1 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2 Alan Hayward <alan.hayward@arm.com>
3 David Sherwood <david.sherwood@arm.com>
5 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
6 polynomial numbers of units.
7 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
8 (valid_vector_subparts_p): New function.
9 (build_vector_type): Remove temporary shim and take the number
10 of units as a poly_uint64 rather than an int.
11 (build_opaque_vector_type): Take the number of units as a
12 poly_uint64 rather than an int.
13 * tree.c (build_vector_from_ctor): Handle polynomial
15 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
16 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
17 (build_vector_from_val): If the number of units is variable,
18 use build_vec_duplicate_cst for constant operands and
19 VEC_DUPLICATE_EXPR otherwise.
20 (make_vector_type): Remove temporary is_constant ().
21 (build_vector_type, build_opaque_vector_type): Take the number of
22 units as a poly_uint64 rather than an int.
23 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
25 * cfgexpand.c (expand_debug_expr): Likewise.
26 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
27 (store_constructor, expand_expr_real_1): Likewise.
28 (const_scalar_mask_from_tree): Likewise.
29 * fold-const-call.c (fold_const_reduction): Likewise.
30 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
31 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
32 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
33 (fold_relational_const): Likewise.
34 (native_interpret_vector): Likewise. Change the size from an
35 int to an unsigned int.
36 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
38 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
39 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
40 duplicating a non-constant operand into a variable-length vector.
41 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
42 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
43 * ipa-icf.c (sem_variable::equals): Likewise.
45 * omp-simd-clone.c (simd_clone_subparts): Likewise.
46 * print-tree.c (print_node): Likewise.
47 * stor-layout.c (layout_type): Likewise.
48 * targhooks.c (default_builtin_vectorization_cost): Likewise.
49 * tree-cfg.c (verify_gimple_comparison): Likewise.
50 (verify_gimple_assign_binary): Likewise.
51 (verify_gimple_assign_ternary): Likewise.
52 (verify_gimple_assign_single): Likewise.
53 * tree-pretty-print.c (dump_generic_node): Likewise.
54 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
55 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
56 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
57 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
58 (vect_shift_permute_load_chain): Likewise.
59 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
60 (expand_vector_condition, optimize_vector_constructor): Likewise.
61 (lower_vec_perm, get_compute_type): Likewise.
62 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
63 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
64 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
65 (vect_recog_mask_conversion_pattern): Likewise.
66 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
67 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
68 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
69 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
70 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
71 (vectorizable_shift, vectorizable_operation, vectorizable_store)
72 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
73 (supportable_widening_operation): Likewise.
74 (supportable_narrowing_operation): Likewise.
75 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
77 * varasm.c (output_constant): Likewise.
79 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
80 Alan Hayward <alan.hayward@arm.com>
81 David Sherwood <david.sherwood@arm.com>
83 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
84 so that both the length == 3 and length != 3 cases set up their
85 own permute vectors. Add comments explaining why we know the
86 number of elements is constant.
87 (vect_permute_load_chain): Likewise.
89 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
90 Alan Hayward <alan.hayward@arm.com>
91 David Sherwood <david.sherwood@arm.com>
93 * machmode.h (mode_nunits): Change from unsigned char to
95 (ONLY_FIXED_SIZE_MODES): New macro.
96 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
97 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
98 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
100 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
101 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
102 or if measurement_type is not polynomial.
103 * genmodes.c (ZERO_COEFFS): New macro.
104 (emit_mode_nunits_inline): Make mode_nunits_inline return a
106 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
107 Use ZERO_COEFFS when emitting initializers.
108 * data-streamer.h (bp_pack_poly_value): New function.
109 (bp_unpack_poly_value): Likewise.
110 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
112 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
114 * tree.c (make_vector_type): Remove temporary shim and make
115 the real function take the number of units as a poly_uint64
117 (build_vector_type_for_mode): Handle polynomial nunits.
118 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
119 * emit-rtl.c (const_vec_series_p_1): Likewise.
120 (gen_rtx_CONST_VECTOR): Likewise.
121 * fold-const.c (test_vec_duplicate_folding): Likewise.
122 * genrecog.c (validate_pattern): Likewise.
123 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
124 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
125 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
126 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
127 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
128 * rtlanal.c (subreg_get_info): Likewise.
129 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
130 (vect_grouped_load_supported): Likewise.
131 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
132 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
133 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
134 (simplify_const_unary_operation, simplify_binary_operation_1)
135 (simplify_const_binary_operation, simplify_ternary_operation)
136 (test_vector_ops_duplicate, test_vector_ops): Likewise.
137 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
138 instead of CONST_VECTOR_NUNITS.
139 * varasm.c (output_constant_pool_2): Likewise.
140 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
141 explicit-encoded elements in the XVEC for variable-length vectors.
143 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
145 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
147 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
148 Alan Hayward <alan.hayward@arm.com>
149 David Sherwood <david.sherwood@arm.com>
151 * coretypes.h (fixed_size_mode): Declare.
152 (fixed_size_mode_pod): New typedef.
153 * builtins.h (target_builtins::x_apply_args_mode)
154 (target_builtins::x_apply_result_mode): Change type to
156 * builtins.c (apply_args_size, apply_result_size, result_vector)
157 (expand_builtin_apply_args_1, expand_builtin_apply)
158 (expand_builtin_return): Update accordingly.
