1 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
6 * tree.c (tree_nop_conversion): Return true for location wrapper
8 (maybe_wrap_with_location): New function.
9 (selftest::check_strip_nops): New function.
10 (selftest::test_location_wrappers): New function.
11 (selftest::tree_c_tests): Call it.
12 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
13 (maybe_wrap_with_location): New decl.
14 (EXPR_LOCATION_WRAPPER_P): New macro.
15 (location_wrapper_p): New inline function.
16 (tree_strip_any_location_wrapper): New inline function.
18 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
21 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
22 stack_realign_offset for the largest alignment of stack slot
24 (ix86_find_max_used_stack_alignment): New function.
25 (ix86_finalize_stack_frame_flags): Use it. Set
26 max_used_stack_alignment if we don't realign stack.
27 * config/i386/i386.h (machine_function): Add
28 max_used_stack_alignment.
30 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
32 * config/arm/arm.opt (-mbranch-cost): New option.
33 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
36 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
39 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
40 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
42 2018-01-10 Richard Biener <rguenther@suse.de>
45 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
46 early out so it also covers the case where we have a non-NULL
49 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
51 PR tree-optimization/83753
52 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
53 for non-strided grouped accesses if the number of elements is 1.
55 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
58 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
59 * i386.h (TARGET_USE_GATHER): Define.
60 * x86-tune.def (X86_TUNE_USE_GATHER): New.
62 2018-01-10 Martin Liska <mliska@suse.cz>
65 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
66 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
68 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
69 CLEANUP_NO_PARTITIONING is not set.
71 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
73 * doc/rtl.texi: Remove documentation of (const ...) wrappers
74 for vectors, as a partial revert of r254296.
75 * rtl.h (const_vec_p): Delete.
76 (const_vec_duplicate_p): Don't test for vector CONSTs.
77 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
78 * expmed.c (make_tree): Likewise.
81 * common.md (E, F): Use CONSTANT_P instead of checking for
83 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
84 checking for CONST_VECTOR.
86 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
89 * predict.c (force_edge_cold): Handle in more sane way edges
92 2018-01-09 Carl Love <cel@us.ibm.com>
94 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
96 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
97 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
98 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
99 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
100 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
101 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
102 * config/rs6000/rs6000-protos.h: Add extern defition for
103 rs6000_generate_float2_double_code.
104 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
106 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
107 (float2_v2df): Add define_expand.
109 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
112 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
113 op_mode in the force_to_mode call.
115 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
117 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
118 instead of checking each element individually.
119 (aarch64_evpc_uzp): Likewise.
120 (aarch64_evpc_zip): Likewise.
121 (aarch64_evpc_ext): Likewise.
122 (aarch64_evpc_rev): Likewise.
123 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
124 instead of checking each element individually. Return true without
126 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
127 whether all selected elements come from the same input, instead of
128 checking each element individually. Remove calls to gen_rtx_REG,
129 start_sequence and end_sequence and instead assert that no rtl is
132 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
134 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
135 order of HIGH and CONST checks.
137 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
139 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
140 if the destination isn't an SSA_NAME.
142 2018-01-09 Richard Biener <rguenther@suse.de>
144 PR tree-optimization/83668
145 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
147 (canonicalize_loop_form): ... here, renamed from ...
148 (canonicalize_loop_closed_ssa_form): ... this and amended to
149 swap successor edges for loop exit blocks to make us use
150 the RPO order we need for initial schedule generation.
152 2018-01-09 Joseph Myers <joseph@codesourcery.com>
154 PR tree-optimization/64811
155 * match.pd: When optimizing comparisons with Inf, avoid
156 introducing or losing exceptions from comparisons with NaN.
158 2018-01-09 Martin Liska <mliska@suse.cz>
161 * asan.c (shadow_mem_size): Add gcc_assert.
163 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
165 Don't save registers in main().
168 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
169 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
170 * config/avr/avr.c (avr_set_current_function): Don't error if
171 naked, OS_task or OS_main are specified at the same time.
172 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
174 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
176 * common/config/avr/avr-common.c (avr_option_optimization_table):
177 Switch on -mmain-is-OS_task for optimizing compilations.
179 2018-01-09 Richard Biener <rguenther@suse.de>
181 PR tree-optimization/83572
182 * graphite.c: Include cfganal.h.
183 (graphite_transform_loops): Connect infinite loops to exit
184 and remove fake edges at the end.
186 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
188 * ipa-inline.c (edge_badness): Revert accidental checkin.
190 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
193 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
194 symbols; not inline clones.
196 2018-01-09 Jakub Jelinek <jakub@redhat.com>
199 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
200 hard registers. Formatting fixes.
