/aarch64-simd.md
542 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
543 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
544 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
545 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
546 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
547 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
548 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
549 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
550 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
551 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
553 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
556 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
557 targ_options->x_arm_arch_string is non NULL.
559 2018-01-11 Tamar Christina <tamar.christina@arm.com>
561 * config/aarch64/aarch64.h
562 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
564 2018-01-11 Sudakshina Das <sudi.das@arm.com>
567 * expmed.c (emit_store_flag_force): Swap if const op0
568 and change VOIDmode to mode of op0.
570 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
572 PR rtl-optimization/83761
573 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
574 than bytes to mode_for_size.
576 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
579 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
580 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
583 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
586 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
588 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
589 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
592 2018-01-10 Michael Collison <michael.collison@arm.com>
594 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
595 * config/aarch64/aarch64-option-extension.def: Add
596 AARCH64_OPT_EXTENSION of 'fp16fml'.
597 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
598 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
599 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
600 * config/aarch64/constraints.md (Ui7): New constraint.
601 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
602 (VFMLA_SEL_W): Ditto.
605 (VFMLA16_LOW): New int iterator.
606 (VFMLA16_HIGH): Ditto.
607 (UNSPEC_FMLAL): New unspec.
608 (UNSPEC_FMLSL): Ditto.
609 (UNSPEC_FMLAL2): Ditto.
610 (UNSPEC_FMLSL2): Ditto.
611 (f16mac): New code attribute.
612 * config/aarch64/aarch64-simd-builtins.def
613 (aarch64_fmlal_lowv2sf): Ditto.
614 (aarch64_fmlsl_lowv2sf): Ditto.
615 (aarch64_fmlalq_lowv4sf): Ditto.
616 (aarch64_fmlslq_lowv4sf): Ditto.
617 (aarch64_fmlal_highv2sf): Ditto.
618 (aarch64_fmlsl_highv2sf): Ditto.
619 (aarch64_fmlalq_highv4sf): Ditto.
620 (aarch64_fmlslq_highv4sf): Ditto.
621 (aarch64_fmlal_lane_lowv2sf): Ditto.
622 (aarch64_fmlsl_lane_lowv2sf): Ditto.
623 (aarch64_fmlal_laneq_lowv2sf): Ditto.
624 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
625 (aarch64_fmlalq_lane_lowv4sf): Ditto.
626 (aarch64_fmlsl_lane_lowv4sf): Ditto.
627 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
628 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
629 (aarch64_fmlal_lane_highv2sf): Ditto.
630 (aarch64_fmlsl_lane_highv2sf): Ditto.
631 (aarch64_fmlal_laneq_highv2sf): Ditto.
632 (aarch64_fmlsl_laneq_highv2sf): Ditto.
633 (aarch64_fmlalq_lane_highv4sf): Ditto.
634 (aarch64_fmlsl_lane_highv4sf): Ditto.
635 (aarch64_fmlalq_laneq_highv4sf): Ditto.
636 (aarch64_fmlsl_laneq_highv4sf): Ditto.
637 * config/aarch64/aarch64-simd.md:
638 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
639 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
640 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
641 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
642 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
643 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
644 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
645 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
646 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
647 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
648 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
649 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
650 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
651 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
652 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
653 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
654 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
655 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
656 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
657 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
658 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
659 (vfmlsl_low_u32): Ditto.
660 (vfmlalq_low_u32): Ditto.
661 (vfmlslq_low_u32): Ditto.
662 (vfmlal_high_u32): Ditto.
663 (vfmlsl_high_u32): Ditto.
664 (vfmlalq_high_u32): Ditto.
665 (vfmlslq_high_u32): Ditto.
666 (vfmlal_lane_low_u32): Ditto.
667 (vfmlsl_lane_low_u32): Ditto.
668 (vfmlal_laneq_low_u32): Ditto.
669 (vfmlsl_laneq_low_u32): Ditto.
670 (vfmlalq_lane_low_u32): Ditto.
671 (vfmlslq_lane_low_u32): Ditto.
672 (vfmlalq_laneq_low_u32): Ditto.
673 (vfmlslq_laneq_low_u32): Ditto.
674 (vfmlal_lane_high_u32): Ditto.
675 (vfmlsl_lane_high_u32): Ditto.
676 (vfmlal_laneq_high_u32): Ditto.
677 (vfmlsl_laneq_high_u32): Ditto.
678 (vfmlalq_lane_high_u32): Ditto.
679 (vfmlslq_lane_high_u32): Ditto.
680 (vfmlalq_laneq_high_u32): Ditto.
681 (vfmlslq_laneq_high_u32): Ditto.
682 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
683 (AARCH64_FL_FOR_ARCH8_4): New.
684 (AARCH64_ISA_F16FML): New ISA flag.
685 (TARGET_F16FML): New feature flag for fp16fml.
686 (doc/invoke.texi): Document new fp16fml option.
688 2018-01-10 Michael Collison <michael.collison@arm.com>
690 * config/aarch64/aarch64-builtins.c:
691 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
692 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
693 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
694 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
695 (AARCH64_ISA_SHA3): New ISA flag.
696 (TARGET_SHA3): New feature flag for sha3.
697 * config/aarch64/iterators.md (sha512_op): New int attribute.
698 (CRYPTO_SHA512): New int iterator.
699 (UNSPEC_SHA512H): New unspec.
700 (UNSPEC_SHA512H2): Ditto.
701 (UNSPEC_SHA512SU0): Ditto.
702 (UNSPEC_SHA512SU1): Ditto.
703 * config/aarch64/aarch64-simd-builtins.def
704 (aarch64_crypto_sha512hqv2di): New builtin.
705 (aarch64_crypto_sha512h2qv2di): Ditto.
706 (aarch64_crypto_sha512su0qv2di): Ditto.
707 (aarch64_crypto_sha512su1qv2di): Ditto.
708 (aarch64_eor3qv8hi): Ditto.
709 (aarch64_rax1qv2di): Ditto.
710 (aarch64_xarqv2di): Ditto.
711 (aarch64_bcaxqv8hi): Ditto.
712 * config/aarch64/aarch64-simd.md:
713 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
714 (aarch64_crypto_sha512su0qv2di): Ditto.
715 (aarch64_crypto_sha512su1qv2di): Ditto.
716 (aarch64_eor3qv8hi): Ditto.
717 (aarch64_rax1qv2di): Ditto.
718 (aarch64_xarqv2di): Ditto.
719 (aarch64_bcaxqv8hi): Ditto.
720 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
721 (vsha512h2q_u64): Ditto.
722 (vsha512su0q_u64): Ditto.
723 (vsha512su1q_u64): Ditto.
728 * config/arm/types.md (crypto_sha512): New type attribute.
729 (crypto_sha3): Ditto.
730 (doc/invoke.texi): Document new sha3 option.
732 2018-01-10 Michael Collison <michael.collison@arm.com>
734 * config/aarch64/aarch64-builtins.c:
735 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
736 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
737 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
738 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
739 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
740 (AARCH64_ISA_SM4): New ISA flag.
741 (TARGET_SM4): New feature flag for sm4.
742 * config/aarch64/aarch64-simd-builtins.def
743 (aarch64_sm3ss1qv4si): Ditto.
744 (aarch64_sm3tt1aq4si): Ditto.
745 (aarch64_sm3tt1bq4si): Ditto.
746 (aarch64_sm3tt2aq4si): Ditto.
747 (aarch64_sm3tt2bq4si): Ditto.
748 (aarch64_sm3partw1qv4si): Ditto.
749 (aarch64_sm3partw2qv4si): Ditto.
750 (aarch64_sm4eqv4si): Ditto.
751 (aarch64_sm4ekeyqv4si): Ditto.
752 * config/aarch64/aarch64-simd.md:
753 (aarch64_sm3ss1qv4si): Ditto.
754 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
755 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
756 (aarch64_sm4eqv4si): Ditto.
757 (aarch64_sm4ekeyqv4si): Ditto.
758 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
760 (CRYPTO_SM3TT): Ditto.
761 (CRYPTO_SM3PART): Ditto.
762 (UNSPEC_SM3SS1): New unspec.
763 (UNSPEC_SM3TT1A): Ditto.
764 (UNSPEC_SM3TT1B): Ditto.
765 (UNSPEC_SM3TT2A): Ditto.
766 (UNSPEC_SM3TT2B): Ditto.
767 (UNSPEC_SM3PARTW1): Ditto.
768 (UNSPEC_SM3PARTW2): Ditto.
769 (UNSPEC_SM4E): Ditto.
770 (UNSPEC_SM4EKEY): Ditto.
771 * config/aarch64/constraints.md (Ui2): New constraint.
772 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
773 * config/arm/types.md (crypto_sm3): New type attribute.
775 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
776 (vsm3tt1aq_u32): Ditto.
777 (vsm3tt1bq_u32): Ditto.
778 (vsm3tt2aq_u32): Ditto.
779 (vsm3tt2bq_u32): Ditto.
780 (vsm3partw1q_u32): Ditto.
781 (vsm3partw2q_u32): Ditto.
783 (vsm4ekeyq_u32): Ditto.
784 (doc/invoke.texi): Document new sm4 option.
786 2018-01-10 Michael Collison <michael.collison@arm.com>
788 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
789 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
790 (AARCH64_FL_FOR_ARCH8_4): New.
791 (AARCH64_FL_V8_4): New flag.
792 (doc/invoke.texi): Document new armv8.4-a option.
794 2018-01-10 Michael Collison <michael.collison@arm.com>
796 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
797 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
798 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
799 * config/aarch64/aarch64-option-extension.def: Add
800 AARCH64_OPT_EXTENSION of 'sha2'.
801 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
802 (crypto): Disable sha2 and aes if crypto disabled.
803 (crypto): Enable aes and sha2 if enabled.
804 (simd): Disable sha2 and aes if simd disabled.
805 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
807 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
808 (TARGET_SHA2): New feature flag for sha2.
809 (TARGET_AES): New feature flag for aes.
810 * config/aarch64/aarch64-simd.md:
811 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
812 conditional on TARGET_AES.
813 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
814 (aarch64_crypto_sha1hsi): Make pattern conditional
816 (aarch64_crypto_sha1hv4si): Ditto.
817 (aarch64_be_crypto_sha1hv4si): Ditto.
818 (aarch64_crypto_sha1su1v4si): Ditto.
819 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
820 (aarch64_crypto_sha1su0v4si): Ditto.
