1 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
7 2018-01-10 Richard Biener <rguenther@suse.de>
10 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
11 early out so it also covers the case where we have a non-NULL
14 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
16 PR tree-optimization/83753
17 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
18 for non-strided grouped accesses if the number of elements is 1.
20 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
23 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
24 * i386.h (TARGET_USE_GATHER): Define.
25 * x86-tune.def (X86_TUNE_USE_GATHER): New.
27 2018-01-10 Martin Liska <mliska@suse.cz>
30 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
31 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
33 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
34 CLEANUP_NO_PARTITIONING is not set.
36 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
38 * doc/rtl.texi: Remove documentation of (const ...) wrappers
39 for vectors, as a partial revert of r254296.
40 * rtl.h (const_vec_p): Delete.
41 (const_vec_duplicate_p): Don't test for vector CONSTs.
42 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
43 * expmed.c (make_tree): Likewise.
46 * common.md (E, F): Use CONSTANT_P instead of checking for
48 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
49 checking for CONST_VECTOR.
51 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
54 * predict.c (force_edge_cold): Handle in more sane way edges
57 2018-01-09 Carl Love <cel@us.ibm.com>
59 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
61 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
62 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
63 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
64 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
65 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
66 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
67 * config/rs6000/rs6000-protos.h: Add extern defition for
68 rs6000_generate_float2_double_code.
69 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
71 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
72 (float2_v2df): Add define_expand.
74 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
77 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
78 op_mode in the force_to_mode call.
80 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
82 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
83 instead of checking each element individually.
84 (aarch64_evpc_uzp): Likewise.
85 (aarch64_evpc_zip): Likewise.
86 (aarch64_evpc_ext): Likewise.
87 (aarch64_evpc_rev): Likewise.
88 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
89 instead of checking each element individually. Return true without
91 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
92 whether all selected elements come from the same input, instead of
93 checking each element individually. Remove calls to gen_rtx_REG,
94 start_sequence and end_sequence and instead assert that no rtl is
97 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
99 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
100 order of HIGH and CONST checks.
102 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
104 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
105 if the destination isn't an SSA_NAME.
107 2018-01-09 Richard Biener <rguenther@suse.de>
109 PR tree-optimization/83668
110 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
112 (canonicalize_loop_form): ... here, renamed from ...
113 (canonicalize_loop_closed_ssa_form): ... this and amended to
114 swap successor edges for loop exit blocks to make us use
115 the RPO order we need for initial schedule generation.
117 2018-01-09 Joseph Myers <joseph@codesourcery.com>
119 PR tree-optimization/64811
120 * match.pd: When optimizing comparisons with Inf, avoid
121 introducing or losing exceptions from comparisons with NaN.
123 2018-01-09 Martin Liska <mliska@suse.cz>
126 * asan.c (shadow_mem_size): Add gcc_assert.
128 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
130 Don't save registers in main().
133 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
134 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
135 * config/avr/avr.c (avr_set_current_function): Don't error if
136 naked, OS_task or OS_main are specified at the same time.
137 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
139 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
141 * common/config/avr/avr-common.c (avr_option_optimization_table):
142 Switch on -mmain-is-OS_task for optimizing compilations.
144 2018-01-09 Richard Biener <rguenther@suse.de>
146 PR tree-optimization/83572
147 * graphite.c: Include cfganal.h.
148 (graphite_transform_loops): Connect infinite loops to exit
149 and remove fake edges at the end.
151 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
153 * ipa-inline.c (edge_badness): Revert accidental checkin.
155 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
158 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
159 symbols; not inline clones.
161 2018-01-09 Jakub Jelinek <jakub@redhat.com>
164 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
165 hard registers. Formatting fixes.
167 PR preprocessor/83722
168 * gcc.c (try_generate_repro): Pass
169 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
170 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
173 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
174 Kito Cheng <kito.cheng@gmail.com>
176 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
177 (riscv_leaf_function_p): Delete.
178 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
180 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
182 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
184 (do_ifelse): New function.
185 (do_isel): New function.
186 (do_sub3): New function.
187 (do_add3): New function.
188 (do_load_mask_compare): New function.
189 (do_overlap_load_compare): New function.
190 (expand_compare_loop): New function.
191 (expand_block_compare): Call expand_compare_loop() when appropriate.
192 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
194 (-mblock-compare-inline-loop-limit): New option.
196 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
199 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
200 Reverse order of second and third operands in first alternative.
201 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
202 of first and second elements in UNSPEC_VPERMR vector.
203 (altivec_expand_vec_perm_le): Likewise.
205 2017-01-08 Jeff Law <law@redhat.com>
207 PR rtl-optimizatin/81308
208 * tree-switch-conversion.c (cfg_altered): New file scoped static.
209 (process_switch): If group_case_labels makes a change, then set
211 (pass_convert_switch::execute): If a switch is converted, then
212 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
214 PR rtl-optimization/81308
215 * recog.c (split_all_insns): Conditionally cleanup the CFG after
218 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
220 PR target/83663 - Revert r255946
221 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
222 generation for cases where splatting a value is not useful.
223 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
224 across a vec_duplicate and a paradoxical subreg forming a vector
225 mode to a vec_concat.
227 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
229 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
230 -march=armv8.3-a variants.
231 * config/arm/t-multilib: Likewise.
232 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
234 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
236 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
238 (cceq_ior_compare_complement): Give it a name so I can use it, and
239 change boolean_or_operator predicate to boolean_operator so it can
240 be used to generate a crand.
241 (eqne): New code iterator.
242 (bd/bd_neg): New code_attrs.
243 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
244 a single define_insn.
245 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
246 decrement (bdnzt/bdnzf/bdzt/bdzf).
247 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
248 with the new names of the branch decrement patterns, and added the
249 names of the branch decrement conditional patterns.
251 2018-01-08 Richard Biener <rguenther@suse.de>
253 PR tree-optimization/83563
254 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
257 2018-01-08 Richard Biener <rguenther@suse.de>
260 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
262 2018-01-08 Richard Biener <rguenther@suse.de>
264 PR tree-optimization/83685
265 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
266 references to abnormals.
268 2018-01-08 Richard Biener <rguenther@suse.de>
271 * dwarf2out.c (output_indirect_strings): Handle empty
272 skeleton_debug_str_hash.
273 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
275 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
277 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
278 (emit_store_direct): Likewise.
279 (arc_trampoline_adjust_address): Likewise.
280 (arc_asm_trampoline_template): New function.
281 (arc_initialize_trampoline): Use asm_trampoline_template.
282 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
283 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
284 * config/arc/arc.md (flush_icache): Delete pattern.
286 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
288 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
289 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
292 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
295 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
296 by not USED_FOR_TARGET.
297 (make_pass_resolve_sw_modes): Likewise.
299 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
301 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
304 2018-01-08 Richard Biener <rguenther@suse.de>
307 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
309 2018-01-08 Richard Biener <rguenther@suse.de>
312 * match.pd ((t * 2) / 2) -> t): Add missing :c.
314 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
317 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
318 basic blocks with a small number of successors.
319 (convert_control_dep_chain_into_preds): Improve handling of
321 (dump_predicates): Split apart into...
322 (dump_pred_chain): ...here...
323 (dump_pred_info): ...and here.
324 (can_one_predicate_be_invalidated_p): Add debugging printfs.
325 (can_chain_union_be_invalidated_p): Improve check for invalidation
327 (uninit_uses_cannot_happen): Avoid unnecessary if
328 convert_control_dep_chain_into_preds yielded nothing.
330 2018-01-06 Martin Sebor <msebor@redhat.com>
332 PR tree-optimization/83640
333 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
334 subtracting negative offset from size.
335 (builtin_access::overlap): Adjust offset bounds of the access to fall
336 within the size of the object if possible.
338 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
340 PR rtl-optimization/83699
341 * expmed.c (extract_bit_field_1): Restrict the vector usage of
342 extract_bit_field_as_subreg to cases in which the extracted
343 value is also a vector.
345 * lra-constraints.c (process_alt_operands): Test for the equivalence
346 substitutions when detecting a possible reload cycle.
348 2018-01-06 Jakub Jelinek <jakub@redhat.com>
351 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
352 by default if flag_selective_schedling{,2}. Formatting fixes.
354 PR rtl-optimization/83682
355 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
356 if it has non-VECTOR_MODE element mode.
357 (vec_duplicate_p): Likewise.