160 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
162 * cse.c (hash_rtx_cb): Hash only the encoded elements.
163 * cselib.c (cselib_hash_rtx): Likewise.
164 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
165 CONST_VECTOR encoding.
167 2017-01-03 Jakub Jelinek <jakub@redhat.com>
168 Jeff Law <law@redhat.com>
171 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
172 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
173 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
174 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
177 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
178 explicitly probe *sp in a noreturn function if there were any callee
179 register saves or frame pointer is needed.
181 2018-01-03 Jakub Jelinek <jakub@redhat.com>
184 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
185 BLKmode for ternary, binary or unary expressions.
188 * var-tracking.c (delete_vta_debug_insn): New inline function.
189 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
190 insns from get_insns () to NULL instead of each bb separately.
191 Use delete_vta_debug_insn. No longer static.
192 (vt_debug_insns_local, variable_tracking_main_1): Adjust
193 delete_vta_debug_insns callers.
194 * rtl.h (delete_vta_debug_insns): Declare.
195 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
196 instead of variable_tracking_main.
198 2018-01-03 Martin Sebor <msebor@redhat.com>
201 * doc/extend.texi (attribute const): Fix a typo.
202 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
203 issuing -Wsuggest-attribute for void functions.
205 2018-01-03 Martin Sebor <msebor@redhat.com>
207 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
208 offset_int::from instead of wide_int::to_shwi.
209 (maybe_diag_overlap): Remove assertion.
210 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
211 * gimple-ssa-sprintf.c (format_directive): Same.
212 (parse_directive): Same.
213 (sprintf_dom_walker::compute_format_length): Same.
214 (try_substitute_return_value): Same.
216 2017-01-03 Jeff Law <law@redhat.com>
219 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
220 non-constant residual for zero at runtime and avoid probing in
221 that case. Reorganize code for trailing problem to mirror handling
224 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
226 PR tree-optimization/83501
227 * tree-ssa-strlen.c (get_string_cst): New.
228 (handle_char_store): Call get_string_cst.
230 2018-01-03 Martin Liska <mliska@suse.cz>
232 PR tree-optimization/83593
233 * tree-ssa-strlen.c: Include tree-cfg.h.
234 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
235 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
236 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
238 (strlen_dom_walker::before_dom_children): Call
239 gimple_purge_dead_eh_edges. Dump tranformation with details
241 (strlen_dom_walker::before_dom_children): Update call by adding
242 new argument cleanup_eh.
243 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
245 2018-01-03 Martin Liska <mliska@suse.cz>
248 * cif-code.def (VARIADIC_THUNK): New enum value.
249 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
252 2018-01-03 Jan Beulich <jbeulich@suse.com>
254 * sse.md (mov<mode>_internal): Tighten condition for when to use
255 vmovdqu<ssescalarsize> for TI and OI modes.
257 2018-01-03 Jakub Jelinek <jakub@redhat.com>
259 Update copyright years.
261 2018-01-03 Martin Liska <mliska@suse.cz>
264 * ipa-visibility.c (function_and_variable_visibility): Skip
265 functions with noipa attribure.
267 2018-01-03 Jakub Jelinek <jakub@redhat.com>
269 * gcc.c (process_command): Update copyright notice dates.
270 * gcov-dump.c (print_version): Ditto.
271 * gcov.c (print_version): Ditto.
272 * gcov-tool.c (print_version): Ditto.
273 * gengtype.c (create_file): Ditto.
274 * doc/cpp.texi: Bump @copying's copyright year.
275 * doc/cppinternals.texi: Ditto.
276 * doc/gcc.texi: Ditto.
277 * doc/gccint.texi: Ditto.
278 * doc/gcov.texi: Ditto.
279 * doc/install.texi: Ditto.
280 * doc/invoke.texi: Ditto.
282 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
284 * vector-builder.h (vector_builder::m_full_nelts): Change from
285 unsigned int to poly_uint64.
286 (vector_builder::full_nelts): Update prototype accordingly.
287 (vector_builder::new_vector): Likewise.
288 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
289 (vector_builder::operator ==): Likewise.
290 (vector_builder::finalize): Likewise.
291 * int-vector-builder.h (int_vector_builder::int_vector_builder):
292 Take the number of elements as a poly_uint64 rather than an
294 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
295 from unsigned int to poly_uint64.
296 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
297 (vec_perm_indices::new_vector): Likewise.
298 (vec_perm_indices::length): Likewise.
299 (vec_perm_indices::nelts_per_input): Likewise.
300 (vec_perm_indices::input_nelts): Likewise.
301 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
302 number of elements per input as a poly_uint64 rather than an
303 unsigned int. Use the original encoding for variable-length
304 vectors, rather than clamping each individual element.
305 For the second and subsequent elements in each pattern,
306 clamp the step and base before clamping their sum.
307 (vec_perm_indices::series_p): Handle polynomial element counts.
308 (vec_perm_indices::all_in_range_p): Likewise.
309 (vec_perm_indices_to_tree): Likewise.
310 (vec_perm_indices_to_rtx): Likewise.