821 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
822 (aarch64_crypto_sha256su0v4si): Ditto.
823 (aarch64_crypto_sha256su1v4si): Ditto.
824 (doc/invoke.texi): Document new aes and sha2 options.
826 2018-01-10 Martin Sebor <msebor@redhat.com>
828 PR tree-optimization/83781
829 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
832 2018-01-11 Martin Sebor <msebor@gmail.com>
833 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
835 PR tree-optimization/83501
836 PR tree-optimization/81703
838 * tree-ssa-strlen.c (get_string_cst): Rename...
839 (get_string_len): ...to this. Handle global constants.
840 (handle_char_store): Adjust.
842 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
843 Jim Wilson <jimw@sifive.com>
845 * config/riscv/riscv-protos.h (riscv_output_return): New.
846 * config/riscv/riscv.c (struct machine_function): New naked_p field.
847 (riscv_attribute_table, riscv_output_return),
848 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
849 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
850 (riscv_compute_frame_info): Only compute frame->mask if not a naked
852 (riscv_expand_prologue): Add early return for naked function.
853 (riscv_expand_epilogue): Likewise.
854 (riscv_function_ok_for_sibcall): Return false for naked function.
855 (riscv_set_current_function): New.
856 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
857 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
858 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
859 * doc/extend.texi (RISC-V Function Attributes): New.
861 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
863 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
864 check for 128-bit long double before checking TCmode.
865 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
866 128-bit long doubles before checking TFmode or TCmode.
867 (FLOAT128_IBM_P): Likewise.
869 2018-01-10 Martin Sebor <msebor@redhat.com>
871 PR tree-optimization/83671
872 * builtins.c (c_strlen): Unconditionally return zero for the empty
874 Use -Warray-bounds for warnings.
875 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
876 for non-constant array indices with COMPONENT_REF, arrays of
877 arrays, and pointers to arrays.
878 (gimple_fold_builtin_strlen): Determine and set length range for
879 non-constant character arrays.
881 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
884 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
887 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
889 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
891 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
894 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
895 VECTOR_MEM_ALTIVEC_OR_VSX_P.
896 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
897 indexed_or_indirect_operand predicate.
898 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
899 (*vsx_le_perm_load_v8hi): Likewise.
900 (*vsx_le_perm_load_v16qi): Likewise.
901 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
902 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
903 (*vsx_le_perm_store_v8hi): Likewise.
904 (*vsx_le_perm_store_v16qi): Likewise.
905 (eight unnamed splitters): Likewise.
907 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
909 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
910 * config/rs6000/emmintrin.h: Likewise.
911 * config/rs6000/mmintrin.h: Likewise.
912 * config/rs6000/xmmintrin.h: Likewise.
914 2018-01-10 David Malcolm <dmalcolm@redhat.com>
917 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
919 * tree.c (tree_nop_conversion): Return true for location wrapper
921 (maybe_wrap_with_location): New function.
922 (selftest::check_strip_nops): New function.
923 (selftest::test_location_wrappers): New function.
924 (selftest::tree_c_tests): Call it.
925 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
926 (maybe_wrap_with_location): New decl.
927 (EXPR_LOCATION_WRAPPER_P): New macro.
928 (location_wrapper_p): New inline function.
929 (tree_strip_any_location_wrapper): New inline function.
931 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
934 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
935 stack_realign_offset for the largest alignment of stack slot
937 (ix86_find_max_used_stack_alignment): New function.
938 (ix86_finalize_stack_frame_flags): Use it. Set
939 max_used_stack_alignment if we don't realign stack.
940 * config/i386/i386.h (machine_function): Add
941 max_used_stack_alignment.
943 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
945 * config/arm/arm.opt (-mbranch-cost): New option.
946 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
949 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
952 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
953 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
955 2018-01-10 Richard Biener <rguenther@suse.de>
958 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
959 early out so it also covers the case where we have a non-NULL
962 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
964 PR tree-optimization/83753
965 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
966 for non-strided grouped accesses if the number of elements is 1.
968 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
971 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
972 * i386.h (TARGET_USE_GATHER): Define.
973 * x86-tune.def (X86_TUNE_USE_GATHER): New.
975 2018-01-10 Martin Liska <mliska@suse.cz>
978 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
979 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
981 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
982 CLEANUP_NO_PARTITIONING is not set.
984 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
986 * doc/rtl.texi: Remove documentation of (const ...) wrappers
987 for vectors, as a partial revert of r254296.
988 * rtl.h (const_vec_p): Delete.
989 (const_vec_duplicate_p): Don't test for vector CONSTs.
990 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
991 * expmed.c (make_tree): Likewise.
994 * common.md (E, F): Use CONSTANT_P instead of checking for
996 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
997 checking for CONST_VECTOR.
999 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
1002 * predict.c (force_edge_cold): Handle in more sane way edges
1005 2018-01-09 Carl Love <cel@us.ibm.com>
1007 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
1009 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
1010 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
1011 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
1012 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
1013 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
1014 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
1015 * config/rs6000/rs6000-protos.h: Add extern defition for
1016 rs6000_generate_float2_double_code.
1017 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
1019 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
1020 (float2_v2df): Add define_expand.
1022 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
1025 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
1026 op_mode in the force_to_mode call.
1028 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
1030 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
1031 instead of checking each element individually.
1032 (aarch64_evpc_uzp): Likewise.
1033 (aarch64_evpc_zip): Likewise.
1034 (aarch64_evpc_ext): Likewise.
1035 (aarch64_evpc_rev): Likewise.
1036 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
1037 instead of checking each element individually. Return true without
1039 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
1040 whether all selected elements come from the same input, instead of
1041 checking each element individually. Remove calls to gen_rtx_REG,
1042 start_sequence and end_sequence and instead assert that no rtl is
1045 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
1047 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
1048 order of HIGH and CONST checks.
1050 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
1052 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
1053 if the destination isn't an SSA_NAME.
1055 2018-01-09 Richard Biener <rguenther@suse.de>
1057 PR tree-optimization/83668
1058 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
1060 (canonicalize_loop_form): ... here, renamed from ...
1061 (canonicalize_loop_closed_ssa_form): ... this and amended to
1062 swap successor edges for loop exit blocks to make us use
1063 the RPO order we need for initial schedule generation.
1065 2018-01-09 Joseph Myers <joseph@codesourcery.com>
1067 PR tree-optimization/64811
1068 * match.pd: When optimizing comparisons with Inf, avoid
1069 introducing or losing exceptions from comparisons with NaN.
1071 2018-01-09 Martin Liska <mliska@suse.cz>
1074 * asan.c (shadow_mem_size): Add gcc_assert.
1076 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
1078 Don't save registers in main().
1081 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
1082 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
1083 * config/avr/avr.c (avr_set_current_function): Don't error if
1084 naked, OS_task or OS_main are specified at the same time.
1085 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
1087 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
1089 * common/config/avr/avr-common.c (avr_option_optimization_table):
1090 Switch on -mmain-is-OS_task for optimizing compilations.
1092 2018-01-09 Richard Biener <rguenther@suse.de>
1094 PR tree-optimization/83572
1095 * graphite.c: Include cfganal.h.
1096 (graphite_transform_loops): Connect infinite loops to exit
1097 and remove fake edges at the end.
1099 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
1101 * ipa-inline.c (edge_badness): Revert accidental checkin.
1103 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
1106 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
1107 symbols; not inline clones.
1109 2018-01-09 Jakub Jelinek <jakub@redhat.com>
1112 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
1113 hard registers. Formatting fixes.
1115 PR preprocessor/83722
1116 * gcc.c (try_generate_repro): Pass
1117 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
1118 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
1121 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
1122 Kito Cheng <kito.cheng@gmail.com>
1124 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
1125 (riscv_leaf_function_p): Delete.
1126 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
1128 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1130 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
1132 (do_ifelse): New function.
1133 (do_isel): New function.
1134 (do_sub3): New function.
1135 (do_add3): New function.
1136 (do_load_mask_compare): New function.
1137 (do_overlap_load_compare): New function.
1138 (expand_compare_loop): New function.
1139 (expand_block_compare): Call expand_compare_loop() when appropriate.
1140 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
1142 (-mblock-compare-inline-loop-limit): New option.
1144 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1147 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
1148 Reverse order of second and third operands in first alternative.
1149 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
1150 of first and second elements in UNSPEC_VPERMR vector.
1151 (altivec_expand_vec_perm_le): Likewise.
1153 2017-01-08 Jeff Law <law@redhat.com>
1155 PR rtl-optimizatin/81308
1156 * tree-switch-conversion.c (cfg_altered): New file scoped static.
1157 (process_switch): If group_case_labels makes a change, then set
1159 (pass_convert_switch::execute): If a switch is converted, then
1160 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
1162 PR rtl-optimization/81308
1163 * recog.c (split_all_insns): Conditionally cleanup the CFG after
1166 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
1168 PR target/83663 - Revert r255946
1169 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
1170 generation for cases where splatting a value is not useful.
1171 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
1172 across a vec_duplicate and a paradoxical subreg forming a vector
1173 mode to a vec_concat.
1175 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1177 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
1178 -march=armv8.3-a variants.
1179 * config/arm/t-multilib: Likewise.
1180 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
1182 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1184 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
1186 (cceq_ior_compare_complement): Give it a name so I can use it, and
1187 change boolean_or_operator predicate to boolean_operator so it can
1188 be used to generate a crand.
1189 (eqne): New code iterator.
1190 (bd/bd_neg): New code_attrs.
1191 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
1192 a single define_insn.
1193 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
1194 decrement (bdnzt/bdnzf/bdzt/bdzf).
1195 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
1196 with the new names of the branch decrement patterns, and added the
1197 names of the branch decrement conditional patterns.
1199 2018-01-08 Richard Biener <rguenther@suse.de>
1201 PR tree-optimization/83563
1202 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
1205 2018-01-08 Richard Biener <rguenther@suse.de>
1208 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
1210 2018-01-08 Richard Biener <rguenther@suse.de>
1212 PR tree-optimization/83685
1213 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
1214 references to abnormals.
1216 2018-01-08 Richard Biener <rguenther@suse.de>
1219 * dwarf2out.c (output_indirect_strings): Handle empty
1220 skeleton_debug_str_hash.
1221 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
1223 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
1225 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
1226 (emit_store_direct): Likewise.
1227 (arc_trampoline_adjust_address): Likewise.