360 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
361 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
363 2018-01-05 Jakub Jelinek <jakub@redhat.com>
366 * config/i386/i386-builtin.def
367 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
368 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
369 Require also OPTION_MASK_ISA_AVX512F in addition to
370 OPTION_MASK_ISA_GFNI.
371 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
372 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
373 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
374 to OPTION_MASK_ISA_GFNI.
375 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
376 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
377 OPTION_MASK_ISA_AVX512BW.
378 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
379 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
380 addition to OPTION_MASK_ISA_GFNI.
381 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
382 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
383 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
384 to OPTION_MASK_ISA_GFNI.
385 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
386 a requirement for all ISAs rather than any of them with a few
388 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
390 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
391 bitmasks to be enabled with 3 exceptions, instead of requiring any
392 enabled ISA with lots of exceptions.
393 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
394 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
395 Change avx512bw in isa attribute to avx512f.
396 * config/i386/sgxintrin.h: Add license boilerplate.
397 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
398 to __AVX512F__ and __AVX512VL to __AVX512VL__.
399 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
400 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
402 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
403 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
404 temporarily sse2 rather than sse if not enabled already.
407 * config/i386/sse.md (VI248_VLBW): Rename to ...
408 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
409 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
410 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
411 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
412 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
413 mode iterator instead of VI248_VLBW.
415 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
417 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
418 (record_modified): Skip clobbers; add debug output.
419 (param_change_prob): Use sreal frequencies.
421 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
423 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
424 punt for user-aligned variables.
426 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
428 * tree-chrec.c (chrec_contains_symbols): Return true for
431 2018-01-05 Sudakshina Das <sudi.das@arm.com>
434 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
435 of (x|y) == x for BICS pattern.
437 2018-01-05 Jakub Jelinek <jakub@redhat.com>
439 PR tree-optimization/83605
440 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
441 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
444 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
446 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
447 * config/epiphany/rtems.h: New file.
449 2018-01-04 Jakub Jelinek <jakub@redhat.com>
450 Uros Bizjak <ubizjak@gmail.com>
453 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
454 QIreg_operand instead of register_operand predicate.
455 * config/i386/i386.c (ix86_rop_should_change_byte_p,
456 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
457 comments instead of -fmitigate[-_]rop.
459 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
462 * cgraphunit.c (symbol_table::compile): Switch to text_section
463 before calling assembly_start debug hook.
464 * run-rtl-passes.c (run_rtl_passes): Likewise.
467 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
469 * tree-vrp.c (extract_range_from_binary_expr_1): Check
470 range_int_cst_p rather than !symbolic_range_p before calling
471 extract_range_from_multiplicative_op_1.
473 2017-01-04 Jeff Law <law@redhat.com>
475 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
476 redundant test in assertion.
478 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
480 * doc/rtl.texi: Document machine_mode wrapper classes.
482 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
484 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
487 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
489 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
490 the VEC_PERM_EXPR fold to fail.
492 2018-01-04 Jakub Jelinek <jakub@redhat.com>
495 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
496 to switched_sections.
498 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
501 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
504 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
507 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
508 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
510 2018-01-04 Jakub Jelinek <jakub@redhat.com>
513 * cfgexpand.c (expand_dbeug_expr) <case BIT_FIELD_REF>: Punt if mode
514 is BLKmode and bitpos not zero or mode change is needed.
516 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
519 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
522 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
525 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
526 instead of MULT rtx. Update all corresponding splitters.
528 (*ssub<modesuffix>): Ditto.
530 (*cmp_sadd_di): Update split patterns.
531 (*cmp_sadd_si): Ditto.
532 (*cmp_sadd_sidi): Ditto.
533 (*cmp_ssub_di): Ditto.
534 (*cmp_ssub_si): Ditto.
535 (*cmp_ssub_sidi): Ditto.
536 * config/alpha/predicates.md (const23_operand): New predicate.
537 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
538 Look for ASHIFT, not MULT inner operand.
539 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
541 2018-01-04 Martin Liska <mliska@suse.cz>
543 PR gcov-profile/83669
544 * gcov.c (output_intermediate_file): Add version to intermediate
546 * doc/gcov.texi: Document new field 'version' in intermediate
547 file format. Fix location of '-k' option of gcov command.
549 2018-01-04 Martin Liska <mliska@suse.cz>
552 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
554 2018-01-04 Jakub Jelinek <jakub@redhat.com>
556 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
558 2018-01-03 Martin Sebor <msebor@redhat.com>
560 PR tree-optimization/83655
561 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
562 checking calls with invalid arguments.
564 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
566 * tree-vect-stmts.c (vect_get_store_rhs): New function.
567 (vectorizable_mask_load_store): Delete.
568 (vectorizable_call): Return false for masked loads and stores.
569 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
570 instead of gimple_assign_rhs1.
571 (vectorizable_load): Handle IFN_MASK_LOAD.
572 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
574 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
576 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
578 (vectorizable_mask_load_store): ...here.
579 (vectorizable_load): ...and here.
581 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
583 * tree-vect-stmts.c (vect_build_all_ones_mask)
584 (vect_build_zero_merge_argument): New functions, split out from...
585 (vectorizable_load): ...here.
587 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
589 * tree-vect-stmts.c (vect_check_store_rhs): New function,
591 (vectorizable_mask_load_store): ...here.
592 (vectorizable_store): ...and here.
594 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
596 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
598 (vectorizable_mask_load_store): ...here.
600 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
602 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
603 (vect_model_store_cost): Take a vec_load_store_type instead of a
605 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
606 (vect_model_store_cost): Take a vec_load_store_type instead of a
608 (vectorizable_mask_load_store): Update accordingly.
609 (vectorizable_store): Likewise.
610 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
612 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
614 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
615 IFN_MASK_LOAD calls here rather than...
616 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
618 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
619 Alan Hayward <alan.hayward@arm.com>
620 David Sherwood <david.sherwood@arm.com>
622 * expmed.c (extract_bit_field_1): For vector extracts,
623 fall back to extract_bit_field_as_subreg if vec_extract
626 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
627 Alan Hayward <alan.hayward@arm.com>
628 David Sherwood <david.sherwood@arm.com>
630 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
631 they are variable or constant sized.
632 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
633 slots for constant-sized data.
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 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
640 handling COND_EXPRs with boolean comparisons, try to find a better
641 basis for the mask type than the boolean itself.
643 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
645 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
646 is calculated and how it can be overridden.
647 * genmodes.c (max_bitsize_mode_any_mode): New variable.
648 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
650 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
653 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
654 Alan Hayward <alan.hayward@arm.com>
655 David Sherwood <david.sherwood@arm.com>
657 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
658 Remove the mode argument.
659 (aarch64_simd_valid_immediate): Remove the mode and inverse
661 * config/aarch64/iterators.md (bitsize): New iterator.
662 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
663 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
664 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
665 aarch64_simd_valid_immediate.
666 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
667 (aarch64_reg_or_bic_imm): Likewise.
668 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
669 with an insn_type enum and msl with a modifier_type enum.
670 Replace element_width with a scalar_mode. Change the shift
671 to unsigned int. Add constructors for scalar_float_mode and
672 scalar_int_mode elements.
673 (aarch64_vect_float_const_representable_p): Delete.
674 (aarch64_can_const_movi_rtx_p)
675 (aarch64_simd_scalar_immediate_valid_for_move)
676 (aarch64_simd_make_constant): Update call to
677 aarch64_simd_valid_immediate.
678 (aarch64_advsimd_valid_immediate_hs): New function.
679 (aarch64_advsimd_valid_immediate): Likewise.
680 (aarch64_simd_valid_immediate): Remove mode and inverse
681 arguments. Rewrite to use the above. Use const_vec_duplicate_p
682 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
683 and aarch64_float_const_representable_p on the result.
684 (aarch64_output_simd_mov_immediate): Remove mode argument.
685 Update call to aarch64_simd_valid_immediate and use of
687 (aarch64_output_scalar_simd_mov_immediate): Update call
690 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
691 Alan Hayward <alan.hayward@arm.com>
692 David Sherwood <david.sherwood@arm.com>
694 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
695 (mode_nunits): Likewise CONST_MODE_NUNITS.
696 * machmode.def (ADJUST_NUNITS): Document.
697 * genmodes.c (mode_data::need_nunits_adj): New field.
698 (blank_mode): Update accordingly.
699 (adj_nunits): New variable.
700 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
702 (emit_mode_size_inline): Set need_bytesize_adj for all modes
703 listed in adj_nunits.
704 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
705 listed in adj_nunits. Don't emit case statements for such modes.