202 PR preprocessor/83722
203 * gcc.c (try_generate_repro): Pass
204 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
205 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
208 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
209 Kito Cheng <kito.cheng@gmail.com>
211 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
212 (riscv_leaf_function_p): Delete.
213 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
215 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
217 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
219 (do_ifelse): New function.
220 (do_isel): New function.
221 (do_sub3): New function.
222 (do_add3): New function.
223 (do_load_mask_compare): New function.
224 (do_overlap_load_compare): New function.
225 (expand_compare_loop): New function.
226 (expand_block_compare): Call expand_compare_loop() when appropriate.
227 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
229 (-mblock-compare-inline-loop-limit): New option.
231 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
234 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
235 Reverse order of second and third operands in first alternative.
236 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
237 of first and second elements in UNSPEC_VPERMR vector.
238 (altivec_expand_vec_perm_le): Likewise.
240 2017-01-08 Jeff Law <law@redhat.com>
242 PR rtl-optimizatin/81308
243 * tree-switch-conversion.c (cfg_altered): New file scoped static.
244 (process_switch): If group_case_labels makes a change, then set
246 (pass_convert_switch::execute): If a switch is converted, then
247 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
249 PR rtl-optimization/81308
250 * recog.c (split_all_insns): Conditionally cleanup the CFG after
253 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
255 PR target/83663 - Revert r255946
256 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
257 generation for cases where splatting a value is not useful.
258 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
259 across a vec_duplicate and a paradoxical subreg forming a vector
260 mode to a vec_concat.
262 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
264 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
265 -march=armv8.3-a variants.
266 * config/arm/t-multilib: Likewise.
267 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
269 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
271 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
273 (cceq_ior_compare_complement): Give it a name so I can use it, and
274 change boolean_or_operator predicate to boolean_operator so it can
275 be used to generate a crand.
276 (eqne): New code iterator.
277 (bd/bd_neg): New code_attrs.
278 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
279 a single define_insn.
280 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
281 decrement (bdnzt/bdnzf/bdzt/bdzf).
282 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
283 with the new names of the branch decrement patterns, and added the
284 names of the branch decrement conditional patterns.
286 2018-01-08 Richard Biener <rguenther@suse.de>
288 PR tree-optimization/83563
289 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
292 2018-01-08 Richard Biener <rguenther@suse.de>
295 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
297 2018-01-08 Richard Biener <rguenther@suse.de>
299 PR tree-optimization/83685
300 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
301 references to abnormals.
303 2018-01-08 Richard Biener <rguenther@suse.de>
306 * dwarf2out.c (output_indirect_strings): Handle empty
307 skeleton_debug_str_hash.
308 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
310 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
312 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
313 (emit_store_direct): Likewise.
314 (arc_trampoline_adjust_address): Likewise.
315 (arc_asm_trampoline_template): New function.
316 (arc_initialize_trampoline): Use asm_trampoline_template.
317 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
318 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
319 * config/arc/arc.md (flush_icache): Delete pattern.
321 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
323 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
324 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
327 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
330 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
331 by not USED_FOR_TARGET.
332 (make_pass_resolve_sw_modes): Likewise.
334 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
336 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
339 2018-01-08 Richard Biener <rguenther@suse.de>
342 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
344 2018-01-08 Richard Biener <rguenther@suse.de>
347 * match.pd ((t * 2) / 2) -> t): Add missing :c.
349 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
352 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
353 basic blocks with a small number of successors.
354 (convert_control_dep_chain_into_preds): Improve handling of
356 (dump_predicates): Split apart into...
357 (dump_pred_chain): ...here...
358 (dump_pred_info): ...and here.
359 (can_one_predicate_be_invalidated_p): Add debugging printfs.
360 (can_chain_union_be_invalidated_p): Improve check for invalidation
362 (uninit_uses_cannot_happen): Avoid unnecessary if
363 convert_control_dep_chain_into_preds yielded nothing.
365 2018-01-06 Martin Sebor <msebor@redhat.com>
367 PR tree-optimization/83640
368 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
369 subtracting negative offset from size.
370 (builtin_access::overlap): Adjust offset bounds of the access to fall
371 within the size of the object if possible.
373 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
375 PR rtl-optimization/83699
376 * expmed.c (extract_bit_field_1): Restrict the vector usage of
377 extract_bit_field_as_subreg to cases in which the extracted
378 value is also a vector.
380 * lra-constraints.c (process_alt_operands): Test for the equivalence
381 substitutions when detecting a possible reload cycle.
383 2018-01-06 Jakub Jelinek <jakub@redhat.com>
386 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
387 by default if flag_selective_schedling{,2}. Formatting fixes.
389 PR rtl-optimization/83682
390 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
391 if it has non-VECTOR_MODE element mode.