1228 (arc_asm_trampoline_template): New function.
1229 (arc_initialize_trampoline): Use asm_trampoline_template.
1230 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
1231 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
1232 * config/arc/arc.md (flush_icache): Delete pattern.
1234 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
1236 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
1237 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
1240 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
1243 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
1244 by not USED_FOR_TARGET.
1245 (make_pass_resolve_sw_modes): Likewise.
1247 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
1249 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
1252 2018-01-08 Richard Biener <rguenther@suse.de>
1255 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
1257 2018-01-08 Richard Biener <rguenther@suse.de>
1260 * match.pd ((t * 2) / 2) -> t): Add missing :c.
1262 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
1265 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
1266 basic blocks with a small number of successors.
1267 (convert_control_dep_chain_into_preds): Improve handling of
1269 (dump_predicates): Split apart into...
1270 (dump_pred_chain): ...here...
1271 (dump_pred_info): ...and here.
1272 (can_one_predicate_be_invalidated_p): Add debugging printfs.
1273 (can_chain_union_be_invalidated_p): Improve check for invalidation
1275 (uninit_uses_cannot_happen): Avoid unnecessary if
1276 convert_control_dep_chain_into_preds yielded nothing.
1278 2018-01-06 Martin Sebor <msebor@redhat.com>
1280 PR tree-optimization/83640
1281 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
1282 subtracting negative offset from size.
1283 (builtin_access::overlap): Adjust offset bounds of the access to fall
1284 within the size of the object if possible.
1286 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
1288 PR rtl-optimization/83699
1289 * expmed.c (extract_bit_field_1): Restrict the vector usage of
1290 extract_bit_field_as_subreg to cases in which the extracted
1291 value is also a vector.
1293 * lra-constraints.c (process_alt_operands): Test for the equivalence
1294 substitutions when detecting a possible reload cycle.
1296 2018-01-06 Jakub Jelinek <jakub@redhat.com>
1299 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
1300 by default if flag_selective_schedling{,2}. Formatting fixes.
1302 PR rtl-optimization/83682
1303 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
1304 if it has non-VECTOR_MODE element mode.
1305 (vec_duplicate_p): Likewise.
1308 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
1309 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
1311 2018-01-05 Jakub Jelinek <jakub@redhat.com>
1314 * config/i386/i386-builtin.def
1315 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
1316 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
1317 Require also OPTION_MASK_ISA_AVX512F in addition to
1318 OPTION_MASK_ISA_GFNI.
1319 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
1320 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
1321 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
1322 to OPTION_MASK_ISA_GFNI.
1323 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
1324 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
1325 OPTION_MASK_ISA_AVX512BW.
1326 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
1327 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
1328 addition to OPTION_MASK_ISA_GFNI.
1329 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
1330 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
1331 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
1332 to OPTION_MASK_ISA_GFNI.
1333 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
1334 a requirement for all ISAs rather than any of them with a few
1336 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
1338 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
1339 bitmasks to be enabled with 3 exceptions, instead of requiring any
1340 enabled ISA with lots of exceptions.
1341 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
1342 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
1343 Change avx512bw in isa attribute to avx512f.
1344 * config/i386/sgxintrin.h: Add license boilerplate.
1345 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
1346 to __AVX512F__ and __AVX512VL to __AVX512VL__.
1347 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
1348 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
1350 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
1351 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
1352 temporarily sse2 rather than sse if not enabled already.
1355 * config/i386/sse.md (VI248_VLBW): Rename to ...
1356 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
1357 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
1358 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
1359 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
1360 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
1361 mode iterator instead of VI248_VLBW.
1363 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
1365 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
1366 (record_modified): Skip clobbers; add debug output.
1367 (param_change_prob): Use sreal frequencies.
1369 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
1371 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
1372 punt for user-aligned variables.
1374 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
1376 * tree-chrec.c (chrec_contains_symbols): Return true for
1379 2018-01-05 Sudakshina Das <sudi.das@arm.com>
1382 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
1383 of (x|y) == x for BICS pattern.
1385 2018-01-05 Jakub Jelinek <jakub@redhat.com>
1387 PR tree-optimization/83605
1388 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
1389 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
1392 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
1394 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
1395 * config/epiphany/rtems.h: New file.
1397 2018-01-04 Jakub Jelinek <jakub@redhat.com>
1398 Uros Bizjak <ubizjak@gmail.com>
1401 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
1402 QIreg_operand instead of register_operand predicate.
1403 * config/i386/i386.c (ix86_rop_should_change_byte_p,
1404 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
1405 comments instead of -fmitigate[-_]rop.
1407 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1410 * cgraphunit.c (symbol_table::compile): Switch to text_section
1411 before calling assembly_start debug hook.
1412 * run-rtl-passes.c (run_rtl_passes): Likewise.
1415 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
1417 * tree-vrp.c (extract_range_from_binary_expr_1): Check
1418 range_int_cst_p rather than !symbolic_range_p before calling
1419 extract_range_from_multiplicative_op_1.
1421 2017-01-04 Jeff Law <law@redhat.com>
1423 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
1424 redundant test in assertion.
1426 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
1428 * doc/rtl.texi: Document machine_mode wrapper classes.
1430 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
1432 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
1435 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
1437 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
1438 the VEC_PERM_EXPR fold to fail.
1440 2018-01-04 Jakub Jelinek <jakub@redhat.com>
1443 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
1444 to switched_sections.
1446 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
1449 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
1452 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
1455 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
1456 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
1458 2018-01-04 Jakub Jelinek <jakub@redhat.com>
1461 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
1462 is BLKmode and bitpos not zero or mode change is needed.
1464 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
1467 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
1470 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
1473 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
1474 instead of MULT rtx. Update all corresponding splitters.
1476 (*ssub<modesuffix>): Ditto.
1478 (*cmp_sadd_di): Update split patterns.
1479 (*cmp_sadd_si): Ditto.
1480 (*cmp_sadd_sidi): Ditto.
1481 (*cmp_ssub_di): Ditto.
1482 (*cmp_ssub_si): Ditto.
1483 (*cmp_ssub_sidi): Ditto.
1484 * config/alpha/predicates.md (const23_operand): New predicate.
1485 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
1486 Look for ASHIFT, not MULT inner operand.
1487 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
1489 2018-01-04 Martin Liska <mliska@suse.cz>
1491 PR gcov-profile/83669
1492 * gcov.c (output_intermediate_file): Add version to intermediate
1494 * doc/gcov.texi: Document new field 'version' in intermediate
1495 file format. Fix location of '-k' option of gcov command.
1497 2018-01-04 Martin Liska <mliska@suse.cz>
1500 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
1502 2018-01-04 Jakub Jelinek <jakub@redhat.com>
1504 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
1506 2018-01-03 Martin Sebor <msebor@redhat.com>
1508 PR tree-optimization/83655
1509 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
1510 checking calls with invalid arguments.
1512 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1514 * tree-vect-stmts.c (vect_get_store_rhs): New function.
1515 (vectorizable_mask_load_store): Delete.
1516 (vectorizable_call): Return false for masked loads and stores.
1517 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
1518 instead of gimple_assign_rhs1.
1519 (vectorizable_load): Handle IFN_MASK_LOAD.
1520 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
1522 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1524 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
1526 (vectorizable_mask_load_store): ...here.
1527 (vectorizable_load): ...and here.
1529 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1531 * tree-vect-stmts.c (vect_build_all_ones_mask)
1532 (vect_build_zero_merge_argument): New functions, split out from...
1533 (vectorizable_load): ...here.
1535 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1537 * tree-vect-stmts.c (vect_check_store_rhs): New function,
1539 (vectorizable_mask_load_store): ...here.
1540 (vectorizable_store): ...and here.
1542 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1544 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
1546 (vectorizable_mask_load_store): ...here.
1548 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1550 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
1551 (vect_model_store_cost): Take a vec_load_store_type instead of a
1553 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
1554 (vect_model_store_cost): Take a vec_load_store_type instead of a
1556 (vectorizable_mask_load_store): Update accordingly.
1557 (vectorizable_store): Likewise.
1558 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
1560 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1562 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
1563 IFN_MASK_LOAD calls here rather than...
1564 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
1566 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1567 Alan Hayward <alan.hayward@arm.com>
1568 David Sherwood <david.sherwood@arm.com>
1570 * expmed.c (extract_bit_field_1): For vector extracts,
1571 fall back to extract_bit_field_as_subreg if vec_extract
1574 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1575 Alan Hayward <alan.hayward@arm.com>
1576 David Sherwood <david.sherwood@arm.com>
1578 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
1579 they are variable or constant sized.
1580 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
1581 slots for constant-sized data.
1583 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1584 Alan Hayward <alan.hayward@arm.com>
1585 David Sherwood <david.sherwood@arm.com>
1587 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
1588 handling COND_EXPRs with boolean comparisons, try to find a better
1589 basis for the mask type than the boolean itself.
1591 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1593 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
1594 is calculated and how it can be overridden.
1595 * genmodes.c (max_bitsize_mode_any_mode): New variable.
1596 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
1598 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
1601 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1602 Alan Hayward <alan.hayward@arm.com>
1603 David Sherwood <david.sherwood@arm.com>
1605 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
1606 Remove the mode argument.
1607 (aarch64_simd_valid_immediate): Remove the mode and inverse
1609 * config/aarch64/iterators.md (bitsize): New iterator.
1610 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
1611 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
1612 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
1613 aarch64_simd_valid_immediate.
1614 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
1615 (aarch64_reg_or_bic_imm): Likewise.
1616 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
1617 with an insn_type enum and msl with a modifier_type enum.
1618 Replace element_width with a scalar_mode. Change the shift
1619 to unsigned int. Add constructors for scalar_float_mode and
1620 scalar_int_mode elements.
1621 (aarch64_vect_float_const_representable_p): Delete.
1622 (aarch64_can_const_movi_rtx_p)
1623 (aarch64_simd_scalar_immediate_valid_for_move)
1624 (aarch64_simd_make_constant): Update call to
1625 aarch64_simd_valid_immediate.
1626 (aarch64_advsimd_valid_immediate_hs): New function.
1627 (aarch64_advsimd_valid_immediate): Likewise.
1628 (aarch64_simd_valid_immediate): Remove mode and inverse
1629 arguments. Rewrite to use the above. Use const_vec_duplicate_p
1630 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
1631 and aarch64_float_const_representable_p on the result.