706 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
707 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
708 nothing if adj_nunits is nonnull.
709 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
710 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
711 (emit_mode_fbit): Update use of print_maybe_const_decl.
712 (emit_move_size): Likewise. Treat the array as non-const
714 (emit_mode_adjustments): Handle adj_nunits.
716 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
718 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
719 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
720 (VECTOR_MODES): Use it.
721 (make_vector_modes): Take the prefix as an argument.
723 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
724 Alan Hayward <alan.hayward@arm.com>
725 David Sherwood <david.sherwood@arm.com>
727 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
728 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
729 for MODE_VECTOR_BOOL.
730 * machmode.def (VECTOR_BOOL_MODE): Document.
731 * genmodes.c (VECTOR_BOOL_MODE): New macro.
732 (make_vector_bool_mode): New function.
733 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
735 * lto-streamer-in.c (lto_input_mode_table): Likewise.
736 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
738 * stor-layout.c (int_mode_for_mode): Likewise.
739 * tree.c (build_vector_type_for_mode): Likewise.
740 * varasm.c (output_constant_pool_2): Likewise.
741 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
742 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
743 for MODE_VECTOR_BOOL.
744 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
745 of mode class checks.
746 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
747 instead of a list of mode class checks.
748 (expand_vector_scalar_condition): Likewise.
749 (type_for_widest_vector_mode): Handle BImode as an inner mode.
751 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
752 Alan Hayward <alan.hayward@arm.com>
753 David Sherwood <david.sherwood@arm.com>
755 * machmode.h (mode_size): Change from unsigned short to
757 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
758 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
759 or if measurement_type is not polynomial.
760 (fixed_size_mode::includes_p): Check for constant-sized modes.
761 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
762 return a poly_uint16 rather than an unsigned short.
763 (emit_mode_size): Change the type of mode_size from unsigned short
764 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
765 (emit_mode_adjustments): Cope with polynomial vector sizes.
766 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
768 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
770 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
771 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
772 * caller-save.c (setup_save_areas): Likewise.
773 (replace_reg_with_saved_mem): Likewise.
774 * calls.c (emit_library_call_value_1): Likewise.
775 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
776 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
777 (gen_lowpart_for_combine): Likewise.
778 * convert.c (convert_to_integer_1): Likewise.
779 * cse.c (equiv_constant, cse_insn): Likewise.
780 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
781 (cselib_subst_to_values): Likewise.
782 * dce.c (word_dce_process_block): Likewise.
783 * df-problems.c (df_word_lr_mark_ref): Likewise.
784 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
785 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
786 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
787 (rtl_for_decl_location): Likewise.
788 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
789 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
790 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
791 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
792 (expand_expr_real_1): Likewise.
793 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
794 (pad_below): Likewise.
795 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
796 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
797 * ira.c (get_subreg_tracking_sizes): Likewise.
798 * ira-build.c (ira_create_allocno_objects): Likewise.
799 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
800 (ira_sort_regnos_for_alter_reg): Likewise.
801 * ira-costs.c (record_operand_costs): Likewise.
802 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
803 (resolve_simple_move): Likewise.
804 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
805 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
806 (lra_constraints): Likewise.
807 (CONST_POOL_OK_P): Reject variable-sized modes.
808 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
809 (add_pseudo_to_slot, lra_spill): Likewise.
810 * omp-low.c (omp_clause_aligned_alignment): Likewise.
811 * optabs-query.c (get_best_extraction_insn): Likewise.
812 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
813 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
814 (expand_mult_highpart, valid_multiword_target_p): Likewise.
815 * recog.c (offsettable_address_addr_space_p): Likewise.
816 * regcprop.c (maybe_mode_change): Likewise.
817 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
818 * regrename.c (build_def_use): Likewise.
819 * regstat.c (dump_reg_info): Likewise.
820 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
821 (find_reloads, find_reloads_subreg_address): Likewise.
822 * reload1.c (eliminate_regs_1): Likewise.
823 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
824 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
825 (simplify_binary_operation_1, simplify_subreg): Likewise.
826 * targhooks.c (default_function_arg_padding): Likewise.
827 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
828 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
829 (verify_gimple_assign_ternary): Likewise.
830 * tree-inline.c (estimate_move_cost): Likewise.
831 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
832 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
833 (get_address_cost_ainc): Likewise.
834 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
835 (vect_supportable_dr_alignment): Likewise.
836 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
837 (vectorizable_reduction): Likewise.
838 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
839 (vectorizable_operation, vectorizable_load): Likewise.
840 * tree.c (build_same_sized_truth_vector_type): Likewise.
841 * valtrack.c (cleanup_auto_inc_dec): Likewise.
842 * var-tracking.c (emit_note_insn_var_location): Likewise.
843 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
844 (ADDR_VEC_ALIGN): Likewise.
846 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
847 Alan Hayward <alan.hayward@arm.com>
848 David Sherwood <david.sherwood@arm.com>
850 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
852 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
853 or if measurement_type is polynomial.
854 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
855 * combine.c (make_extraction): Likewise.
856 * dse.c (find_shift_sequence): Likewise.
857 * dwarf2out.c (mem_loc_descriptor): Likewise.
858 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
859 (extract_bit_field, extract_low_bits): Likewise.
860 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
861 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
862 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
863 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
864 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
865 * reload.c (find_reloads): Likewise.
866 * reload1.c (alter_reg): Likewise.
867 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
868 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
869 * tree-if-conv.c (predicate_mem_writes): Likewise.
870 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
871 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
872 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
873 * valtrack.c (dead_debug_insert_temp): Likewise.
874 * varasm.c (mergeable_constant_section): Likewise.
875 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
877 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
878 Alan Hayward <alan.hayward@arm.com>
879 David Sherwood <david.sherwood@arm.com>
881 * expr.c (expand_assignment): Cope with polynomial mode sizes
882 when assigning to a CONCAT.
884 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
885 Alan Hayward <alan.hayward@arm.com>
886 David Sherwood <david.sherwood@arm.com>
888 * machmode.h (mode_precision): Change from unsigned short to
890 (mode_to_precision): Return a poly_uint16 rather than an unsigned
892 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
893 or if measurement_type is not polynomial.
894 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
895 in which the mode is already known to be a scalar_int_mode.
896 * genmodes.c (emit_mode_precision): Change the type of mode_precision
897 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
899 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
900 for GET_MODE_PRECISION.
901 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
902 for GET_MODE_PRECISION.
903 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
905 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
906 (expand_field_assignment, make_extraction): Likewise.
907 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
908 (get_last_value): Likewise.
909 * convert.c (convert_to_integer_1): Likewise.
910 * cse.c (cse_insn): Likewise.
911 * expr.c (expand_expr_real_1): Likewise.
912 * lra-constraints.c (simplify_operand_subreg): Likewise.
913 * optabs-query.c (can_atomic_load_p): Likewise.
914 * optabs.c (expand_atomic_load): Likewise.
915 (expand_atomic_store): Likewise.
916 * ree.c (combine_reaching_defs): Likewise.
917 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
918 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
919 * tree.h (type_has_mode_precision_p): Likewise.
920 * ubsan.c (instrument_si_overflow): Likewise.
922 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
923 Alan Hayward <alan.hayward@arm.com>
924 David Sherwood <david.sherwood@arm.com>
926 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
927 polynomial numbers of units.
928 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
929 (valid_vector_subparts_p): New function.
930 (build_vector_type): Remove temporary shim and take the number
931 of units as a poly_uint64 rather than an int.
932 (build_opaque_vector_type): Take the number of units as a
933 poly_uint64 rather than an int.
934 * tree.c (build_vector_from_ctor): Handle polynomial
935 TYPE_VECTOR_SUBPARTS.
936 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
937 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
938 (build_vector_from_val): If the number of units is variable,
939 use build_vec_duplicate_cst for constant operands and
940 VEC_DUPLICATE_EXPR otherwise.
941 (make_vector_type): Remove temporary is_constant ().
942 (build_vector_type, build_opaque_vector_type): Take the number of
943 units as a poly_uint64 rather than an int.
944 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
946 * cfgexpand.c (expand_debug_expr): Likewise.
947 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
948 (store_constructor, expand_expr_real_1): Likewise.
949 (const_scalar_mask_from_tree): Likewise.
950 * fold-const-call.c (fold_const_reduction): Likewise.
951 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
952 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
953 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
954 (fold_relational_const): Likewise.
955 (native_interpret_vector): Likewise. Change the size from an
956 int to an unsigned int.