311 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
312 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
313 (tree_vector_builder::new_binary_operation): Handle polynomial
314 element counts. Return false if we need to know the number
315 of elements at compile time.
316 * fold-const.c (fold_vec_perm): Punt if the number of elements
317 isn't known at compile time.
319 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
321 * vec-perm-indices.h (vec_perm_builder): Change element type
322 from HOST_WIDE_INT to poly_int64.
323 (vec_perm_indices::element_type): Update accordingly.
324 (vec_perm_indices::clamp): Handle polynomial element_types.
325 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
326 (vec_perm_indices::all_in_range_p): Likewise.
327 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
329 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
330 polynomial vec_perm_indices element types.
331 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
332 * fold-const.c (fold_vec_perm): Likewise.
333 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
334 * tree-vect-generic.c (lower_vec_perm): Likewise.
335 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
336 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
337 element type to HOST_WIDE_INT.
339 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
340 Alan Hayward <alan.hayward@arm.com>
341 David Sherwood <david.sherwood@arm.com>
343 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
344 rather than an int. Use plus_constant.
345 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
346 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
348 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
349 Alan Hayward <alan.hayward@arm.com>
350 David Sherwood <david.sherwood@arm.com>
352 * calls.c (emit_call_1, expand_call): Change struct_value_size from
353 a HOST_WIDE_INT to a poly_int64.
355 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
356 Alan Hayward <alan.hayward@arm.com>
357 David Sherwood <david.sherwood@arm.com>
359 * calls.c (load_register_parameters): Cope with polynomial
360 mode sizes. Require a constant size for BLKmode parameters
361 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
362 forces a parameter to be padded at the lsb end in order to
363 fill a complete number of words, require the parameter size
364 to be ordered wrt UNITS_PER_WORD.
366 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
367 Alan Hayward <alan.hayward@arm.com>
368 David Sherwood <david.sherwood@arm.com>
370 * reload1.c (spill_stack_slot_width): Change element type
371 from unsigned int to poly_uint64_pod.
372 (alter_reg): Treat mode sizes as polynomial.
374 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
375 Alan Hayward <alan.hayward@arm.com>
376 David Sherwood <david.sherwood@arm.com>
378 * reload.c (complex_word_subreg_p): New function.
379 (reload_inner_reg_of_subreg, push_reload): Use it.
381 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
382 Alan Hayward <alan.hayward@arm.com>
383 David Sherwood <david.sherwood@arm.com>
385 * lra-constraints.c (process_alt_operands): Reject matched
386 operands whose sizes aren't ordered.
387 (match_reload): Refer to this check here.
389 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
390 Alan Hayward <alan.hayward@arm.com>
391 David Sherwood <david.sherwood@arm.com>
393 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
394 that the mode size is in the set {1, 2, 4, 8, 16}.
396 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
397 Alan Hayward <alan.hayward@arm.com>
398 David Sherwood <david.sherwood@arm.com>
400 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
401 Use plus_constant instead of gen_rtx_PLUS.
403 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
404 Alan Hayward <alan.hayward@arm.com>
405 David Sherwood <david.sherwood@arm.com>
407 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
408 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
409 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
410 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
411 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
412 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
413 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
414 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
415 * config/i386/i386.c (ix86_push_rounding): ...this new function.
416 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
418 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
419 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
420 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
421 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
422 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
423 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
424 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
425 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
426 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
427 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
429 * expr.c (emit_move_resolve_push): Treat the input and result
430 of PUSH_ROUNDING as a poly_int64.
431 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
432 (emit_push_insn): Likewise.
433 * lra-eliminations.c (mark_not_eliminable): Likewise.
434 * recog.c (push_operand): Likewise.
435 * reload1.c (elimination_effects): Likewise.
436 * rtlanal.c (nonzero_bits1): Likewise.
437 * calls.c (store_one_arg): Likewise. Require the padding to be
438 known at compile time.
440 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
441 Alan Hayward <alan.hayward@arm.com>
442 David Sherwood <david.sherwood@arm.com>
444 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
445 Use plus_constant instead of gen_rtx_PLUS.
447 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
448 Alan Hayward <alan.hayward@arm.com>
449 David Sherwood <david.sherwood@arm.com>
451 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
454 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
455 Alan Hayward <alan.hayward@arm.com>
456 David Sherwood <david.sherwood@arm.com>
458 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
459 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
460 via stack temporaries. Treat the mode size as polynomial too.
462 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
463 Alan Hayward <alan.hayward@arm.com>
464 David Sherwood <david.sherwood@arm.com>
466 * expr.c (expand_expr_real_2): When handling conversions involving
467 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
468 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
469 as a poly_uint64 too.
471 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
472 Alan Hayward <alan.hayward@arm.com>
473 David Sherwood <david.sherwood@arm.com>
475 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
477 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
478 Alan Hayward <alan.hayward@arm.com>
479 David Sherwood <david.sherwood@arm.com>
481 * combine.c (can_change_dest_mode): Handle polynomial
482 REGMODE_NATURAL_SIZE.
483 * expmed.c (store_bit_field_1): Likewise.
484 * expr.c (store_constructor): Likewise.
485 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
486 and polynomial REGMODE_NATURAL_SIZE.
487 (gen_lowpart_common): Likewise.