1632 (aarch64_output_simd_mov_immediate): Remove mode argument.
1633 Update call to aarch64_simd_valid_immediate and use of
1634 simd_immediate_info.
1635 (aarch64_output_scalar_simd_mov_immediate): Update call
1638 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1639 Alan Hayward <alan.hayward@arm.com>
1640 David Sherwood <david.sherwood@arm.com>
1642 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
1643 (mode_nunits): Likewise CONST_MODE_NUNITS.
1644 * machmode.def (ADJUST_NUNITS): Document.
1645 * genmodes.c (mode_data::need_nunits_adj): New field.
1646 (blank_mode): Update accordingly.
1647 (adj_nunits): New variable.
1648 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
1650 (emit_mode_size_inline): Set need_bytesize_adj for all modes
1651 listed in adj_nunits.
1652 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
1653 listed in adj_nunits. Don't emit case statements for such modes.
1654 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
1655 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
1656 nothing if adj_nunits is nonnull.
1657 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
1658 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
1659 (emit_mode_fbit): Update use of print_maybe_const_decl.
1660 (emit_move_size): Likewise. Treat the array as non-const
1662 (emit_mode_adjustments): Handle adj_nunits.
1664 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1666 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
1667 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
1668 (VECTOR_MODES): Use it.
1669 (make_vector_modes): Take the prefix as an argument.
1671 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1672 Alan Hayward <alan.hayward@arm.com>
1673 David Sherwood <david.sherwood@arm.com>
1675 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
1676 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
1677 for MODE_VECTOR_BOOL.
1678 * machmode.def (VECTOR_BOOL_MODE): Document.
1679 * genmodes.c (VECTOR_BOOL_MODE): New macro.
1680 (make_vector_bool_mode): New function.
1681 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
1683 * lto-streamer-in.c (lto_input_mode_table): Likewise.
1684 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
1686 * stor-layout.c (int_mode_for_mode): Likewise.
1687 * tree.c (build_vector_type_for_mode): Likewise.
1688 * varasm.c (output_constant_pool_2): Likewise.
1689 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
1690 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
1691 for MODE_VECTOR_BOOL.
1692 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
1693 of mode class checks.
1694 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
1695 instead of a list of mode class checks.
1696 (expand_vector_scalar_condition): Likewise.
1697 (type_for_widest_vector_mode): Handle BImode as an inner mode.
1699 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1700 Alan Hayward <alan.hayward@arm.com>
1701 David Sherwood <david.sherwood@arm.com>
1703 * machmode.h (mode_size): Change from unsigned short to
1705 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
1706 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
1707 or if measurement_type is not polynomial.
1708 (fixed_size_mode::includes_p): Check for constant-sized modes.
1709 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
1710 return a poly_uint16 rather than an unsigned short.
1711 (emit_mode_size): Change the type of mode_size from unsigned short
1712 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
1713 (emit_mode_adjustments): Cope with polynomial vector sizes.
1714 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
1716 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
1718 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
1719 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
1720 * caller-save.c (setup_save_areas): Likewise.
1721 (replace_reg_with_saved_mem): Likewise.
1722 * calls.c (emit_library_call_value_1): Likewise.
1723 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
1724 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
1725 (gen_lowpart_for_combine): Likewise.
1726 * convert.c (convert_to_integer_1): Likewise.
1727 * cse.c (equiv_constant, cse_insn): Likewise.
1728 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
1729 (cselib_subst_to_values): Likewise.
1730 * dce.c (word_dce_process_block): Likewise.
1731 * df-problems.c (df_word_lr_mark_ref): Likewise.
1732 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
1733 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
1734 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
1735 (rtl_for_decl_location): Likewise.
1736 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
1737 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
1738 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
1739 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
1740 (expand_expr_real_1): Likewise.
1741 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
1742 (pad_below): Likewise.
1743 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
1744 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
1745 * ira.c (get_subreg_tracking_sizes): Likewise.
1746 * ira-build.c (ira_create_allocno_objects): Likewise.
1747 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
1748 (ira_sort_regnos_for_alter_reg): Likewise.
1749 * ira-costs.c (record_operand_costs): Likewise.
1750 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
1751 (resolve_simple_move): Likewise.
1752 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
1753 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
1754 (lra_constraints): Likewise.
1755 (CONST_POOL_OK_P): Reject variable-sized modes.
1756 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
1757 (add_pseudo_to_slot, lra_spill): Likewise.
1758 * omp-low.c (omp_clause_aligned_alignment): Likewise.
1759 * optabs-query.c (get_best_extraction_insn): Likewise.
1760 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
1761 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
1762 (expand_mult_highpart, valid_multiword_target_p): Likewise.
1763 * recog.c (offsettable_address_addr_space_p): Likewise.
1764 * regcprop.c (maybe_mode_change): Likewise.
1765 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
1766 * regrename.c (build_def_use): Likewise.
1767 * regstat.c (dump_reg_info): Likewise.
1768 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
1769 (find_reloads, find_reloads_subreg_address): Likewise.
1770 * reload1.c (eliminate_regs_1): Likewise.
1771 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
1772 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
1773 (simplify_binary_operation_1, simplify_subreg): Likewise.
1774 * targhooks.c (default_function_arg_padding): Likewise.
1775 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
1776 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
1777 (verify_gimple_assign_ternary): Likewise.
1778 * tree-inline.c (estimate_move_cost): Likewise.
1779 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1780 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
1781 (get_address_cost_ainc): Likewise.
1782 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
1783 (vect_supportable_dr_alignment): Likewise.
1784 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
1785 (vectorizable_reduction): Likewise.
1786 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
1787 (vectorizable_operation, vectorizable_load): Likewise.
1788 * tree.c (build_same_sized_truth_vector_type): Likewise.
1789 * valtrack.c (cleanup_auto_inc_dec): Likewise.
1790 * var-tracking.c (emit_note_insn_var_location): Likewise.
1791 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
1792 (ADDR_VEC_ALIGN): Likewise.
1794 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1795 Alan Hayward <alan.hayward@arm.com>
1796 David Sherwood <david.sherwood@arm.com>
1798 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
1800 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
1801 or if measurement_type is polynomial.
1802 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
1803 * combine.c (make_extraction): Likewise.
1804 * dse.c (find_shift_sequence): Likewise.
1805 * dwarf2out.c (mem_loc_descriptor): Likewise.
1806 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
1807 (extract_bit_field, extract_low_bits): Likewise.
1808 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
1809 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
1810 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
1811 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
1812 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
1813 * reload.c (find_reloads): Likewise.
1814 * reload1.c (alter_reg): Likewise.
1815 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
1816 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
1817 * tree-if-conv.c (predicate_mem_writes): Likewise.
1818 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
1819 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
1820 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
1821 * valtrack.c (dead_debug_insert_temp): Likewise.
1822 * varasm.c (mergeable_constant_section): Likewise.
1823 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
1825 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1826 Alan Hayward <alan.hayward@arm.com>
1827 David Sherwood <david.sherwood@arm.com>
1829 * expr.c (expand_assignment): Cope with polynomial mode sizes
1830 when assigning to a CONCAT.
1832 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1833 Alan Hayward <alan.hayward@arm.com>
1834 David Sherwood <david.sherwood@arm.com>
1836 * machmode.h (mode_precision): Change from unsigned short to
1838 (mode_to_precision): Return a poly_uint16 rather than an unsigned
1840 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
1841 or if measurement_type is not polynomial.
1842 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
1843 in which the mode is already known to be a scalar_int_mode.
1844 * genmodes.c (emit_mode_precision): Change the type of mode_precision
1845 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
1847 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
1848 for GET_MODE_PRECISION.
1849 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
1850 for GET_MODE_PRECISION.
1851 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
1853 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
1854 (expand_field_assignment, make_extraction): Likewise.
1855 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
1856 (get_last_value): Likewise.
1857 * convert.c (convert_to_integer_1): Likewise.
1858 * cse.c (cse_insn): Likewise.
1859 * expr.c (expand_expr_real_1): Likewise.
1860 * lra-constraints.c (simplify_operand_subreg): Likewise.
1861 * optabs-query.c (can_atomic_load_p): Likewise.
1862 * optabs.c (expand_atomic_load): Likewise.
1863 (expand_atomic_store): Likewise.
1864 * ree.c (combine_reaching_defs): Likewise.
1865 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
1866 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
1867 * tree.h (type_has_mode_precision_p): Likewise.
1868 * ubsan.c (instrument_si_overflow): Likewise.
1870 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1871 Alan Hayward <alan.hayward@arm.com>
1872 David Sherwood <david.sherwood@arm.com>
1874 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
1875 polynomial numbers of units.
1876 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
1877 (valid_vector_subparts_p): New function.
1878 (build_vector_type): Remove temporary shim and take the number
1879 of units as a poly_uint64 rather than an int.
1880 (build_opaque_vector_type): Take the number of units as a
1881 poly_uint64 rather than an int.
1882 * tree.c (build_vector_from_ctor): Handle polynomial
1883 TYPE_VECTOR_SUBPARTS.
1884 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
1885 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
1886 (build_vector_from_val): If the number of units is variable,
1887 use build_vec_duplicate_cst for constant operands and
1888 VEC_DUPLICATE_EXPR otherwise.
1889 (make_vector_type): Remove temporary is_constant ().
1890 (build_vector_type, build_opaque_vector_type): Take the number of
1891 units as a poly_uint64 rather than an int.
1892 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
1894 * cfgexpand.c (expand_debug_expr): Likewise.
1895 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
1896 (store_constructor, expand_expr_real_1): Likewise.
1897 (const_scalar_mask_from_tree): Likewise.
1898 * fold-const-call.c (fold_const_reduction): Likewise.
1899 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
1900 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
1901 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
1902 (fold_relational_const): Likewise.
1903 (native_interpret_vector): Likewise. Change the size from an
1904 int to an unsigned int.
1905 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
1906 TYPE_VECTOR_SUBPARTS.
1907 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
1908 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
1909 duplicating a non-constant operand into a variable-length vector.
1910 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
1911 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
1912 * ipa-icf.c (sem_variable::equals): Likewise.
1913 * match.pd: Likewise.
1914 * omp-simd-clone.c (simd_clone_subparts): Likewise.
1915 * print-tree.c (print_node): Likewise.
1916 * stor-layout.c (layout_type): Likewise.
1917 * targhooks.c (default_builtin_vectorization_cost): Likewise.