957 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
958 TYPE_VECTOR_SUBPARTS.
959 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
960 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
961 duplicating a non-constant operand into a variable-length vector.
962 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
963 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
964 * ipa-icf.c (sem_variable::equals): Likewise.
965 * match.pd: Likewise.
966 * omp-simd-clone.c (simd_clone_subparts): Likewise.
967 * print-tree.c (print_node): Likewise.
968 * stor-layout.c (layout_type): Likewise.
969 * targhooks.c (default_builtin_vectorization_cost): Likewise.
970 * tree-cfg.c (verify_gimple_comparison): Likewise.
971 (verify_gimple_assign_binary): Likewise.
972 (verify_gimple_assign_ternary): Likewise.
973 (verify_gimple_assign_single): Likewise.
974 * tree-pretty-print.c (dump_generic_node): Likewise.
975 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
976 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
977 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
978 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
979 (vect_shift_permute_load_chain): Likewise.
980 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
981 (expand_vector_condition, optimize_vector_constructor): Likewise.
982 (lower_vec_perm, get_compute_type): Likewise.
983 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
984 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
985 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
986 (vect_recog_mask_conversion_pattern): Likewise.
987 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
988 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
989 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
990 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
991 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
992 (vectorizable_shift, vectorizable_operation, vectorizable_store)
993 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
994 (supportable_widening_operation): Likewise.
995 (supportable_narrowing_operation): Likewise.
996 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
998 * varasm.c (output_constant): Likewise.
1000 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1001 Alan Hayward <alan.hayward@arm.com>
1002 David Sherwood <david.sherwood@arm.com>
1004 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
1005 so that both the length == 3 and length != 3 cases set up their
1006 own permute vectors. Add comments explaining why we know the
1007 number of elements is constant.
1008 (vect_permute_load_chain): Likewise.
1010 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1011 Alan Hayward <alan.hayward@arm.com>
1012 David Sherwood <david.sherwood@arm.com>
1014 * machmode.h (mode_nunits): Change from unsigned char to
1016 (ONLY_FIXED_SIZE_MODES): New macro.
1017 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
1018 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
1019 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
1021 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
1022 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
1023 or if measurement_type is not polynomial.
1024 * genmodes.c (ZERO_COEFFS): New macro.
1025 (emit_mode_nunits_inline): Make mode_nunits_inline return a
1027 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
1028 Use ZERO_COEFFS when emitting initializers.
1029 * data-streamer.h (bp_pack_poly_value): New function.
1030 (bp_unpack_poly_value): Likewise.
1031 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
1032 for GET_MODE_NUNITS.
1033 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
1034 for GET_MODE_NUNITS.
1035 * tree.c (make_vector_type): Remove temporary shim and make
1036 the real function take the number of units as a poly_uint64
1038 (build_vector_type_for_mode): Handle polynomial nunits.
1039 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
1040 * emit-rtl.c (const_vec_series_p_1): Likewise.
1041 (gen_rtx_CONST_VECTOR): Likewise.
1042 * fold-const.c (test_vec_duplicate_folding): Likewise.
1043 * genrecog.c (validate_pattern): Likewise.
1044 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
1045 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
1046 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
1047 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
1048 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
1049 * rtlanal.c (subreg_get_info): Likewise.
1050 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1051 (vect_grouped_load_supported): Likewise.
1052 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
1053 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
1054 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
1055 (simplify_const_unary_operation, simplify_binary_operation_1)
1056 (simplify_const_binary_operation, simplify_ternary_operation)
1057 (test_vector_ops_duplicate, test_vector_ops): Likewise.
1058 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
1059 instead of CONST_VECTOR_NUNITS.
1060 * varasm.c (output_constant_pool_2): Likewise.
1061 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
1062 explicit-encoded elements in the XVEC for variable-length vectors.
1064 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1066 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
1068 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1069 Alan Hayward <alan.hayward@arm.com>
1070 David Sherwood <david.sherwood@arm.com>
1072 * coretypes.h (fixed_size_mode): Declare.
1073 (fixed_size_mode_pod): New typedef.
1074 * builtins.h (target_builtins::x_apply_args_mode)
1075 (target_builtins::x_apply_result_mode): Change type to
1076 fixed_size_mode_pod.
1077 * builtins.c (apply_args_size, apply_result_size, result_vector)
1078 (expand_builtin_apply_args_1, expand_builtin_apply)
1079 (expand_builtin_return): Update accordingly.
1081 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1083 * cse.c (hash_rtx_cb): Hash only the encoded elements.
1084 * cselib.c (cselib_hash_rtx): Likewise.
1085 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
1086 CONST_VECTOR encoding.
1088 2017-01-03 Jakub Jelinek <jakub@redhat.com>
1089 Jeff Law <law@redhat.com>
1092 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
1093 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
1094 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
1095 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
1098 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
1099 explicitly probe *sp in a noreturn function if there were any callee
1100 register saves or frame pointer is needed.
1102 2018-01-03 Jakub Jelinek <jakub@redhat.com>
1105 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
1106 BLKmode for ternary, binary or unary expressions.
1109 * var-tracking.c (delete_vta_debug_insn): New inline function.
1110 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
1111 insns from get_insns () to NULL instead of each bb separately.
1112 Use delete_vta_debug_insn. No longer static.
1113 (vt_debug_insns_local, variable_tracking_main_1): Adjust
1114 delete_vta_debug_insns callers.
1115 * rtl.h (delete_vta_debug_insns): Declare.
1116 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
1117 instead of variable_tracking_main.
1119 2018-01-03 Martin Sebor <msebor@redhat.com>
1121 PR tree-optimization/83603
1122 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
1123 arguments past the endof the argument list in functions declared
1124 without a prototype.
1125 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
1126 Avoid checking when arguments are null.
1128 2018-01-03 Martin Sebor <msebor@redhat.com>
1131 * doc/extend.texi (attribute const): Fix a typo.
1132 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
1133 issuing -Wsuggest-attribute for void functions.
1135 2018-01-03 Martin Sebor <msebor@redhat.com>
1137 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
1138 offset_int::from instead of wide_int::to_shwi.
1139 (maybe_diag_overlap): Remove assertion.
1140 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
1141 * gimple-ssa-sprintf.c (format_directive): Same.
1142 (parse_directive): Same.
1143 (sprintf_dom_walker::compute_format_length): Same.
1144 (try_substitute_return_value): Same.
1146 2017-01-03 Jeff Law <law@redhat.com>
1149 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
1150 non-constant residual for zero at runtime and avoid probing in
1151 that case. Reorganize code for trailing problem to mirror handling
1154 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
1156 PR tree-optimization/83501
1157 * tree-ssa-strlen.c (get_string_cst): New.
1158 (handle_char_store): Call get_string_cst.
1160 2018-01-03 Martin Liska <mliska@suse.cz>
1162 PR tree-optimization/83593
1163 * tree-ssa-strlen.c: Include tree-cfg.h.
1164 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
1165 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
1166 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
1168 (strlen_dom_walker::before_dom_children): Call
1169 gimple_purge_dead_eh_edges. Dump tranformation with details
1171 (strlen_dom_walker::before_dom_children): Update call by adding
1172 new argument cleanup_eh.
1173 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
1175 2018-01-03 Martin Liska <mliska@suse.cz>
1178 * cif-code.def (VARIADIC_THUNK): New enum value.
1179 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
1182 2018-01-03 Jan Beulich <jbeulich@suse.com>
1184 * sse.md (mov<mode>_internal): Tighten condition for when to use
1185 vmovdqu<ssescalarsize> for TI and OI modes.
1187 2018-01-03 Jakub Jelinek <jakub@redhat.com>
1189 Update copyright years.
1191 2018-01-03 Martin Liska <mliska@suse.cz>
1194 * ipa-visibility.c (function_and_variable_visibility): Skip
1195 functions with noipa attribure.
1197 2018-01-03 Jakub Jelinek <jakub@redhat.com>
1199 * gcc.c (process_command): Update copyright notice dates.
1200 * gcov-dump.c (print_version): Ditto.
1201 * gcov.c (print_version): Ditto.
1202 * gcov-tool.c (print_version): Ditto.
1203 * gengtype.c (create_file): Ditto.
1204 * doc/cpp.texi: Bump @copying's copyright year.
1205 * doc/cppinternals.texi: Ditto.
1206 * doc/gcc.texi: Ditto.
1207 * doc/gccint.texi: Ditto.