488 * reginfo.c (record_subregs_of_mode): Likewise.
489 * rtlanal.c (read_modify_subreg_p): Likewise.
491 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
492 Alan Hayward <alan.hayward@arm.com>
493 David Sherwood <david.sherwood@arm.com>
495 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
498 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
499 Alan Hayward <alan.hayward@arm.com>
500 David Sherwood <david.sherwood@arm.com>
502 * match.pd: Cope with polynomial numbers of vector elements.
504 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
505 Alan Hayward <alan.hayward@arm.com>
506 David Sherwood <david.sherwood@arm.com>
508 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
509 in a POINTER_PLUS_EXPR.
511 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
512 Alan Hayward <alan.hayward@arm.com>
513 David Sherwood <david.sherwood@arm.com>
515 * omp-simd-clone.c (simd_clone_subparts): New function.
516 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
517 (ipa_simd_modify_function_body): Likewise.
519 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
520 Alan Hayward <alan.hayward@arm.com>
521 David Sherwood <david.sherwood@arm.com>
523 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
524 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
525 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
526 (expand_vector_condition, vector_element): Likewise.
527 (subparts_gt): New function.
528 (get_compute_type): Use subparts_gt.
529 (count_type_subparts): Delete.
530 (expand_vector_operations_1): Use subparts_gt instead of
533 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
534 Alan Hayward <alan.hayward@arm.com>
535 David Sherwood <david.sherwood@arm.com>
537 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
538 (vect_compile_time_alias): ...this new function. Do the calculation
539 on poly_ints rather than trees.
540 (vect_prune_runtime_alias_test_list): Update call accordingly.
542 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
543 Alan Hayward <alan.hayward@arm.com>
544 David Sherwood <david.sherwood@arm.com>
546 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
548 (vect_schedule_slp_instance): Likewise.
550 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
551 Alan Hayward <alan.hayward@arm.com>
552 David Sherwood <david.sherwood@arm.com>
554 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
555 constant and extern definitions for variable-length vectors.
556 (vect_get_constant_vectors): Note that the number of units
557 is known to be constant.
559 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
560 Alan Hayward <alan.hayward@arm.com>
561 David Sherwood <david.sherwood@arm.com>
563 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
564 of units as polynomial. Choose between WIDE and NARROW based
567 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
568 Alan Hayward <alan.hayward@arm.com>
569 David Sherwood <david.sherwood@arm.com>
571 * tree-vect-stmts.c (simd_clone_subparts): New function.
572 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
574 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
575 Alan Hayward <alan.hayward@arm.com>
576 David Sherwood <david.sherwood@arm.com>
578 * tree-vect-stmts.c (vectorizable_call): Treat the number of
579 vectors as polynomial. Use build_index_vector for
582 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
583 Alan Hayward <alan.hayward@arm.com>
584 David Sherwood <david.sherwood@arm.com>
586 * tree-vect-stmts.c (get_load_store_type): Treat the number of
587 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
588 for variable-length vectors.
589 (vectorizable_mask_load_store): Treat the number of units as
590 polynomial, asserting that it is constant if the condition has
591 already been enforced.
592 (vectorizable_store, vectorizable_load): Likewise.
594 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
595 Alan Hayward <alan.hayward@arm.com>
596 David Sherwood <david.sherwood@arm.com>
598 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
599 of units as polynomial. Punt if we can't tell at compile time
600 which vector contains the final result.
602 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
603 Alan Hayward <alan.hayward@arm.com>
604 David Sherwood <david.sherwood@arm.com>
606 * tree-vect-loop.c (vectorizable_induction): Treat the number
607 of units as polynomial. Punt on SLP inductions. Use an integer
608 VEC_SERIES_EXPR for variable-length integer reductions. Use a
609 cast of such a series for variable-length floating-point
612 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
613 Alan Hayward <alan.hayward@arm.com>
614 David Sherwood <david.sherwood@arm.com>
616 * tree.h (build_index_vector): Declare.
617 * tree.c (build_index_vector): New function.
618 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
619 of units as polynomial, forcibly converting it to a constant if
620 vectorizable_reduction has already enforced the condition.
621 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
622 to create a {1,2,3,...} vector.
623 (vectorizable_reduction): Treat the number of units as polynomial.
624 Choose vectype_in based on the largest scalar element size rather
625 than the smallest number of units. Enforce the restrictions
628 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
629 Alan Hayward <alan.hayward@arm.com>
630 David Sherwood <david.sherwood@arm.com>
632 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
633 number of units as polynomial.
635 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
636 Alan Hayward <alan.hayward@arm.com>
637 David Sherwood <david.sherwood@arm.com>
639 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
640 * target.def (autovectorize_vector_sizes): Return the vector sizes
641 by pointer, using vector_sizes rather than a bitmask.
642 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
643 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
644 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
646 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
647 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
648 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
649 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
650 * omp-general.c (omp_max_vf): Likewise.
651 * omp-low.c (omp_clause_aligned_alignment): Likewise.
652 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
653 * tree-vect-loop.c (vect_analyze_loop): Likewise.
654 * tree-vect-slp.c (vect_slp_bb): Likewise.
655 * doc/tm.texi: Regenerate.