1918 * tree-cfg.c (verify_gimple_comparison): Likewise.
1919 (verify_gimple_assign_binary): Likewise.
1920 (verify_gimple_assign_ternary): Likewise.
1921 (verify_gimple_assign_single): Likewise.
1922 * tree-pretty-print.c (dump_generic_node): Likewise.
1923 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1924 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
1925 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
1926 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
1927 (vect_shift_permute_load_chain): Likewise.
1928 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
1929 (expand_vector_condition, optimize_vector_constructor): Likewise.
1930 (lower_vec_perm, get_compute_type): Likewise.
1931 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
1932 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
1933 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
1934 (vect_recog_mask_conversion_pattern): Likewise.
1935 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
1936 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
1937 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1938 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
1939 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
1940 (vectorizable_shift, vectorizable_operation, vectorizable_store)
1941 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
1942 (supportable_widening_operation): Likewise.
1943 (supportable_narrowing_operation): Likewise.
1944 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
1946 * varasm.c (output_constant): Likewise.
1948 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1949 Alan Hayward <alan.hayward@arm.com>
1950 David Sherwood <david.sherwood@arm.com>
1952 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
1953 so that both the length == 3 and length != 3 cases set up their
1954 own permute vectors. Add comments explaining why we know the
1955 number of elements is constant.
1956 (vect_permute_load_chain): Likewise.
1958 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1959 Alan Hayward <alan.hayward@arm.com>
1960 David Sherwood <david.sherwood@arm.com>
1962 * machmode.h (mode_nunits): Change from unsigned char to
1964 (ONLY_FIXED_SIZE_MODES): New macro.
1965 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
1966 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
1967 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
1969 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
1970 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
1971 or if measurement_type is not polynomial.
1972 * genmodes.c (ZERO_COEFFS): New macro.
1973 (emit_mode_nunits_inline): Make mode_nunits_inline return a
1975 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
1976 Use ZERO_COEFFS when emitting initializers.
1977 * data-streamer.h (bp_pack_poly_value): New function.
1978 (bp_unpack_poly_value): Likewise.
1979 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
1980 for GET_MODE_NUNITS.
1981 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
1982 for GET_MODE_NUNITS.
1983 * tree.c (make_vector_type): Remove temporary shim and make
1984 the real function take the number of units as a poly_uint64
1986 (build_vector_type_for_mode): Handle polynomial nunits.
1987 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
1988 * emit-rtl.c (const_vec_series_p_1): Likewise.
1989 (gen_rtx_CONST_VECTOR): Likewise.
1990 * fold-const.c (test_vec_duplicate_folding): Likewise.
1991 * genrecog.c (validate_pattern): Likewise.
1992 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
1993 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
1994 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
1995 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
1996 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
1997 * rtlanal.c (subreg_get_info): Likewise.
1998 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1999 (vect_grouped_load_supported): Likewise.
2000 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
2001 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
2002 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
2003 (simplify_const_unary_operation, simplify_binary_operation_1)
2004 (simplify_const_binary_operation, simplify_ternary_operation)
2005 (test_vector_ops_duplicate, test_vector_ops): Likewise.
2006 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
2007 instead of CONST_VECTOR_NUNITS.
2008 * varasm.c (output_constant_pool_2): Likewise.
2009 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
2010 explicit-encoded elements in the XVEC for variable-length vectors.
2012 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2014 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
2016 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2017 Alan Hayward <alan.hayward@arm.com>
2018 David Sherwood <david.sherwood@arm.com>
2020 * coretypes.h (fixed_size_mode): Declare.
2021 (fixed_size_mode_pod): New typedef.
2022 * builtins.h (target_builtins::x_apply_args_mode)
2023 (target_builtins::x_apply_result_mode): Change type to
2024 fixed_size_mode_pod.
2025 * builtins.c (apply_args_size, apply_result_size, result_vector)
2026 (expand_builtin_apply_args_1, expand_builtin_apply)
2027 (expand_builtin_return): Update accordingly.
2029 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2031 * cse.c (hash_rtx_cb): Hash only the encoded elements.
2032 * cselib.c (cselib_hash_rtx): Likewise.
2033 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
2034 CONST_VECTOR encoding.
2036 2017-01-03 Jakub Jelinek <jakub@redhat.com>
2037 Jeff Law <law@redhat.com>
2040 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
2041 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
2042 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
2043 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
2046 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
2047 explicitly probe *sp in a noreturn function if there were any callee
2048 register saves or frame pointer is needed.
2050 2018-01-03 Jakub Jelinek <jakub@redhat.com>
2053 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
2054 BLKmode for ternary, binary or unary expressions.
2057 * var-tracking.c (delete_vta_debug_insn): New inline function.
2058 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
2059 insns from get_insns () to NULL instead of each bb separately.
2060 Use delete_vta_debug_insn. No longer static.
2061 (vt_debug_insns_local, variable_tracking_main_1): Adjust
2062 delete_vta_debug_insns callers.
2063 * rtl.h (delete_vta_debug_insns): Declare.
2064 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
2065 instead of variable_tracking_main.
2067 2018-01-03 Martin Sebor <msebor@redhat.com>
2069 PR tree-optimization/83603
2070 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
2071 arguments past the endof the argument list in functions declared
2072 without a prototype.
2073 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
2074 Avoid checking when arguments are null.
2076 2018-01-03 Martin Sebor <msebor@redhat.com>
2079 * doc/extend.texi (attribute const): Fix a typo.
2080 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
2081 issuing -Wsuggest-attribute for void functions.
2083 2018-01-03 Martin Sebor <msebor@redhat.com>
2085 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
2086 offset_int::from instead of wide_int::to_shwi.
2087 (maybe_diag_overlap): Remove assertion.
2088 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
2089 * gimple-ssa-sprintf.c (format_directive): Same.
2090 (parse_directive): Same.
2091 (sprintf_dom_walker::compute_format_length): Same.
2092 (try_substitute_return_value): Same.
2094 2017-01-03 Jeff Law <law@redhat.com>
2097 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
2098 non-constant residual for zero at runtime and avoid probing in
2099 that case. Reorganize code for trailing problem to mirror handling
2102 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
2104 PR tree-optimization/83501
2105 * tree-ssa-strlen.c (get_string_cst): New.
2106 (handle_char_store): Call get_string_cst.
2108 2018-01-03 Martin Liska <mliska@suse.cz>
2110 PR tree-optimization/83593
2111 * tree-ssa-strlen.c: Include tree-cfg.h.
2112 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
2113 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
2114 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
2116 (strlen_dom_walker::before_dom_children): Call
2117 gimple_purge_dead_eh_edges. Dump tranformation with details
2119 (strlen_dom_walker::before_dom_children): Update call by adding
2120 new argument cleanup_eh.
2121 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
2123 2018-01-03 Martin Liska <mliska@suse.cz>
2126 * cif-code.def (VARIADIC_THUNK): New enum value.
2127 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
2130 2018-01-03 Jan Beulich <jbeulich@suse.com>
2132 * sse.md (mov<mode>_internal): Tighten condition for when to use
2133 vmovdqu<ssescalarsize> for TI and OI modes.
2135 2018-01-03 Jakub Jelinek <jakub@redhat.com>
2137 Update copyright years.
2139 2018-01-03 Martin Liska <mliska@suse.cz>
2142 * ipa-visibility.c (function_and_variable_visibility): Skip
2143 functions with noipa attribure.
2145 2018-01-03 Jakub Jelinek <jakub@redhat.com>
2147 * gcc.c (process_command): Update copyright notice dates.
2148 * gcov-dump.c (print_version): Ditto.
2149 * gcov.c (print_version): Ditto.
2150 * gcov-tool.c (print_version): Ditto.
2151 * gengtype.c (create_file): Ditto.
2152 * doc/cpp.texi: Bump @copying's copyright year.
2153 * doc/cppinternals.texi: Ditto.
2154 * doc/gcc.texi: Ditto.
2155 * doc/gccint.texi: Ditto.
2156 * doc/gcov.texi: Ditto.
2157 * doc/install.texi: Ditto.
2158 * doc/invoke.texi: Ditto.
2160 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2162 * vector-builder.h (vector_builder::m_full_nelts): Change from
2163 unsigned int to poly_uint64.
2164 (vector_builder::full_nelts): Update prototype accordingly.
2165 (vector_builder::new_vector): Likewise.
2166 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
2167 (vector_builder::operator ==): Likewise.
2168 (vector_builder::finalize): Likewise.
2169 * int-vector-builder.h (int_vector_builder::int_vector_builder):
2170 Take the number of elements as a poly_uint64 rather than an
2172 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
2173 from unsigned int to poly_uint64.
2174 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
2175 (vec_perm_indices::new_vector): Likewise.
2176 (vec_perm_indices::length): Likewise.
2177 (vec_perm_indices::nelts_per_input): Likewise.
2178 (vec_perm_indices::input_nelts): Likewise.
2179 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
2180 number of elements per input as a poly_uint64 rather than an
2181 unsigned int. Use the original encoding for variable-length
2182 vectors, rather than clamping each individual element.
2183 For the second and subsequent elements in each pattern,
2184 clamp the step and base before clamping their sum.
2185 (vec_perm_indices::series_p): Handle polynomial element counts.
2186 (vec_perm_indices::all_in_range_p): Likewise.
2187 (vec_perm_indices_to_tree): Likewise.
2188 (vec_perm_indices_to_rtx): Likewise.
2189 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
2190 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
2191 (tree_vector_builder::new_binary_operation): Handle polynomial
2192 element counts. Return false if we need to know the number
2193 of elements at compile time.
2194 * fold-const.c (fold_vec_perm): Punt if the number of elements
2195 isn't known at compile time.
2197 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2199 * vec-perm-indices.h (vec_perm_builder): Change element type
2200 from HOST_WIDE_INT to poly_int64.
2201 (vec_perm_indices::element_type): Update accordingly.
2202 (vec_perm_indices::clamp): Handle polynomial element_types.
2203 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
2204 (vec_perm_indices::all_in_range_p): Likewise.
2205 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
2207 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
2208 polynomial vec_perm_indices element types.
2209 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
2210 * fold-const.c (fold_vec_perm): Likewise.
2211 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
2212 * tree-vect-generic.c (lower_vec_perm): Likewise.
2213 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
2214 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
2215 element type to HOST_WIDE_INT.