1208 * doc/gcov.texi: Ditto.
1209 * doc/install.texi: Ditto.
1210 * doc/invoke.texi: Ditto.
1212 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1214 * vector-builder.h (vector_builder::m_full_nelts): Change from
1215 unsigned int to poly_uint64.
1216 (vector_builder::full_nelts): Update prototype accordingly.
1217 (vector_builder::new_vector): Likewise.
1218 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
1219 (vector_builder::operator ==): Likewise.
1220 (vector_builder::finalize): Likewise.
1221 * int-vector-builder.h (int_vector_builder::int_vector_builder):
1222 Take the number of elements as a poly_uint64 rather than an
1224 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
1225 from unsigned int to poly_uint64.
1226 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
1227 (vec_perm_indices::new_vector): Likewise.
1228 (vec_perm_indices::length): Likewise.
1229 (vec_perm_indices::nelts_per_input): Likewise.
1230 (vec_perm_indices::input_nelts): Likewise.
1231 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
1232 number of elements per input as a poly_uint64 rather than an
1233 unsigned int. Use the original encoding for variable-length
1234 vectors, rather than clamping each individual element.
1235 For the second and subsequent elements in each pattern,
1236 clamp the step and base before clamping their sum.
1237 (vec_perm_indices::series_p): Handle polynomial element counts.
1238 (vec_perm_indices::all_in_range_p): Likewise.
1239 (vec_perm_indices_to_tree): Likewise.
1240 (vec_perm_indices_to_rtx): Likewise.
1241 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
1242 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
1243 (tree_vector_builder::new_binary_operation): Handle polynomial
1244 element counts. Return false if we need to know the number
1245 of elements at compile time.
1246 * fold-const.c (fold_vec_perm): Punt if the number of elements
1247 isn't known at compile time.
1249 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1251 * vec-perm-indices.h (vec_perm_builder): Change element type
1252 from HOST_WIDE_INT to poly_int64.
1253 (vec_perm_indices::element_type): Update accordingly.
1254 (vec_perm_indices::clamp): Handle polynomial element_types.
1255 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
1256 (vec_perm_indices::all_in_range_p): Likewise.
1257 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
1259 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
1260 polynomial vec_perm_indices element types.
1261 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
1262 * fold-const.c (fold_vec_perm): Likewise.
1263 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
1264 * tree-vect-generic.c (lower_vec_perm): Likewise.
1265 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
1266 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
1267 element type to HOST_WIDE_INT.
1269 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1270 Alan Hayward <alan.hayward@arm.com>
1271 David Sherwood <david.sherwood@arm.com>
1273 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
1274 rather than an int. Use plus_constant.
1275 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
1276 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
1278 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1279 Alan Hayward <alan.hayward@arm.com>
1280 David Sherwood <david.sherwood@arm.com>
1282 * calls.c (emit_call_1, expand_call): Change struct_value_size from
1283 a HOST_WIDE_INT to a poly_int64.
1285 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1286 Alan Hayward <alan.hayward@arm.com>
1287 David Sherwood <david.sherwood@arm.com>
1289 * calls.c (load_register_parameters): Cope with polynomial
1290 mode sizes. Require a constant size for BLKmode parameters
1291 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
1292 forces a parameter to be padded at the lsb end in order to
1293 fill a complete number of words, require the parameter size
1294 to be ordered wrt UNITS_PER_WORD.
1296 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1297 Alan Hayward <alan.hayward@arm.com>
1298 David Sherwood <david.sherwood@arm.com>
1300 * reload1.c (spill_stack_slot_width): Change element type
1301 from unsigned int to poly_uint64_pod.
1302 (alter_reg): Treat mode sizes as polynomial.
1304 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1305 Alan Hayward <alan.hayward@arm.com>
1306 David Sherwood <david.sherwood@arm.com>
1308 * reload.c (complex_word_subreg_p): New function.
1309 (reload_inner_reg_of_subreg, push_reload): Use it.
1311 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1312 Alan Hayward <alan.hayward@arm.com>
1313 David Sherwood <david.sherwood@arm.com>
1315 * lra-constraints.c (process_alt_operands): Reject matched
1316 operands whose sizes aren't ordered.
1317 (match_reload): Refer to this check here.
1319 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1320 Alan Hayward <alan.hayward@arm.com>
1321 David Sherwood <david.sherwood@arm.com>
1323 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
1324 that the mode size is in the set {1, 2, 4, 8, 16}.
1326 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1327 Alan Hayward <alan.hayward@arm.com>
1328 David Sherwood <david.sherwood@arm.com>
1330 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
1331 Use plus_constant instead of gen_rtx_PLUS.
1333 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1334 Alan Hayward <alan.hayward@arm.com>
1335 David Sherwood <david.sherwood@arm.com>
1337 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
1338 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
1339 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
1340 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
1341 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
1342 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
1343 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
1344 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
1345 * config/i386/i386.c (ix86_push_rounding): ...this new function.
1346 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
1348 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
1349 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
1350 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
1351 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
1352 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
1353 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
1354 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
1355 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
1356 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
1357 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
1359 * expr.c (emit_move_resolve_push): Treat the input and result
1360 of PUSH_ROUNDING as a poly_int64.
1361 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
1362 (emit_push_insn): Likewise.
1363 * lra-eliminations.c (mark_not_eliminable): Likewise.
1364 * recog.c (push_operand): Likewise.
1365 * reload1.c (elimination_effects): Likewise.
1366 * rtlanal.c (nonzero_bits1): Likewise.
1367 * calls.c (store_one_arg): Likewise. Require the padding to be
1368 known at compile time.
1370 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1371 Alan Hayward <alan.hayward@arm.com>
1372 David Sherwood <david.sherwood@arm.com>
1374 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
1375 Use plus_constant instead of gen_rtx_PLUS.
1377 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1378 Alan Hayward <alan.hayward@arm.com>
1379 David Sherwood <david.sherwood@arm.com>
1381 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
1384 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1385 Alan Hayward <alan.hayward@arm.com>
1386 David Sherwood <david.sherwood@arm.com>
1388 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
1389 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
1390 via stack temporaries. Treat the mode size as polynomial too.
1392 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1393 Alan Hayward <alan.hayward@arm.com>
1394 David Sherwood <david.sherwood@arm.com>
1396 * expr.c (expand_expr_real_2): When handling conversions involving
1397 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
1398 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
1399 as a poly_uint64 too.
1401 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1402 Alan Hayward <alan.hayward@arm.com>
1403 David Sherwood <david.sherwood@arm.com>
1405 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
1407 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1408 Alan Hayward <alan.hayward@arm.com>
1409 David Sherwood <david.sherwood@arm.com>
1411 * combine.c (can_change_dest_mode): Handle polynomial
1412 REGMODE_NATURAL_SIZE.
1413 * expmed.c (store_bit_field_1): Likewise.
1414 * expr.c (store_constructor): Likewise.
1415 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
1416 and polynomial REGMODE_NATURAL_SIZE.
1417 (gen_lowpart_common): Likewise.
1418 * reginfo.c (record_subregs_of_mode): Likewise.
1419 * rtlanal.c (read_modify_subreg_p): Likewise.
1421 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1422 Alan Hayward <alan.hayward@arm.com>
1423 David Sherwood <david.sherwood@arm.com>
1425 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
1426 numbers of elements.
1428 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1429 Alan Hayward <alan.hayward@arm.com>
1430 David Sherwood <david.sherwood@arm.com>
1432 * match.pd: Cope with polynomial numbers of vector elements.
1434 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1435 Alan Hayward <alan.hayward@arm.com>
1436 David Sherwood <david.sherwood@arm.com>
1438 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
1439 in a POINTER_PLUS_EXPR.
1441 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1442 Alan Hayward <alan.hayward@arm.com>
1443 David Sherwood <david.sherwood@arm.com>
1445 * omp-simd-clone.c (simd_clone_subparts): New function.
1446 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
1447 (ipa_simd_modify_function_body): Likewise.
1449 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1450 Alan Hayward <alan.hayward@arm.com>
1451 David Sherwood <david.sherwood@arm.com>
1453 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
1454 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
1455 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
1456 (expand_vector_condition, vector_element): Likewise.
1457 (subparts_gt): New function.
1458 (get_compute_type): Use subparts_gt.
1459 (count_type_subparts): Delete.
1460 (expand_vector_operations_1): Use subparts_gt instead of
1461 count_type_subparts.