392 (vec_duplicate_p): Likewise.
395 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
396 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
398 2018-01-05 Jakub Jelinek <jakub@redhat.com>
401 * config/i386/i386-builtin.def
402 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
403 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
404 Require also OPTION_MASK_ISA_AVX512F in addition to
405 OPTION_MASK_ISA_GFNI.
406 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
407 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
408 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
409 to OPTION_MASK_ISA_GFNI.
410 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
411 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
412 OPTION_MASK_ISA_AVX512BW.
413 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
414 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
415 addition to OPTION_MASK_ISA_GFNI.
416 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
417 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
418 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
419 to OPTION_MASK_ISA_GFNI.
420 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
421 a requirement for all ISAs rather than any of them with a few
423 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
425 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
426 bitmasks to be enabled with 3 exceptions, instead of requiring any
427 enabled ISA with lots of exceptions.
428 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
429 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
430 Change avx512bw in isa attribute to avx512f.
431 * config/i386/sgxintrin.h: Add license boilerplate.
432 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
433 to __AVX512F__ and __AVX512VL to __AVX512VL__.
434 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
435 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
437 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
438 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
439 temporarily sse2 rather than sse if not enabled already.
442 * config/i386/sse.md (VI248_VLBW): Rename to ...
443 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
444 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
445 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
446 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
447 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
448 mode iterator instead of VI248_VLBW.
450 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
452 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
453 (record_modified): Skip clobbers; add debug output.
454 (param_change_prob): Use sreal frequencies.
456 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
458 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
459 punt for user-aligned variables.
461 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
463 * tree-chrec.c (chrec_contains_symbols): Return true for
466 2018-01-05 Sudakshina Das <sudi.das@arm.com>
469 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
470 of (x|y) == x for BICS pattern.
472 2018-01-05 Jakub Jelinek <jakub@redhat.com>
474 PR tree-optimization/83605
475 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
476 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
479 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
481 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
482 * config/epiphany/rtems.h: New file.
484 2018-01-04 Jakub Jelinek <jakub@redhat.com>
485 Uros Bizjak <ubizjak@gmail.com>
488 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
489 QIreg_operand instead of register_operand predicate.
490 * config/i386/i386.c (ix86_rop_should_change_byte_p,
491 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
492 comments instead of -fmitigate[-_]rop.
494 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
497 * cgraphunit.c (symbol_table::compile): Switch to text_section
498 before calling assembly_start debug hook.
499 * run-rtl-passes.c (run_rtl_passes): Likewise.
502 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
504 * tree-vrp.c (extract_range_from_binary_expr_1): Check
505 range_int_cst_p rather than !symbolic_range_p before calling
506 extract_range_from_multiplicative_op_1.
508 2017-01-04 Jeff Law <law@redhat.com>
510 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
511 redundant test in assertion.
513 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
515 * doc/rtl.texi: Document machine_mode wrapper classes.
517 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
519 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
522 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
524 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
525 the VEC_PERM_EXPR fold to fail.
527 2018-01-04 Jakub Jelinek <jakub@redhat.com>
530 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
531 to switched_sections.
533 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
536 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
539 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
542 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
543 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
545 2018-01-04 Jakub Jelinek <jakub@redhat.com>
548 * cfgexpand.c (expand_dbeug_expr) <case BIT_FIELD_REF>: Punt if mode
549 is BLKmode and bitpos not zero or mode change is needed.
551 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
554 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
557 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
560 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
561 instead of MULT rtx. Update all corresponding splitters.
563 (*ssub<modesuffix>): Ditto.
565 (*cmp_sadd_di): Update split patterns.
566 (*cmp_sadd_si): Ditto.
567 (*cmp_sadd_sidi): Ditto.
568 (*cmp_ssub_di): Ditto.
569 (*cmp_ssub_si): Ditto.
570 (*cmp_ssub_sidi): Ditto.
571 * config/alpha/predicates.md (const23_operand): New predicate.
572 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
573 Look for ASHIFT, not MULT inner operand.
574 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
576 2018-01-04 Martin Liska <mliska@suse.cz>
578 PR gcov-profile/83669
579 * gcov.c (output_intermediate_file): Add version to intermediate
581 * doc/gcov.texi: Document new field 'version' in intermediate
582 file format. Fix location of '-k' option of gcov command.
584 2018-01-04 Martin Liska <mliska@suse.cz>
587 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
589 2018-01-04 Jakub Jelinek <jakub@redhat.com>
591 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
593 2018-01-03 Martin Sebor <msebor@redhat.com>
595 PR tree-optimization/83655
596 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
597 checking calls with invalid arguments.
599 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
601 * tree-vect-stmts.c (vect_get_store_rhs): New function.