2217 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2218 Alan Hayward <alan.hayward@arm.com>
2219 David Sherwood <david.sherwood@arm.com>
2221 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
2222 rather than an int. Use plus_constant.
2223 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
2224 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
2226 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2227 Alan Hayward <alan.hayward@arm.com>
2228 David Sherwood <david.sherwood@arm.com>
2230 * calls.c (emit_call_1, expand_call): Change struct_value_size from
2231 a HOST_WIDE_INT to a poly_int64.
2233 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2234 Alan Hayward <alan.hayward@arm.com>
2235 David Sherwood <david.sherwood@arm.com>
2237 * calls.c (load_register_parameters): Cope with polynomial
2238 mode sizes. Require a constant size for BLKmode parameters
2239 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
2240 forces a parameter to be padded at the lsb end in order to
2241 fill a complete number of words, require the parameter size
2242 to be ordered wrt UNITS_PER_WORD.
2244 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2245 Alan Hayward <alan.hayward@arm.com>
2246 David Sherwood <david.sherwood@arm.com>
2248 * reload1.c (spill_stack_slot_width): Change element type
2249 from unsigned int to poly_uint64_pod.
2250 (alter_reg): Treat mode sizes as polynomial.
2252 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2253 Alan Hayward <alan.hayward@arm.com>
2254 David Sherwood <david.sherwood@arm.com>
2256 * reload.c (complex_word_subreg_p): New function.
2257 (reload_inner_reg_of_subreg, push_reload): Use it.
2259 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2260 Alan Hayward <alan.hayward@arm.com>
2261 David Sherwood <david.sherwood@arm.com>
2263 * lra-constraints.c (process_alt_operands): Reject matched
2264 operands whose sizes aren't ordered.
2265 (match_reload): Refer to this check here.
2267 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2268 Alan Hayward <alan.hayward@arm.com>
2269 David Sherwood <david.sherwood@arm.com>
2271 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
2272 that the mode size is in the set {1, 2, 4, 8, 16}.
2274 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2275 Alan Hayward <alan.hayward@arm.com>
2276 David Sherwood <david.sherwood@arm.com>
2278 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
2279 Use plus_constant instead of gen_rtx_PLUS.
2281 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2282 Alan Hayward <alan.hayward@arm.com>
2283 David Sherwood <david.sherwood@arm.com>
2285 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
2286 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
2287 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
2288 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
2289 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
2290 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
2291 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
2292 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
2293 * config/i386/i386.c (ix86_push_rounding): ...this new function.
2294 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
2296 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
2297 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
2298 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
2299 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
2300 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
2301 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
2302 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
2303 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
2304 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
2305 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
2307 * expr.c (emit_move_resolve_push): Treat the input and result
2308 of PUSH_ROUNDING as a poly_int64.
2309 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
2310 (emit_push_insn): Likewise.
2311 * lra-eliminations.c (mark_not_eliminable): Likewise.
2312 * recog.c (push_operand): Likewise.
2313 * reload1.c (elimination_effects): Likewise.
2314 * rtlanal.c (nonzero_bits1): Likewise.
2315 * calls.c (store_one_arg): Likewise. Require the padding to be
2316 known at compile time.
2318 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2319 Alan Hayward <alan.hayward@arm.com>
2320 David Sherwood <david.sherwood@arm.com>
2322 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
2323 Use plus_constant instead of gen_rtx_PLUS.
2325 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2326 Alan Hayward <alan.hayward@arm.com>
2327 David Sherwood <david.sherwood@arm.com>
2329 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
2332 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2333 Alan Hayward <alan.hayward@arm.com>
2334 David Sherwood <david.sherwood@arm.com>
2336 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
2337 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
2338 via stack temporaries. Treat the mode size as polynomial too.
2340 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2341 Alan Hayward <alan.hayward@arm.com>
2342 David Sherwood <david.sherwood@arm.com>
2344 * expr.c (expand_expr_real_2): When handling conversions involving
2345 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
2346 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
2347 as a poly_uint64 too.
2349 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2350 Alan Hayward <alan.hayward@arm.com>
2351 David Sherwood <david.sherwood@arm.com>
2353 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
2355 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2356 Alan Hayward <alan.hayward@arm.com>
2357 David Sherwood <david.sherwood@arm.com>
2359 * combine.c (can_change_dest_mode): Handle polynomial
2360 REGMODE_NATURAL_SIZE.
2361 * expmed.c (store_bit_field_1): Likewise.
2362 * expr.c (store_constructor): Likewise.
2363 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
2364 and polynomial REGMODE_NATURAL_SIZE.
2365 (gen_lowpart_common): Likewise.
2366 * reginfo.c (record_subregs_of_mode): Likewise.
2367 * rtlanal.c (read_modify_subreg_p): Likewise.
2369 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2370 Alan Hayward <alan.hayward@arm.com>
2371 David Sherwood <david.sherwood@arm.com>
2373 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
2374 numbers of elements.
2376 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2377 Alan Hayward <alan.hayward@arm.com>
2378 David Sherwood <david.sherwood@arm.com>
2380 * match.pd: Cope with polynomial numbers of vector elements.
2382 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2383 Alan Hayward <alan.hayward@arm.com>
2384 David Sherwood <david.sherwood@arm.com>
2386 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
2387 in a POINTER_PLUS_EXPR.
2389 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2390 Alan Hayward <alan.hayward@arm.com>
2391 David Sherwood <david.sherwood@arm.com>
2393 * omp-simd-clone.c (simd_clone_subparts): New function.
2394 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
2395 (ipa_simd_modify_function_body): 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-vect-generic.c (nunits_for_known_piecewise_op): New function.
2402 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
2403 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
2404 (expand_vector_condition, vector_element): Likewise.
2405 (subparts_gt): New function.
2406 (get_compute_type): Use subparts_gt.
2407 (count_type_subparts): Delete.
2408 (expand_vector_operations_1): Use subparts_gt instead of
2409 count_type_subparts.
2411 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2412 Alan Hayward <alan.hayward@arm.com>
2413 David Sherwood <david.sherwood@arm.com>
2415 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
2416 (vect_compile_time_alias): ...this new function. Do the calculation
2417 on poly_ints rather than trees.
2418 (vect_prune_runtime_alias_test_list): Update call accordingly.
2420 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2421 Alan Hayward <alan.hayward@arm.com>
2422 David Sherwood <david.sherwood@arm.com>
2424 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
2426 (vect_schedule_slp_instance): Likewise.
2428 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2429 Alan Hayward <alan.hayward@arm.com>
2430 David Sherwood <david.sherwood@arm.com>
2432 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
2433 constant and extern definitions for variable-length vectors.
2434 (vect_get_constant_vectors): Note that the number of units
2435 is known to be constant.
2437 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2438 Alan Hayward <alan.hayward@arm.com>
2439 David Sherwood <david.sherwood@arm.com>
2441 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
2442 of units as polynomial. Choose between WIDE and NARROW based
2445 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2446 Alan Hayward <alan.hayward@arm.com>
2447 David Sherwood <david.sherwood@arm.com>
2449 * tree-vect-stmts.c (simd_clone_subparts): New function.
2450 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
2452 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2453 Alan Hayward <alan.hayward@arm.com>
2454 David Sherwood <david.sherwood@arm.com>
2456 * tree-vect-stmts.c (vectorizable_call): Treat the number of
2457 vectors as polynomial. Use build_index_vector for
2460 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2461 Alan Hayward <alan.hayward@arm.com>
2462 David Sherwood <david.sherwood@arm.com>
2464 * tree-vect-stmts.c (get_load_store_type): Treat the number of
2465 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
2466 for variable-length vectors.
2467 (vectorizable_mask_load_store): Treat the number of units as
2468 polynomial, asserting that it is constant if the condition has
2469 already been enforced.
2470 (vectorizable_store, vectorizable_load): Likewise.
2472 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2473 Alan Hayward <alan.hayward@arm.com>
2474 David Sherwood <david.sherwood@arm.com>
2476 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
2477 of units as polynomial. Punt if we can't tell at compile time
2478 which vector contains the final result.
2480 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2481 Alan Hayward <alan.hayward@arm.com>
2482 David Sherwood <david.sherwood@arm.com>
2484 * tree-vect-loop.c (vectorizable_induction): Treat the number
2485 of units as polynomial. Punt on SLP inductions. Use an integer
2486 VEC_SERIES_EXPR for variable-length integer reductions. Use a
2487 cast of such a series for variable-length floating-point
2490 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2491 Alan Hayward <alan.hayward@arm.com>
2492 David Sherwood <david.sherwood@arm.com>
2494 * tree.h (build_index_vector): Declare.
2495 * tree.c (build_index_vector): New function.
2496 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
2497 of units as polynomial, forcibly converting it to a constant if
2498 vectorizable_reduction has already enforced the condition.
2499 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
2500 to create a {1,2,3,...} vector.
2501 (vectorizable_reduction): Treat the number of units as polynomial.
2502 Choose vectype_in based on the largest scalar element size rather
2503 than the smallest number of units. Enforce the restrictions
2506 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2507 Alan Hayward <alan.hayward@arm.com>
2508 David Sherwood <david.sherwood@arm.com>
2510 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
2511 number of units as polynomial.
2513 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2514 Alan Hayward <alan.hayward@arm.com>
2515 David Sherwood <david.sherwood@arm.com>
2517 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
2518 * target.def (autovectorize_vector_sizes): Return the vector sizes
2519 by pointer, using vector_sizes rather than a bitmask.
2520 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
2521 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
2522 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
2524 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
2525 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
2526 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
2527 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
2528 * omp-general.c (omp_max_vf): Likewise.
2529 * omp-low.c (omp_clause_aligned_alignment): Likewise.
2530 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
2531 * tree-vect-loop.c (vect_analyze_loop): Likewise.
2532 * tree-vect-slp.c (vect_slp_bb): Likewise.
2533 * doc/tm.texi: Regenerate.
2534 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
2536 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
2537 the vector size as a poly_uint64 rather than an unsigned int.
2538 (current_vector_size): Change from an unsigned int to a poly_uint64.
2539 (get_vectype_for_scalar_type): Update accordingly.
2540 * tree.h (build_truth_vector_type): Take the size and number of
2541 units as a poly_uint64 rather than an unsigned int.
2542 (build_vector_type): Add a temporary overload that takes
2543 the number of units as a poly_uint64 rather than an unsigned int.
2544 * tree.c (make_vector_type): Likewise.