1463 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1464 Alan Hayward <alan.hayward@arm.com>
1465 David Sherwood <david.sherwood@arm.com>
1467 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
1468 (vect_compile_time_alias): ...this new function. Do the calculation
1469 on poly_ints rather than trees.
1470 (vect_prune_runtime_alias_test_list): Update call accordingly.
1472 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1473 Alan Hayward <alan.hayward@arm.com>
1474 David Sherwood <david.sherwood@arm.com>
1476 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
1478 (vect_schedule_slp_instance): Likewise.
1480 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1481 Alan Hayward <alan.hayward@arm.com>
1482 David Sherwood <david.sherwood@arm.com>
1484 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
1485 constant and extern definitions for variable-length vectors.
1486 (vect_get_constant_vectors): Note that the number of units
1487 is known to be constant.
1489 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1490 Alan Hayward <alan.hayward@arm.com>
1491 David Sherwood <david.sherwood@arm.com>
1493 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
1494 of units as polynomial. Choose between WIDE and NARROW based
1497 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1498 Alan Hayward <alan.hayward@arm.com>
1499 David Sherwood <david.sherwood@arm.com>
1501 * tree-vect-stmts.c (simd_clone_subparts): New function.
1502 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
1504 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1505 Alan Hayward <alan.hayward@arm.com>
1506 David Sherwood <david.sherwood@arm.com>
1508 * tree-vect-stmts.c (vectorizable_call): Treat the number of
1509 vectors as polynomial. Use build_index_vector for
1512 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1513 Alan Hayward <alan.hayward@arm.com>
1514 David Sherwood <david.sherwood@arm.com>
1516 * tree-vect-stmts.c (get_load_store_type): Treat the number of
1517 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
1518 for variable-length vectors.
1519 (vectorizable_mask_load_store): Treat the number of units as
1520 polynomial, asserting that it is constant if the condition has
1521 already been enforced.
1522 (vectorizable_store, vectorizable_load): Likewise.
1524 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1525 Alan Hayward <alan.hayward@arm.com>
1526 David Sherwood <david.sherwood@arm.com>
1528 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
1529 of units as polynomial. Punt if we can't tell at compile time
1530 which vector contains the final result.
1532 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1533 Alan Hayward <alan.hayward@arm.com>
1534 David Sherwood <david.sherwood@arm.com>
1536 * tree-vect-loop.c (vectorizable_induction): Treat the number
1537 of units as polynomial. Punt on SLP inductions. Use an integer
1538 VEC_SERIES_EXPR for variable-length integer reductions. Use a
1539 cast of such a series for variable-length floating-point
1542 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1543 Alan Hayward <alan.hayward@arm.com>
1544 David Sherwood <david.sherwood@arm.com>
1546 * tree.h (build_index_vector): Declare.
1547 * tree.c (build_index_vector): New function.
1548 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
1549 of units as polynomial, forcibly converting it to a constant if
1550 vectorizable_reduction has already enforced the condition.
1551 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
1552 to create a {1,2,3,...} vector.
1553 (vectorizable_reduction): Treat the number of units as polynomial.
1554 Choose vectype_in based on the largest scalar element size rather
1555 than the smallest number of units. Enforce the restrictions
1558 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1559 Alan Hayward <alan.hayward@arm.com>
1560 David Sherwood <david.sherwood@arm.com>
1562 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
1563 number of units as polynomial.
1565 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1566 Alan Hayward <alan.hayward@arm.com>
1567 David Sherwood <david.sherwood@arm.com>
1569 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
1570 * target.def (autovectorize_vector_sizes): Return the vector sizes
1571 by pointer, using vector_sizes rather than a bitmask.
1572 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
1573 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
1574 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
1576 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
1577 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
1578 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
1579 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
1580 * omp-general.c (omp_max_vf): Likewise.
1581 * omp-low.c (omp_clause_aligned_alignment): Likewise.
1582 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
1583 * tree-vect-loop.c (vect_analyze_loop): Likewise.
1584 * tree-vect-slp.c (vect_slp_bb): Likewise.
1585 * doc/tm.texi: Regenerate.
1586 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
1588 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
1589 the vector size as a poly_uint64 rather than an unsigned int.
1590 (current_vector_size): Change from an unsigned int to a poly_uint64.
1591 (get_vectype_for_scalar_type): Update accordingly.
1592 * tree.h (build_truth_vector_type): Take the size and number of
1593 units as a poly_uint64 rather than an unsigned int.
1594 (build_vector_type): Add a temporary overload that takes
1595 the number of units as a poly_uint64 rather than an unsigned int.
1596 * tree.c (make_vector_type): Likewise.
1597 (build_truth_vector_type): Take the number of units as a poly_uint64
1598 rather than an unsigned int.
1600 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1601 Alan Hayward <alan.hayward@arm.com>
1602 David Sherwood <david.sherwood@arm.com>
1604 * target.def (get_mask_mode): Take the number of units and length
1605 as poly_uint64s rather than unsigned ints.
1606 * targhooks.h (default_get_mask_mode): Update accordingly.
1607 * targhooks.c (default_get_mask_mode): Likewise.
1608 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
1609 * doc/tm.texi: Regenerate.
1611 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1612 Alan Hayward <alan.hayward@arm.com>
1613 David Sherwood <david.sherwood@arm.com>
1615 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
1616 * omp-general.c (omp_max_vf): Likewise.
1617 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
1618 (expand_omp_simd): Handle polynomial safelen.
1619 * omp-low.c (omplow_simd_context): Add a default constructor.
1620 (omplow_simd_context::max_vf): Change from int to poly_uint64.
1621 (lower_rec_simd_input_clauses): Update accordingly.
1622 (lower_rec_input_clauses): Likewise.
1624 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1625 Alan Hayward <alan.hayward@arm.com>
1626 David Sherwood <david.sherwood@arm.com>
1628 * tree-vectorizer.h (vect_nunits_for_cost): New function.
1629 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
1630 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
1631 (vect_analyze_slp_cost): Likewise.
1632 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
1633 (vect_model_load_cost): Likewise.
1635 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1636 Alan Hayward <alan.hayward@arm.com>
1637 David Sherwood <david.sherwood@arm.com>
1639 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
1640 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
1641 from an unsigned int * to a poly_uint64_pod *.
1642 (calculate_unrolling_factor): New function.
1643 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
1645 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1646 Alan Hayward <alan.hayward@arm.com>
1647 David Sherwood <david.sherwood@arm.com>
1649 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
1650 from an unsigned int to a poly_uint64.
1651 (_loop_vec_info::slp_unrolling_factor): Likewise.
1652 (_loop_vec_info::vectorization_factor): Change from an int
1654 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
1655 (vect_get_num_vectors): New function.
1656 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
1657 (vect_get_num_copies): Use vect_get_num_vectors.
1658 (vect_analyze_data_ref_dependences): Change max_vf from an int *
1659 to an unsigned int *.
1660 (vect_analyze_data_refs): Change min_vf from an int * to a
1662 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1663 than an unsigned HOST_WIDE_INT.
1664 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
1665 (vect_analyze_data_ref_dependence): Change max_vf from an int *
1666 to an unsigned int *.
1667 (vect_analyze_data_ref_dependences): Likewise.
1668 (vect_compute_data_ref_alignment): Handle polynomial vf.
1669 (vect_enhance_data_refs_alignment): Likewise.
1670 (vect_prune_runtime_alias_test_list): Likewise.
1671 (vect_shift_permute_load_chain): Likewise.
1672 (vect_supportable_dr_alignment): Likewise.
1673 (dependence_distance_ge_vf): Take the vectorization factor as a
1674 poly_uint64 rather than an unsigned HOST_WIDE_INT.
1675 (vect_analyze_data_refs): Change min_vf from an int * to a
1677 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
1678 vfm1 as a poly_uint64 rather than an int. Make the same change
1679 for the returned bound_scalar.
1680 (vect_gen_vector_loop_niters): Handle polynomial vf.
1681 (vect_do_peeling): Likewise. Update call to
1682 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
1683 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
1685 * tree-vect-loop.c (vect_determine_vectorization_factor)
1686 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
1687 (vect_get_known_peeling_cost): Likewise.
1688 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
1689 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
1690 (vect_transform_loop): Likewise. Use the lowest possible VF when
1691 updating the upper bounds of the loop.
1692 (vect_min_worthwhile_factor): Make static. Return an unsigned int
1694 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
1695 polynomial unroll factors.
1696 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
1697 (vect_make_slp_decision): Likewise.