656 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
658 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
659 the vector size as a poly_uint64 rather than an unsigned int.
660 (current_vector_size): Change from an unsigned int to a poly_uint64.
661 (get_vectype_for_scalar_type): Update accordingly.
662 * tree.h (build_truth_vector_type): Take the size and number of
663 units as a poly_uint64 rather than an unsigned int.
664 (build_vector_type): Add a temporary overload that takes
665 the number of units as a poly_uint64 rather than an unsigned int.
666 * tree.c (make_vector_type): Likewise.
667 (build_truth_vector_type): Take the number of units as a poly_uint64
668 rather than an unsigned int.
670 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
671 Alan Hayward <alan.hayward@arm.com>
672 David Sherwood <david.sherwood@arm.com>
674 * target.def (get_mask_mode): Take the number of units and length
675 as poly_uint64s rather than unsigned ints.
676 * targhooks.h (default_get_mask_mode): Update accordingly.
677 * targhooks.c (default_get_mask_mode): Likewise.
678 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
679 * doc/tm.texi: Regenerate.
681 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
682 Alan Hayward <alan.hayward@arm.com>
683 David Sherwood <david.sherwood@arm.com>
685 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
686 * omp-general.c (omp_max_vf): Likewise.
687 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
688 (expand_omp_simd): Handle polynomial safelen.
689 * omp-low.c (omplow_simd_context): Add a default constructor.
690 (omplow_simd_context::max_vf): Change from int to poly_uint64.
691 (lower_rec_simd_input_clauses): Update accordingly.
692 (lower_rec_input_clauses): Likewise.
694 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
695 Alan Hayward <alan.hayward@arm.com>
696 David Sherwood <david.sherwood@arm.com>
698 * tree-vectorizer.h (vect_nunits_for_cost): New function.
699 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
700 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
701 (vect_analyze_slp_cost): Likewise.
702 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
703 (vect_model_load_cost): Likewise.
705 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
706 Alan Hayward <alan.hayward@arm.com>
707 David Sherwood <david.sherwood@arm.com>
709 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
710 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
711 from an unsigned int * to a poly_uint64_pod *.
712 (calculate_unrolling_factor): New function.
713 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
715 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
716 Alan Hayward <alan.hayward@arm.com>
717 David Sherwood <david.sherwood@arm.com>
719 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
720 from an unsigned int to a poly_uint64.
721 (_loop_vec_info::slp_unrolling_factor): Likewise.
722 (_loop_vec_info::vectorization_factor): Change from an int
724 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
725 (vect_get_num_vectors): New function.
726 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
727 (vect_get_num_copies): Use vect_get_num_vectors.
728 (vect_analyze_data_ref_dependences): Change max_vf from an int *
729 to an unsigned int *.
730 (vect_analyze_data_refs): Change min_vf from an int * to a
732 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
733 than an unsigned HOST_WIDE_INT.
734 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
735 (vect_analyze_data_ref_dependence): Change max_vf from an int *
736 to an unsigned int *.
737 (vect_analyze_data_ref_dependences): Likewise.
738 (vect_compute_data_ref_alignment): Handle polynomial vf.
739 (vect_enhance_data_refs_alignment): Likewise.
740 (vect_prune_runtime_alias_test_list): Likewise.
741 (vect_shift_permute_load_chain): Likewise.
742 (vect_supportable_dr_alignment): Likewise.
743 (dependence_distance_ge_vf): Take the vectorization factor as a
744 poly_uint64 rather than an unsigned HOST_WIDE_INT.
745 (vect_analyze_data_refs): Change min_vf from an int * to a
747 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
748 vfm1 as a poly_uint64 rather than an int. Make the same change
749 for the returned bound_scalar.
750 (vect_gen_vector_loop_niters): Handle polynomial vf.
751 (vect_do_peeling): Likewise. Update call to
752 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
753 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
755 * tree-vect-loop.c (vect_determine_vectorization_factor)
756 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
757 (vect_get_known_peeling_cost): Likewise.
758 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
759 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
760 (vect_transform_loop): Likewise. Use the lowest possible VF when
761 updating the upper bounds of the loop.
762 (vect_min_worthwhile_factor): Make static. Return an unsigned int
764 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
765 polynomial unroll factors.
766 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
767 (vect_make_slp_decision): Likewise.
768 (vect_supported_load_permutation_p): Likewise, and polynomial
770 (vect_analyze_slp_cost): Handle polynomial vf.
771 (vect_slp_analyze_node_operations): Likewise.
772 (vect_slp_analyze_bb_1): Likewise.
773 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
774 than an unsigned HOST_WIDE_INT.
775 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
776 (vectorizable_load): Handle polynomial vf.
777 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
779 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
781 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
782 Alan Hayward <alan.hayward@arm.com>
783 David Sherwood <david.sherwood@arm.com>
785 * match.pd: Handle bit operations involving three constants
786 and try to fold one pair.
788 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
790 * tree-vect-loop-manip.c: Include gimple-fold.h.
791 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
792 niters_maybe_zero parameters. Handle other cases besides a step of 1.
793 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
794 Add a path that uses a step of VF instead of 1, but disable it
796 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
797 and niters_no_overflow parameters. Update calls to
798 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
799 Create a new SSA name if the latter choses to use a ste other
800 than zero, and return it via niters_vector_mult_vf_var.