2545 (build_truth_vector_type): Take the number of units as a poly_uint64
2546 rather than an unsigned int.
2548 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2549 Alan Hayward <alan.hayward@arm.com>
2550 David Sherwood <david.sherwood@arm.com>
2552 * target.def (get_mask_mode): Take the number of units and length
2553 as poly_uint64s rather than unsigned ints.
2554 * targhooks.h (default_get_mask_mode): Update accordingly.
2555 * targhooks.c (default_get_mask_mode): Likewise.
2556 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
2557 * doc/tm.texi: Regenerate.
2559 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2560 Alan Hayward <alan.hayward@arm.com>
2561 David Sherwood <david.sherwood@arm.com>
2563 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
2564 * omp-general.c (omp_max_vf): Likewise.
2565 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
2566 (expand_omp_simd): Handle polynomial safelen.
2567 * omp-low.c (omplow_simd_context): Add a default constructor.
2568 (omplow_simd_context::max_vf): Change from int to poly_uint64.
2569 (lower_rec_simd_input_clauses): Update accordingly.
2570 (lower_rec_input_clauses): Likewise.
2572 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2573 Alan Hayward <alan.hayward@arm.com>
2574 David Sherwood <david.sherwood@arm.com>
2576 * tree-vectorizer.h (vect_nunits_for_cost): New function.
2577 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
2578 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
2579 (vect_analyze_slp_cost): Likewise.
2580 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
2581 (vect_model_load_cost): Likewise.
2583 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2584 Alan Hayward <alan.hayward@arm.com>
2585 David Sherwood <david.sherwood@arm.com>
2587 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
2588 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
2589 from an unsigned int * to a poly_uint64_pod *.
2590 (calculate_unrolling_factor): New function.
2591 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
2593 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2594 Alan Hayward <alan.hayward@arm.com>
2595 David Sherwood <david.sherwood@arm.com>
2597 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
2598 from an unsigned int to a poly_uint64.
2599 (_loop_vec_info::slp_unrolling_factor): Likewise.
2600 (_loop_vec_info::vectorization_factor): Change from an int
2602 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
2603 (vect_get_num_vectors): New function.
2604 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
2605 (vect_get_num_copies): Use vect_get_num_vectors.
2606 (vect_analyze_data_ref_dependences): Change max_vf from an int *
2607 to an unsigned int *.
2608 (vect_analyze_data_refs): Change min_vf from an int * to a
2610 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
2611 than an unsigned HOST_WIDE_INT.
2612 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
2613 (vect_analyze_data_ref_dependence): Change max_vf from an int *
2614 to an unsigned int *.
2615 (vect_analyze_data_ref_dependences): Likewise.
2616 (vect_compute_data_ref_alignment): Handle polynomial vf.
2617 (vect_enhance_data_refs_alignment): Likewise.
2618 (vect_prune_runtime_alias_test_list): Likewise.
2619 (vect_shift_permute_load_chain): Likewise.
2620 (vect_supportable_dr_alignment): Likewise.
2621 (dependence_distance_ge_vf): Take the vectorization factor as a
2622 poly_uint64 rather than an unsigned HOST_WIDE_INT.
2623 (vect_analyze_data_refs): Change min_vf from an int * to a
2625 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
2626 vfm1 as a poly_uint64 rather than an int. Make the same change
2627 for the returned bound_scalar.
2628 (vect_gen_vector_loop_niters): Handle polynomial vf.
2629 (vect_do_peeling): Likewise. Update call to
2630 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
2631 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
2633 * tree-vect-loop.c (vect_determine_vectorization_factor)
2634 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
2635 (vect_get_known_peeling_cost): Likewise.
2636 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
2637 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
2638 (vect_transform_loop): Likewise. Use the lowest possible VF when
2639 updating the upper bounds of the loop.
2640 (vect_min_worthwhile_factor): Make static. Return an unsigned int
2642 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
2643 polynomial unroll factors.
2644 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
2645 (vect_make_slp_decision): Likewise.
2646 (vect_supported_load_permutation_p): Likewise, and polynomial
2648 (vect_analyze_slp_cost): Handle polynomial vf.
2649 (vect_slp_analyze_node_operations): Likewise.
2650 (vect_slp_analyze_bb_1): Likewise.
2651 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
2652 than an unsigned HOST_WIDE_INT.
2653 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
2654 (vectorizable_load): Handle polynomial vf.
2655 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
2657 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
2659 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2660 Alan Hayward <alan.hayward@arm.com>
2661 David Sherwood <david.sherwood@arm.com>
2663 * match.pd: Handle bit operations involving three constants
2664 and try to fold one pair.
2666 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2668 * tree-vect-loop-manip.c: Include gimple-fold.h.
2669 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
2670 niters_maybe_zero parameters. Handle other cases besides a step of 1.
2671 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
2672 Add a path that uses a step of VF instead of 1, but disable it
2674 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
2675 and niters_no_overflow parameters. Update calls to
2676 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
2677 Create a new SSA name if the latter choses to use a ste other
2678 than zero, and return it via niters_vector_mult_vf_var.
2679 * tree-vect-loop.c (vect_transform_loop): Update calls to
2680 vect_do_peeling, vect_gen_vector_loop_niters and
2681 slpeel_make_loop_iterate_ntimes.
2682 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
2683 (vect_gen_vector_loop_niters): Update declarations after above changes.
2685 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
2687 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
2688 128-bit round to integer instructions.
2689 (ceil<mode>2): Likewise.
2690 (btrunc<mode>2): Likewise.
2691 (round<mode>2): Likewise.
2693 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2695 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
2696 unaligned VSX load/store on P8/P9.
2697 (expand_block_clear): Allow the use of unaligned VSX
2698 load/store on P8/P9.
2700 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2702 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
2704 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
2705 swap associated with both a load and a store.
2707 2018-01-02 Andrew Waterman <andrew@sifive.com>
2709 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
2710 * config/riscv/riscv.md (clear_cache): Use it.
2712 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
2714 * web.c: Remove out-of-date comment.
2716 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2718 * expr.c (fixup_args_size_notes): Check that any existing
2719 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
2720 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
2721 (emit_single_push_insn): ...here.
2723 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2725 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
2726 (const_vector_encoded_nelts): New function.
2727 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
2728 (const_vector_int_elt, const_vector_elt): Declare.
2729 * emit-rtl.c (const_vector_int_elt_1): New function.
2730 (const_vector_elt): Likewise.
2731 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
2732 of CONST_VECTOR_ELT.
2734 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2736 * expr.c: Include rtx-vector-builder.h.
2737 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
2738 directly on the tree encoding.
2739 (const_vector_from_tree): Likewise.
2740 * optabs.c: Include rtx-vector-builder.h.
2741 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
2742 sequence of "u" values.
2743 * vec-perm-indices.c: Include rtx-vector-builder.h.
2744 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
2745 directly on the vec_perm_indices encoding.
2747 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2749 * doc/rtl.texi (const_vector): Describe new encoding scheme.
2750 * Makefile.in (OBJS): Add rtx-vector-builder.o.
2751 * rtx-vector-builder.h: New file.
2752 * rtx-vector-builder.c: Likewise.
2753 * rtl.h (rtx_def::u2): Add a const_vector field.
2754 (CONST_VECTOR_NPATTERNS): New macro.
2755 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
2756 (CONST_VECTOR_DUPLICATE_P): Likewise.
2757 (CONST_VECTOR_STEPPED_P): Likewise.
2758 (CONST_VECTOR_ENCODED_ELT): Likewise.
2759 (const_vec_duplicate_p): Check for a duplicated vector encoding.
2760 (unwrap_const_vec_duplicate): Likewise.
2761 (const_vec_series_p): Check for a non-duplicated vector encoding.
2762 Say that the function only returns true for integer vectors.
2763 * emit-rtl.c: Include rtx-vector-builder.h.
2764 (gen_const_vec_duplicate_1): Delete.
2765 (gen_const_vector): Call gen_const_vec_duplicate instead of
2766 gen_const_vec_duplicate_1.
2767 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
2768 (gen_const_vec_duplicate): Use rtx_vector_builder.
2769 (gen_const_vec_series): Likewise.
2770 (gen_rtx_CONST_VECTOR): Likewise.
2771 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
2772 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
2773 Build a new vector rather than modifying a CONST_VECTOR in-place.
2774 (handle_special_swappables): Update call accordingly.
2775 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
2776 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
2777 Build a new vector rather than modifying a CONST_VECTOR in-place.
2778 (handle_special_swappables): Update call accordingly.
2780 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2782 * simplify-rtx.c (simplify_const_binary_operation): Use
2783 CONST_VECTOR_ELT instead of XVECEXP.
2785 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2787 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
2788 the selector elements to be different from the data elements
2789 if the selector is a VECTOR_CST.
2790 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
2791 ssizetype for the selector.
2793 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2795 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
2796 before testing each element individually.
2797 * tree-vect-generic.c (lower_vec_perm): Likewise.
2799 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2801 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
2802 * selftest-run-tests.c (selftest::run_tests): Call it.
2803 * vector-builder.h (vector_builder::operator ==): New function.
2804 (vector_builder::operator !=): Likewise.
2805 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
2806 (vec_perm_indices::all_from_input_p): New function.
2807 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
2808 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
2809 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
2810 instead of reading the VECTOR_CST directly. Detect whether both
2811 vector inputs are the same before constructing the vec_perm_indices,
2812 and update the number of inputs argument accordingly. Use the
2813 utility functions added above. Only construct sel2 if we need to.
2815 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2817 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
2818 the broadcast of the low byte.
2819 (expand_mult_highpart): Use an explicit encoding for the permutes.
2820 * optabs-query.c (can_mult_highpart_p): Likewise.
2821 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
2822 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
2823 (vectorizable_bswap): Likewise.
2824 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
2825 explicit encoding for the power-of-2 permutes.
2826 (vect_permute_store_chain): Likewise.
2827 (vect_grouped_load_supported): Likewise.
2828 (vect_permute_load_chain): Likewise.
2830 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2832 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
2833 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
2834 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
2835 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
2836 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
2837 (vect_gen_perm_mask_any): Likewise.
2839 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2841 * int-vector-builder.h: New file.
2842 * vec-perm-indices.h: Include int-vector-builder.h.
2843 (vec_perm_indices): Redefine as an int_vector_builder.
2844 (auto_vec_perm_indices): Delete.
2845 (vec_perm_builder): Redefine as a stand-alone class.