1698 (vect_supported_load_permutation_p): Likewise, and polynomial
1700 (vect_analyze_slp_cost): Handle polynomial vf.
1701 (vect_slp_analyze_node_operations): Likewise.
1702 (vect_slp_analyze_bb_1): Likewise.
1703 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1704 than an unsigned HOST_WIDE_INT.
1705 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
1706 (vectorizable_load): Handle polynomial vf.
1707 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
1709 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
1711 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1712 Alan Hayward <alan.hayward@arm.com>
1713 David Sherwood <david.sherwood@arm.com>
1715 * match.pd: Handle bit operations involving three constants
1716 and try to fold one pair.
1718 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1720 * tree-vect-loop-manip.c: Include gimple-fold.h.
1721 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
1722 niters_maybe_zero parameters. Handle other cases besides a step of 1.
1723 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
1724 Add a path that uses a step of VF instead of 1, but disable it
1726 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
1727 and niters_no_overflow parameters. Update calls to
1728 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
1729 Create a new SSA name if the latter choses to use a ste other
1730 than zero, and return it via niters_vector_mult_vf_var.
1731 * tree-vect-loop.c (vect_transform_loop): Update calls to
1732 vect_do_peeling, vect_gen_vector_loop_niters and
1733 slpeel_make_loop_iterate_ntimes.
1734 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
1735 (vect_gen_vector_loop_niters): Update declarations after above changes.
1737 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
1739 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
1740 128-bit round to integer instructions.
1741 (ceil<mode>2): Likewise.
1742 (btrunc<mode>2): Likewise.
1743 (round<mode>2): Likewise.
1745 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1747 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
1748 unaligned VSX load/store on P8/P9.
1749 (expand_block_clear): Allow the use of unaligned VSX
1750 load/store on P8/P9.
1752 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1754 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
1756 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
1757 swap associated with both a load and a store.
1759 2018-01-02 Andrew Waterman <andrew@sifive.com>
1761 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
1762 * config/riscv/riscv.md (clear_cache): Use it.
1764 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
1766 * web.c: Remove out-of-date comment.
1768 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1770 * expr.c (fixup_args_size_notes): Check that any existing
1771 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
1772 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
1773 (emit_single_push_insn): ...here.
1775 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1777 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
1778 (const_vector_encoded_nelts): New function.
1779 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
1780 (const_vector_int_elt, const_vector_elt): Declare.
1781 * emit-rtl.c (const_vector_int_elt_1): New function.
1782 (const_vector_elt): Likewise.
1783 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
1784 of CONST_VECTOR_ELT.
1786 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1788 * expr.c: Include rtx-vector-builder.h.
1789 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
1790 directly on the tree encoding.
1791 (const_vector_from_tree): Likewise.
1792 * optabs.c: Include rtx-vector-builder.h.
1793 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
1794 sequence of "u" values.
1795 * vec-perm-indices.c: Include rtx-vector-builder.h.
1796 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
1797 directly on the vec_perm_indices encoding.
1799 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1801 * doc/rtl.texi (const_vector): Describe new encoding scheme.
1802 * Makefile.in (OBJS): Add rtx-vector-builder.o.
1803 * rtx-vector-builder.h: New file.
1804 * rtx-vector-builder.c: Likewise.
1805 * rtl.h (rtx_def::u2): Add a const_vector field.
1806 (CONST_VECTOR_NPATTERNS): New macro.
1807 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
1808 (CONST_VECTOR_DUPLICATE_P): Likewise.
1809 (CONST_VECTOR_STEPPED_P): Likewise.
1810 (CONST_VECTOR_ENCODED_ELT): Likewise.
1811 (const_vec_duplicate_p): Check for a duplicated vector encoding.
1812 (unwrap_const_vec_duplicate): Likewise.
1813 (const_vec_series_p): Check for a non-duplicated vector encoding.
1814 Say that the function only returns true for integer vectors.
1815 * emit-rtl.c: Include rtx-vector-builder.h.
1816 (gen_const_vec_duplicate_1): Delete.
1817 (gen_const_vector): Call gen_const_vec_duplicate instead of
1818 gen_const_vec_duplicate_1.
1819 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
1820 (gen_const_vec_duplicate): Use rtx_vector_builder.
1821 (gen_const_vec_series): Likewise.
1822 (gen_rtx_CONST_VECTOR): Likewise.
1823 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
1824 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1825 Build a new vector rather than modifying a CONST_VECTOR in-place.
1826 (handle_special_swappables): Update call accordingly.
1827 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
1828 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1829 Build a new vector rather than modifying a CONST_VECTOR in-place.
1830 (handle_special_swappables): Update call accordingly.
1832 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1834 * simplify-rtx.c (simplify_const_binary_operation): Use
1835 CONST_VECTOR_ELT instead of XVECEXP.
1837 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1839 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
1840 the selector elements to be different from the data elements
1841 if the selector is a VECTOR_CST.
1842 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
1843 ssizetype for the selector.
1845 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1847 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
1848 before testing each element individually.
1849 * tree-vect-generic.c (lower_vec_perm): Likewise.
1851 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1853 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
1854 * selftest-run-tests.c (selftest::run_tests): Call it.
1855 * vector-builder.h (vector_builder::operator ==): New function.
1856 (vector_builder::operator !=): Likewise.
1857 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
1858 (vec_perm_indices::all_from_input_p): New function.
1859 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
1860 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
1861 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
1862 instead of reading the VECTOR_CST directly. Detect whether both
1863 vector inputs are the same before constructing the vec_perm_indices,
1864 and update the number of inputs argument accordingly. Use the
1865 utility functions added above. Only construct sel2 if we need to.
1867 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1869 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
1870 the broadcast of the low byte.
1871 (expand_mult_highpart): Use an explicit encoding for the permutes.
1872 * optabs-query.c (can_mult_highpart_p): Likewise.
1873 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
1874 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1875 (vectorizable_bswap): Likewise.
1876 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
1877 explicit encoding for the power-of-2 permutes.
1878 (vect_permute_store_chain): Likewise.
1879 (vect_grouped_load_supported): Likewise.
1880 (vect_permute_load_chain): Likewise.
1882 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1884 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
1885 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
1886 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
1887 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
1888 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
1889 (vect_gen_perm_mask_any): Likewise.
1891 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1893 * int-vector-builder.h: New file.
1894 * vec-perm-indices.h: Include int-vector-builder.h.
1895 (vec_perm_indices): Redefine as an int_vector_builder.
1896 (auto_vec_perm_indices): Delete.
1897 (vec_perm_builder): Redefine as a stand-alone class.
1898 (vec_perm_indices::vec_perm_indices): New function.
1899 (vec_perm_indices::clamp): Likewise.
1900 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
1901 (vec_perm_indices::new_vector): New function.
1902 (vec_perm_indices::new_expanded_vector): Update for new
1903 vec_perm_indices class.
1904 (vec_perm_indices::rotate_inputs): New function.
1905 (vec_perm_indices::all_in_range_p): Operate directly on the
1906 encoded form, without computing elided elements.
1907 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
1908 encoding. Update for new vec_perm_indices class.
1909 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
1910 the given vec_perm_builder.
1911 (expand_vec_perm_var): Update vec_perm_builder constructor.
1912 (expand_mult_highpart): Use vec_perm_builder instead of
1913 auto_vec_perm_indices.
1914 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
1915 vec_perm_indices instead of auto_vec_perm_indices. Use a single
1916 or double series encoding as appropriate.
1917 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
1918 vec_perm_indices instead of auto_vec_perm_indices.
1919 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1920 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1921 (vect_permute_store_chain): Likewise.
1922 (vect_grouped_load_supported): Likewise.
1923 (vect_permute_load_chain): Likewise.
1924 (vect_shift_permute_load_chain): Likewise.
1925 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1926 (vect_transform_slp_perm_load): Likewise.
1927 (vect_schedule_slp_instance): Likewise.
1928 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1929 (vectorizable_mask_load_store): Likewise.
1930 (vectorizable_bswap): Likewise.
1931 (vectorizable_store): Likewise.
1932 (vectorizable_load): Likewise.
1933 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
1934 vec_perm_indices instead of auto_vec_perm_indices. Use
1935 tree_to_vec_perm_builder to read the vector from a tree.
1936 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
1937 vec_perm_builder instead of a vec_perm_indices.
1938 (have_whole_vector_shift): Use vec_perm_builder and
1939 vec_perm_indices instead of auto_vec_perm_indices. Leave the
1940 truncation to calc_vec_perm_mask_for_shift.