801 * tree-vect-loop.c (vect_transform_loop): Update calls to
802 vect_do_peeling, vect_gen_vector_loop_niters and
803 slpeel_make_loop_iterate_ntimes.
804 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
805 (vect_gen_vector_loop_niters): Update declarations after above changes.
807 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
809 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
810 128-bit round to integer instructions.
811 (ceil<mode>2): Likewise.
812 (btrunc<mode>2): Likewise.
813 (round<mode>2): Likewise.
815 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
817 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
818 unaligned VSX load/store on P8/P9.
819 (expand_block_clear): Allow the use of unaligned VSX
822 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
824 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
826 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
827 swap associated with both a load and a store.
829 2018-01-02 Andrew Waterman <andrew@sifive.com>
831 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
832 * config/riscv/riscv.md (clear_cache): Use it.
834 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
836 * web.c: Remove out-of-date comment.
838 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
840 * expr.c (fixup_args_size_notes): Check that any existing
841 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
842 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
843 (emit_single_push_insn): ...here.
845 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
847 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
848 (const_vector_encoded_nelts): New function.
849 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
850 (const_vector_int_elt, const_vector_elt): Declare.
851 * emit-rtl.c (const_vector_int_elt_1): New function.
852 (const_vector_elt): Likewise.
853 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
856 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
858 * expr.c: Include rtx-vector-builder.h.
859 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
860 directly on the tree encoding.
861 (const_vector_from_tree): Likewise.
862 * optabs.c: Include rtx-vector-builder.h.
863 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
864 sequence of "u" values.
865 * vec-perm-indices.c: Include rtx-vector-builder.h.
866 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
867 directly on the vec_perm_indices encoding.
869 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
871 * doc/rtl.texi (const_vector): Describe new encoding scheme.
872 * Makefile.in (OBJS): Add rtx-vector-builder.o.
873 * rtx-vector-builder.h: New file.
874 * rtx-vector-builder.c: Likewise.
875 * rtl.h (rtx_def::u2): Add a const_vector field.
876 (CONST_VECTOR_NPATTERNS): New macro.
877 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
878 (CONST_VECTOR_DUPLICATE_P): Likewise.
879 (CONST_VECTOR_STEPPED_P): Likewise.
880 (CONST_VECTOR_ENCODED_ELT): Likewise.
881 (const_vec_duplicate_p): Check for a duplicated vector encoding.
882 (unwrap_const_vec_duplicate): Likewise.
883 (const_vec_series_p): Check for a non-duplicated vector encoding.
884 Say that the function only returns true for integer vectors.
885 * emit-rtl.c: Include rtx-vector-builder.h.
886 (gen_const_vec_duplicate_1): Delete.
887 (gen_const_vector): Call gen_const_vec_duplicate instead of
888 gen_const_vec_duplicate_1.
889 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
890 (gen_const_vec_duplicate): Use rtx_vector_builder.
891 (gen_const_vec_series): Likewise.
892 (gen_rtx_CONST_VECTOR): Likewise.
893 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
894 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
895 Build a new vector rather than modifying a CONST_VECTOR in-place.
896 (handle_special_swappables): Update call accordingly.
897 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
898 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
899 Build a new vector rather than modifying a CONST_VECTOR in-place.
900 (handle_special_swappables): Update call accordingly.
902 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
904 * simplify-rtx.c (simplify_const_binary_operation): Use
905 CONST_VECTOR_ELT instead of XVECEXP.
907 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
909 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
910 the selector elements to be different from the data elements
911 if the selector is a VECTOR_CST.
912 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
913 ssizetype for the selector.
915 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
917 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
918 before testing each element individually.
919 * tree-vect-generic.c (lower_vec_perm): Likewise.
921 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
923 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
924 * selftest-run-tests.c (selftest::run_tests): Call it.
925 * vector-builder.h (vector_builder::operator ==): New function.
926 (vector_builder::operator !=): Likewise.
927 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
928 (vec_perm_indices::all_from_input_p): New function.
929 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
930 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
931 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
932 instead of reading the VECTOR_CST directly. Detect whether both
933 vector inputs are the same before constructing the vec_perm_indices,
934 and update the number of inputs argument accordingly. Use the
935 utility functions added above. Only construct sel2 if we need to.
937 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
939 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
940 the broadcast of the low byte.
941 (expand_mult_highpart): Use an explicit encoding for the permutes.
942 * optabs-query.c (can_mult_highpart_p): Likewise.
943 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
944 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
945 (vectorizable_bswap): Likewise.
946 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
947 explicit encoding for the power-of-2 permutes.
948 (vect_permute_store_chain): Likewise.
949 (vect_grouped_load_supported): Likewise.
950 (vect_permute_load_chain): Likewise.
952 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
954 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
955 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
956 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
957 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
958 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
959 (vect_gen_perm_mask_any): Likewise.
961 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
963 * int-vector-builder.h: New file.
964 * vec-perm-indices.h: Include int-vector-builder.h.
965 (vec_perm_indices): Redefine as an int_vector_builder.
966 (auto_vec_perm_indices): Delete.
967 (vec_perm_builder): Redefine as a stand-alone class.