2846 (vec_perm_indices::vec_perm_indices): New function.
2847 (vec_perm_indices::clamp): Likewise.
2848 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
2849 (vec_perm_indices::new_vector): New function.
2850 (vec_perm_indices::new_expanded_vector): Update for new
2851 vec_perm_indices class.
2852 (vec_perm_indices::rotate_inputs): New function.
2853 (vec_perm_indices::all_in_range_p): Operate directly on the
2854 encoded form, without computing elided elements.
2855 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
2856 encoding. Update for new vec_perm_indices class.
2857 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
2858 the given vec_perm_builder.
2859 (expand_vec_perm_var): Update vec_perm_builder constructor.
2860 (expand_mult_highpart): Use vec_perm_builder instead of
2861 auto_vec_perm_indices.
2862 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
2863 vec_perm_indices instead of auto_vec_perm_indices. Use a single
2864 or double series encoding as appropriate.
2865 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
2866 vec_perm_indices instead of auto_vec_perm_indices.
2867 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
2868 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
2869 (vect_permute_store_chain): Likewise.
2870 (vect_grouped_load_supported): Likewise.
2871 (vect_permute_load_chain): Likewise.
2872 (vect_shift_permute_load_chain): Likewise.
2873 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
2874 (vect_transform_slp_perm_load): Likewise.
2875 (vect_schedule_slp_instance): Likewise.
2876 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
2877 (vectorizable_mask_load_store): Likewise.
2878 (vectorizable_bswap): Likewise.
2879 (vectorizable_store): Likewise.
2880 (vectorizable_load): Likewise.
2881 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
2882 vec_perm_indices instead of auto_vec_perm_indices. Use
2883 tree_to_vec_perm_builder to read the vector from a tree.
2884 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
2885 vec_perm_builder instead of a vec_perm_indices.
2886 (have_whole_vector_shift): Use vec_perm_builder and
2887 vec_perm_indices instead of auto_vec_perm_indices. Leave the
2888 truncation to calc_vec_perm_mask_for_shift.
2889 (vect_create_epilog_for_reduction): Likewise.
2890 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
2891 from auto_vec_perm_indices to vec_perm_indices.
2892 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
2893 instead of changing individual elements.
2894 (aarch64_vectorize_vec_perm_const): Use new_vector to install
2895 the vector in d.perm.
2896 * config/arm/arm.c (expand_vec_perm_d::perm): Change
2897 from auto_vec_perm_indices to vec_perm_indices.
2898 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
2899 instead of changing individual elements.
2900 (arm_vectorize_vec_perm_const): Use new_vector to install
2901 the vector in d.perm.
2902 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
2903 Update vec_perm_builder constructor.
2904 (rs6000_expand_interleave): Likewise.
2905 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
2906 (rs6000_expand_interleave): Likewise.
2908 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2910 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
2911 to qimode could truncate the indices.
2912 * optabs.c (expand_vec_perm_var): Likewise.
2914 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2916 * Makefile.in (OBJS): Add vec-perm-indices.o.
2917 * vec-perm-indices.h: New file.
2918 * vec-perm-indices.c: Likewise.
2919 * target.h (vec_perm_indices): Replace with a forward class
2921 (auto_vec_perm_indices): Move to vec-perm-indices.h.
2922 * optabs.h: Include vec-perm-indices.h.
2923 (expand_vec_perm): Delete.
2924 (selector_fits_mode_p, expand_vec_perm_var): Declare.
2925 (expand_vec_perm_const): Declare.
2926 * target.def (vec_perm_const_ok): Replace with...
2927 (vec_perm_const): ...this new hook.
2928 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
2929 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
2930 * doc/tm.texi: Regenerate.
2931 * optabs.def (vec_perm_const): Delete.
2932 * doc/md.texi (vec_perm_const): Likewise.
2933 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
2934 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
2935 expand_vec_perm for constant permutation vectors. Assert that
2936 the mode of variable permutation vectors is the integer equivalent
2937 of the mode that is being permuted.
2938 * optabs-query.h (selector_fits_mode_p): Declare.
2939 * optabs-query.c: Include vec-perm-indices.h.
2940 (selector_fits_mode_p): New function.
2941 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
2942 is defined, instead of checking whether the vec_perm_const_optab
2943 exists. Use targetm.vectorize.vec_perm_const instead of
2944 targetm.vectorize.vec_perm_const_ok. Check whether the indices
2945 fit in the vector mode before using a variable permute.
2946 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
2947 vec_perm_indices instead of an rtx.
2948 (expand_vec_perm): Replace with...
2949 (expand_vec_perm_const): ...this new function. Take the selector
2950 as a vec_perm_indices rather than an rtx. Also take the mode of
2951 the selector. Update call to shift_amt_for_vec_perm_mask.
2952 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
2953 Use vec_perm_indices::new_expanded_vector to expand the original
2954 selector into bytes. Check whether the indices fit in the vector
2955 mode before using a variable permute.
2956 (expand_vec_perm_var): Make global.
2957 (expand_mult_highpart): Use expand_vec_perm_const.
2958 * fold-const.c: Includes vec-perm-indices.h.
2959 * tree-ssa-forwprop.c: Likewise.
2960 * tree-vect-data-refs.c: Likewise.
2961 * tree-vect-generic.c: Likewise.
2962 * tree-vect-loop.c: Likewise.
2963 * tree-vect-slp.c: Likewise.
2964 * tree-vect-stmts.c: Likewise.
2965 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
2967 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
2968 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
2969 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
2970 (aarch64_vectorize_vec_perm_const): ...this new function.
2971 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
2972 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2973 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
2974 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
2975 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
2976 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2977 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
2979 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
2980 check for NEON modes.
2981 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
2982 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
2983 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
2984 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
2986 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
2987 the old VEC_PERM_CONST conditions.
2988 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
2989 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
2990 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
2991 (ia64_vectorize_vec_perm_const_ok): Merge into...
2992 (ia64_vectorize_vec_perm_const): ...this new function.
2993 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
2994 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
2995 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
2996 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
2997 * config/mips/mips.c (mips_expand_vec_perm_const)
2998 (mips_vectorize_vec_perm_const_ok): Merge into...
2999 (mips_vectorize_vec_perm_const): ...this new function.
3000 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
3001 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
3002 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
3003 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
3004 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
3005 (rs6000_expand_vec_perm_const): Delete.
3006 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
3008 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
3009 (altivec_expand_vec_perm_const_le): Take each operand individually.
3010 Operate on constant selectors rather than rtxes.
3011 (altivec_expand_vec_perm_const): Likewise. Update call to
3012 altivec_expand_vec_perm_const_le.
3013 (rs6000_expand_vec_perm_const): Delete.
3014 (rs6000_vectorize_vec_perm_const_ok): Delete.
3015 (rs6000_vectorize_vec_perm_const): New function.
3016 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
3017 an element count and rtx array.
3018 (rs6000_expand_extract_even): Update call accordingly.
3019 (rs6000_expand_interleave): Likewise.
3020 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
3021 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
3022 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
3023 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
3024 (rs6000_expand_vec_perm_const): Delete.
3025 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
3026 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
3027 (altivec_expand_vec_perm_const_le): Take each operand individually.
3028 Operate on constant selectors rather than rtxes.
3029 (altivec_expand_vec_perm_const): Likewise. Update call to
3030 altivec_expand_vec_perm_const_le.
3031 (rs6000_expand_vec_perm_const): Delete.
3032 (rs6000_vectorize_vec_perm_const_ok): Delete.
3033 (rs6000_vectorize_vec_perm_const): New function. Remove stray
3034 reference to the SPE evmerge intructions.
3035 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
3036 an element count and rtx array.
3037 (rs6000_expand_extract_even): Update call accordingly.
3038 (rs6000_expand_interleave): Likewise.
3039 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
3040 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
3042 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
3044 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3046 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
3047 vector mode and that that mode matches the mode of the data
3049 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
3050 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
3051 directly using expand_vec_perm_1 when forcing selectors into
3053 (expand_vec_perm_var): New function, split out from expand_vec_perm.
3055 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3057 * optabs-query.h (can_vec_perm_p): Delete.
3058 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
3059 * optabs-query.c (can_vec_perm_p): Split into...
3060 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
3061 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
3062 particular selector is valid.
3063 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
3064 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
3065 (vect_grouped_load_supported): Likewise.
3066 (vect_shift_permute_load_chain): Likewise.
3067 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
3068 (vect_transform_slp_perm_load): Likewise.
3069 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
3070 (vectorizable_bswap): Likewise.
3071 (vect_gen_perm_mask_checked): Likewise.
3072 * fold-const.c (fold_ternary_loc): Likewise. Don't take
3073 implementations of variable permutation vectors into account
3074 when deciding which selector to use.
3075 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
3076 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
3077 with a false third argument.
3078 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
3079 to test whether the constant selector is valid and can_vec_perm_var_p
3080 to test whether a variable selector is valid.
3082 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3084 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
3085 * optabs-query.c (can_vec_perm_p): Likewise.
3086 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
3087 instead of vec_perm_indices.
3088 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
3089 (vect_gen_perm_mask_checked): Likewise,
3090 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
3091 (vect_gen_perm_mask_checked): Likewise,
3093 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
3095 * optabs-query.h (qimode_for_vec_perm): Declare.
3096 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
3097 (qimode_for_vec_perm): ...this new function.
3098 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
3100 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3102 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
3103 does not have a conditional at the top.
3105 2018-01-02 Richard Biener <rguenther@suse.de>
3107 * ipa-inline.c (big_speedup_p): Fix expression.
3109 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
3112 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
3115 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
3119 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
3120 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
3121 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
3122 cond_taken_branch_cost 3->4.
3124 2018-01-01 Jakub Jelinek <jakub@redhat.com>
3126 PR tree-optimization/83581
3127 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
3128 TODO_cleanup_cfg if any changes have been made.
3131 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
3132 convert_modes if target mode has the right side, but different mode
3136 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
3137 last argument when extracting from CONCAT. If either from_real or
3138 from_imag is NULL, use expansion through memory. If result is not
3139 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
3140 the parts directly to inner mode, if even that fails, use expansion
3144 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
3145 check for bswap in mode rather than HImode and use that in expand_unop
3148 Copyright (C) 2018 Free Software Foundation, Inc.
3150 Copying and distribution of this file, with or without modification,
3151 are permitted in any medium without royalty provided the copyright
3152 notice and this notice are preserved.