1941 (vect_create_epilog_for_reduction): Likewise.
1942 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
1943 from auto_vec_perm_indices to vec_perm_indices.
1944 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1945 instead of changing individual elements.
1946 (aarch64_vectorize_vec_perm_const): Use new_vector to install
1947 the vector in d.perm.
1948 * config/arm/arm.c (expand_vec_perm_d::perm): Change
1949 from auto_vec_perm_indices to vec_perm_indices.
1950 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1951 instead of changing individual elements.
1952 (arm_vectorize_vec_perm_const): Use new_vector to install
1953 the vector in d.perm.
1954 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
1955 Update vec_perm_builder constructor.
1956 (rs6000_expand_interleave): Likewise.
1957 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
1958 (rs6000_expand_interleave): Likewise.
1960 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1962 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
1963 to qimode could truncate the indices.
1964 * optabs.c (expand_vec_perm_var): Likewise.
1966 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1968 * Makefile.in (OBJS): Add vec-perm-indices.o.
1969 * vec-perm-indices.h: New file.
1970 * vec-perm-indices.c: Likewise.
1971 * target.h (vec_perm_indices): Replace with a forward class
1973 (auto_vec_perm_indices): Move to vec-perm-indices.h.
1974 * optabs.h: Include vec-perm-indices.h.
1975 (expand_vec_perm): Delete.
1976 (selector_fits_mode_p, expand_vec_perm_var): Declare.
1977 (expand_vec_perm_const): Declare.
1978 * target.def (vec_perm_const_ok): Replace with...
1979 (vec_perm_const): ...this new hook.
1980 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
1981 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
1982 * doc/tm.texi: Regenerate.
1983 * optabs.def (vec_perm_const): Delete.
1984 * doc/md.texi (vec_perm_const): Likewise.
1985 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
1986 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
1987 expand_vec_perm for constant permutation vectors. Assert that
1988 the mode of variable permutation vectors is the integer equivalent
1989 of the mode that is being permuted.
1990 * optabs-query.h (selector_fits_mode_p): Declare.
1991 * optabs-query.c: Include vec-perm-indices.h.
1992 (selector_fits_mode_p): New function.
1993 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
1994 is defined, instead of checking whether the vec_perm_const_optab
1995 exists. Use targetm.vectorize.vec_perm_const instead of
1996 targetm.vectorize.vec_perm_const_ok. Check whether the indices
1997 fit in the vector mode before using a variable permute.
1998 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
1999 vec_perm_indices instead of an rtx.
2000 (expand_vec_perm): Replace with...
2001 (expand_vec_perm_const): ...this new function. Take the selector
2002 as a vec_perm_indices rather than an rtx. Also take the mode of
2003 the selector. Update call to shift_amt_for_vec_perm_mask.
2004 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
2005 Use vec_perm_indices::new_expanded_vector to expand the original
2006 selector into bytes. Check whether the indices fit in the vector
2007 mode before using a variable permute.
2008 (expand_vec_perm_var): Make global.
2009 (expand_mult_highpart): Use expand_vec_perm_const.
2010 * fold-const.c: Includes vec-perm-indices.h.
2011 * tree-ssa-forwprop.c: Likewise.
2012 * tree-vect-data-refs.c: Likewise.
2013 * tree-vect-generic.c: Likewise.
2014 * tree-vect-loop.c: Likewise.
2015 * tree-vect-slp.c: Likewise.
2016 * tree-vect-stmts.c: Likewise.
2017 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
2019 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
2020 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
2021 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
2022 (aarch64_vectorize_vec_perm_const): ...this new function.
2023 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
2024 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2025 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
2026 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
2027 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
2028 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2029 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
2031 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
2032 check for NEON modes.
2033 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
2034 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
2035 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
2036 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
2038 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
2039 the old VEC_PERM_CONST conditions.
2040 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
2041 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
2042 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
2043 (ia64_vectorize_vec_perm_const_ok): Merge into...
2044 (ia64_vectorize_vec_perm_const): ...this new function.
2045 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
2046 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
2047 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
2048 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
2049 * config/mips/mips.c (mips_expand_vec_perm_const)
2050 (mips_vectorize_vec_perm_const_ok): Merge into...
2051 (mips_vectorize_vec_perm_const): ...this new function.
2052 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
2053 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
2054 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
2055 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
2056 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
2057 (rs6000_expand_vec_perm_const): Delete.
2058 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
2060 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2061 (altivec_expand_vec_perm_const_le): Take each operand individually.
2062 Operate on constant selectors rather than rtxes.
2063 (altivec_expand_vec_perm_const): Likewise. Update call to
2064 altivec_expand_vec_perm_const_le.
2065 (rs6000_expand_vec_perm_const): Delete.
2066 (rs6000_vectorize_vec_perm_const_ok): Delete.
2067 (rs6000_vectorize_vec_perm_const): New function.
2068 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
2069 an element count and rtx array.
2070 (rs6000_expand_extract_even): Update call accordingly.
2071 (rs6000_expand_interleave): Likewise.
2072 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
2073 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
2074 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
2075 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
2076 (rs6000_expand_vec_perm_const): Delete.
2077 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
2078 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2079 (altivec_expand_vec_perm_const_le): Take each operand individually.
2080 Operate on constant selectors rather than rtxes.
2081 (altivec_expand_vec_perm_const): Likewise. Update call to
2082 altivec_expand_vec_perm_const_le.
2083 (rs6000_expand_vec_perm_const): Delete.
2084 (rs6000_vectorize_vec_perm_const_ok): Delete.
2085 (rs6000_vectorize_vec_perm_const): New function. Remove stray
2086 reference to the SPE evmerge intructions.
2087 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
2088 an element count and rtx array.
2089 (rs6000_expand_extract_even): Update call accordingly.
2090 (rs6000_expand_interleave): Likewise.
2091 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
2092 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
2094 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2096 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2098 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
2099 vector mode and that that mode matches the mode of the data
2101 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
2102 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
2103 directly using expand_vec_perm_1 when forcing selectors into
2105 (expand_vec_perm_var): New function, split out from expand_vec_perm.
2107 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2109 * optabs-query.h (can_vec_perm_p): Delete.
2110 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
2111 * optabs-query.c (can_vec_perm_p): Split into...
2112 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
2113 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
2114 particular selector is valid.
2115 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
2116 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
2117 (vect_grouped_load_supported): Likewise.
2118 (vect_shift_permute_load_chain): Likewise.
2119 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
2120 (vect_transform_slp_perm_load): Likewise.
2121 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
2122 (vectorizable_bswap): Likewise.
2123 (vect_gen_perm_mask_checked): Likewise.
2124 * fold-const.c (fold_ternary_loc): Likewise. Don't take
2125 implementations of variable permutation vectors into account
2126 when deciding which selector to use.
2127 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
2128 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
2129 with a false third argument.
2130 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
2131 to test whether the constant selector is valid and can_vec_perm_var_p
2132 to test whether a variable selector is valid.
2134 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2136 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
2137 * optabs-query.c (can_vec_perm_p): Likewise.
2138 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
2139 instead of vec_perm_indices.
2140 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
2141 (vect_gen_perm_mask_checked): Likewise,
2142 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
2143 (vect_gen_perm_mask_checked): Likewise,
2145 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2147 * optabs-query.h (qimode_for_vec_perm): Declare.
2148 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
2149 (qimode_for_vec_perm): ...this new function.
2150 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
2152 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2154 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
2155 does not have a conditional at the top.
2157 2018-01-02 Richard Biener <rguenther@suse.de>
2159 * ipa-inline.c (big_speedup_p): Fix expression.
2161 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
2164 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
2167 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
2171 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
2172 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
2173 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
2174 cond_taken_branch_cost 3->4.
2176 2018-01-01 Jakub Jelinek <jakub@redhat.com>
2178 PR tree-optimization/83581
2179 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
2180 TODO_cleanup_cfg if any changes have been made.
2183 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
2184 convert_modes if target mode has the right side, but different mode
2188 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
2189 last argument when extracting from CONCAT. If either from_real or
2190 from_imag is NULL, use expansion through memory. If result is not
2191 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
2192 the parts directly to inner mode, if even that fails, use expansion
2196 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
2197 check for bswap in mode rather than HImode and use that in expand_unop
2200 Copyright (C) 2018 Free Software Foundation, Inc.
2202 Copying and distribution of this file, with or without modification,
2203 are permitted in any medium without royalty provided the copyright
2204 notice and this notice are preserved.