968 (vec_perm_indices::vec_perm_indices): New function.
969 (vec_perm_indices::clamp): Likewise.
970 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
971 (vec_perm_indices::new_vector): New function.
972 (vec_perm_indices::new_expanded_vector): Update for new
973 vec_perm_indices class.
974 (vec_perm_indices::rotate_inputs): New function.
975 (vec_perm_indices::all_in_range_p): Operate directly on the
976 encoded form, without computing elided elements.
977 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
978 encoding. Update for new vec_perm_indices class.
979 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
980 the given vec_perm_builder.
981 (expand_vec_perm_var): Update vec_perm_builder constructor.
982 (expand_mult_highpart): Use vec_perm_builder instead of
983 auto_vec_perm_indices.
984 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
985 vec_perm_indices instead of auto_vec_perm_indices. Use a single
986 or double series encoding as appropriate.
987 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
988 vec_perm_indices instead of auto_vec_perm_indices.
989 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
990 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
991 (vect_permute_store_chain): Likewise.
992 (vect_grouped_load_supported): Likewise.
993 (vect_permute_load_chain): Likewise.
994 (vect_shift_permute_load_chain): Likewise.
995 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
996 (vect_transform_slp_perm_load): Likewise.
997 (vect_schedule_slp_instance): Likewise.
998 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
999 (vectorizable_mask_load_store): Likewise.
1000 (vectorizable_bswap): Likewise.
1001 (vectorizable_store): Likewise.
1002 (vectorizable_load): Likewise.
1003 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
1004 vec_perm_indices instead of auto_vec_perm_indices. Use
1005 tree_to_vec_perm_builder to read the vector from a tree.
1006 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
1007 vec_perm_builder instead of a vec_perm_indices.
1008 (have_whole_vector_shift): Use vec_perm_builder and
1009 vec_perm_indices instead of auto_vec_perm_indices. Leave the
1010 truncation to calc_vec_perm_mask_for_shift.
1011 (vect_create_epilog_for_reduction): Likewise.
1012 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
1013 from auto_vec_perm_indices to vec_perm_indices.
1014 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1015 instead of changing individual elements.
1016 (aarch64_vectorize_vec_perm_const): Use new_vector to install
1017 the vector in d.perm.
1018 * config/arm/arm.c (expand_vec_perm_d::perm): Change
1019 from auto_vec_perm_indices to vec_perm_indices.
1020 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1021 instead of changing individual elements.
1022 (arm_vectorize_vec_perm_const): Use new_vector to install
1023 the vector in d.perm.
1024 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
1025 Update vec_perm_builder constructor.
1026 (rs6000_expand_interleave): Likewise.
1027 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
1028 (rs6000_expand_interleave): Likewise.
1030 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1032 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
1033 to qimode could truncate the indices.
1034 * optabs.c (expand_vec_perm_var): Likewise.
1036 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1038 * Makefile.in (OBJS): Add vec-perm-indices.o.
1039 * vec-perm-indices.h: New file.
1040 * vec-perm-indices.c: Likewise.
1041 * target.h (vec_perm_indices): Replace with a forward class
1043 (auto_vec_perm_indices): Move to vec-perm-indices.h.
1044 * optabs.h: Include vec-perm-indices.h.
1045 (expand_vec_perm): Delete.
1046 (selector_fits_mode_p, expand_vec_perm_var): Declare.
1047 (expand_vec_perm_const): Declare.
1048 * target.def (vec_perm_const_ok): Replace with...
1049 (vec_perm_const): ...this new hook.
1050 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
1051 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
1052 * doc/tm.texi: Regenerate.
1053 * optabs.def (vec_perm_const): Delete.
1054 * doc/md.texi (vec_perm_const): Likewise.
1055 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
1056 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
1057 expand_vec_perm for constant permutation vectors. Assert that
1058 the mode of variable permutation vectors is the integer equivalent
1059 of the mode that is being permuted.
1060 * optabs-query.h (selector_fits_mode_p): Declare.
1061 * optabs-query.c: Include vec-perm-indices.h.
1062 (selector_fits_mode_p): New function.
1063 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
1064 is defined, instead of checking whether the vec_perm_const_optab
1065 exists. Use targetm.vectorize.vec_perm_const instead of
1066 targetm.vectorize.vec_perm_const_ok. Check whether the indices
1067 fit in the vector mode before using a variable permute.
1068 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
1069 vec_perm_indices instead of an rtx.
1070 (expand_vec_perm): Replace with...
1071 (expand_vec_perm_const): ...this new function. Take the selector
1072 as a vec_perm_indices rather than an rtx. Also take the mode of
1073 the selector. Update call to shift_amt_for_vec_perm_mask.
1074 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
1075 Use vec_perm_indices::new_expanded_vector to expand the original
1076 selector into bytes. Check whether the indices fit in the vector
1077 mode before using a variable permute.
1078 (expand_vec_perm_var): Make global.
1079 (expand_mult_highpart): Use expand_vec_perm_const.
1080 * fold-const.c: Includes vec-perm-indices.h.
1081 * tree-ssa-forwprop.c: Likewise.
1082 * tree-vect-data-refs.c: Likewise.
1083 * tree-vect-generic.c: Likewise.
1084 * tree-vect-loop.c: Likewise.
1085 * tree-vect-slp.c: Likewise.
1086 * tree-vect-stmts.c: Likewise.
1087 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
1089 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
1090 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
1091 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
1092 (aarch64_vectorize_vec_perm_const): ...this new function.
1093 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1094 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1095 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
1096 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
1097 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1098 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1099 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
1101 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
1102 check for NEON modes.
1103 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
1104 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
1105 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
1106 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
1108 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
1109 the old VEC_PERM_CONST conditions.
1110 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
1111 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
1112 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
1113 (ia64_vectorize_vec_perm_const_ok): Merge into...
1114 (ia64_vectorize_vec_perm_const): ...this new function.
1115 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
1116 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
1117 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
1118 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
1119 * config/mips/mips.c (mips_expand_vec_perm_const)
1120 (mips_vectorize_vec_perm_const_ok): Merge into...
1121 (mips_vectorize_vec_perm_const): ...this new function.
1122 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
1123 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
1124 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
1125 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
1126 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
1127 (rs6000_expand_vec_perm_const): Delete.
1128 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
1130 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1131 (altivec_expand_vec_perm_const_le): Take each operand individually.
1132 Operate on constant selectors rather than rtxes.
1133 (altivec_expand_vec_perm_const): Likewise. Update call to
1134 altivec_expand_vec_perm_const_le.
1135 (rs6000_expand_vec_perm_const): Delete.
1136 (rs6000_vectorize_vec_perm_const_ok): Delete.
1137 (rs6000_vectorize_vec_perm_const): New function.
1138 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1139 an element count and rtx array.
1140 (rs6000_expand_extract_even): Update call accordingly.
1141 (rs6000_expand_interleave): Likewise.
1142 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
1143 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
1144 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
1145 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
1146 (rs6000_expand_vec_perm_const): Delete.
1147 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1148 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1149 (altivec_expand_vec_perm_const_le): Take each operand individually.
1150 Operate on constant selectors rather than rtxes.
1151 (altivec_expand_vec_perm_const): Likewise. Update call to
1152 altivec_expand_vec_perm_const_le.
1153 (rs6000_expand_vec_perm_const): Delete.
1154 (rs6000_vectorize_vec_perm_const_ok): Delete.
1155 (rs6000_vectorize_vec_perm_const): New function. Remove stray
1156 reference to the SPE evmerge intructions.
1157 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1158 an element count and rtx array.
1159 (rs6000_expand_extract_even): Update call accordingly.
1160 (rs6000_expand_interleave): Likewise.
1161 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
1162 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
1164 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1166 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1168 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
1169 vector mode and that that mode matches the mode of the data
1171 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
1172 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
1173 directly using expand_vec_perm_1 when forcing selectors into
1175 (expand_vec_perm_var): New function, split out from expand_vec_perm.
1177 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1179 * optabs-query.h (can_vec_perm_p): Delete.
1180 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
1181 * optabs-query.c (can_vec_perm_p): Split into...
1182 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
1183 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
1184 particular selector is valid.
1185 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1186 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1187 (vect_grouped_load_supported): Likewise.
1188 (vect_shift_permute_load_chain): Likewise.
1189 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1190 (vect_transform_slp_perm_load): Likewise.
1191 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1192 (vectorizable_bswap): Likewise.
1193 (vect_gen_perm_mask_checked): Likewise.
1194 * fold-const.c (fold_ternary_loc): Likewise. Don't take
1195 implementations of variable permutation vectors into account
1196 when deciding which selector to use.
1197 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
1198 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
1199 with a false third argument.
1200 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
1201 to test whether the constant selector is valid and can_vec_perm_var_p
1202 to test whether a variable selector is valid.
1204 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1206 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
1207 * optabs-query.c (can_vec_perm_p): Likewise.
1208 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
1209 instead of vec_perm_indices.
1210 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
1211 (vect_gen_perm_mask_checked): Likewise,
1212 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
1213 (vect_gen_perm_mask_checked): Likewise,
1215 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1217 * optabs-query.h (qimode_for_vec_perm): Declare.
1218 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
1219 (qimode_for_vec_perm): ...this new function.
1220 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
1222 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1224 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
1225 does not have a conditional at the top.
1227 2018-01-02 Richard Biener <rguenther@suse.de>
1229 * ipa-inline.c (big_speedup_p): Fix expression.
1231 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1234 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
1237 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1241 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
1242 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
1243 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
1244 cond_taken_branch_cost 3->4.
1246 2018-01-01 Jakub Jelinek <jakub@redhat.com>
1248 PR tree-optimization/83581
1249 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
1250 TODO_cleanup_cfg if any changes have been made.
1253 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
1254 convert_modes if target mode has the right side, but different mode
1258 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
1259 last argument when extracting from CONCAT. If either from_real or
1260 from_imag is NULL, use expansion through memory. If result is not
1261 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
1262 the parts directly to inner mode, if even that fails, use expansion
1266 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
1267 check for bswap in mode rather than HImode and use that in expand_unop
1270 Copyright (C) 2018 Free Software Foundation, Inc.
1272 Copying and distribution of this file, with or without modification,
1273 are permitted in any medium without royalty provided the copyright
1274 notice and this notice are preserved.