602 (vectorizable_mask_load_store): Delete.
603 (vectorizable_call): Return false for masked loads and stores.
604 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
605 instead of gimple_assign_rhs1.
606 (vectorizable_load): Handle IFN_MASK_LOAD.
607 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
609 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
611 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
613 (vectorizable_mask_load_store): ...here.
614 (vectorizable_load): ...and here.
616 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
618 * tree-vect-stmts.c (vect_build_all_ones_mask)
619 (vect_build_zero_merge_argument): New functions, split out from...
620 (vectorizable_load): ...here.
622 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
624 * tree-vect-stmts.c (vect_check_store_rhs): New function,
626 (vectorizable_mask_load_store): ...here.
627 (vectorizable_store): ...and here.
629 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
631 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
633 (vectorizable_mask_load_store): ...here.
635 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
637 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
638 (vect_model_store_cost): Take a vec_load_store_type instead of a
640 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
641 (vect_model_store_cost): Take a vec_load_store_type instead of a
643 (vectorizable_mask_load_store): Update accordingly.
644 (vectorizable_store): Likewise.
645 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
647 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
649 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
650 IFN_MASK_LOAD calls here rather than...
651 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
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 * expmed.c (extract_bit_field_1): For vector extracts,
658 fall back to extract_bit_field_as_subreg if vec_extract
661 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
662 Alan Hayward <alan.hayward@arm.com>
663 David Sherwood <david.sherwood@arm.com>
665 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
666 they are variable or constant sized.
667 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
668 slots for constant-sized data.
670 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
671 Alan Hayward <alan.hayward@arm.com>
672 David Sherwood <david.sherwood@arm.com>
674 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
675 handling COND_EXPRs with boolean comparisons, try to find a better
676 basis for the mask type than the boolean itself.
678 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
680 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
681 is calculated and how it can be overridden.
682 * genmodes.c (max_bitsize_mode_any_mode): New variable.
683 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
685 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
688 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
689 Alan Hayward <alan.hayward@arm.com>
690 David Sherwood <david.sherwood@arm.com>
692 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
693 Remove the mode argument.
694 (aarch64_simd_valid_immediate): Remove the mode and inverse
696 * config/aarch64/iterators.md (bitsize): New iterator.
697 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
698 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
699 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
700 aarch64_simd_valid_immediate.
701 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
702 (aarch64_reg_or_bic_imm): Likewise.
703 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
704 with an insn_type enum and msl with a modifier_type enum.
705 Replace element_width with a scalar_mode. Change the shift
706 to unsigned int. Add constructors for scalar_float_mode and
707 scalar_int_mode elements.
708 (aarch64_vect_float_const_representable_p): Delete.
709 (aarch64_can_const_movi_rtx_p)
710 (aarch64_simd_scalar_immediate_valid_for_move)
711 (aarch64_simd_make_constant): Update call to
712 aarch64_simd_valid_immediate.
713 (aarch64_advsimd_valid_immediate_hs): New function.
714 (aarch64_advsimd_valid_immediate): Likewise.
715 (aarch64_simd_valid_immediate): Remove mode and inverse
716 arguments. Rewrite to use the above. Use const_vec_duplicate_p
717 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
718 and aarch64_float_const_representable_p on the result.
719 (aarch64_output_simd_mov_immediate): Remove mode argument.
720 Update call to aarch64_simd_valid_immediate and use of
722 (aarch64_output_scalar_simd_mov_immediate): Update call
725 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
726 Alan Hayward <alan.hayward@arm.com>
727 David Sherwood <david.sherwood@arm.com>
729 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
730 (mode_nunits): Likewise CONST_MODE_NUNITS.
731 * machmode.def (ADJUST_NUNITS): Document.
732 * genmodes.c (mode_data::need_nunits_adj): New field.
733 (blank_mode): Update accordingly.
734 (adj_nunits): New variable.
735 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
737 (emit_mode_size_inline): Set need_bytesize_adj for all modes
738 listed in adj_nunits.
739 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
740 listed in adj_nunits. Don't emit case statements for such modes.
741 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
742 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
743 nothing if adj_nunits is nonnull.
744 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
745 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
746 (emit_mode_fbit): Update use of print_maybe_const_decl.
747 (emit_move_size): Likewise. Treat the array as non-const
749 (emit_mode_adjustments): Handle adj_nunits.
751 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
753 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
754 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
755 (VECTOR_MODES): Use it.
756 (make_vector_modes): Take the prefix as an argument.
758 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
759 Alan Hayward <alan.hayward@arm.com>
760 David Sherwood <david.sherwood@arm.com>
762 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
763 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
764 for MODE_VECTOR_BOOL.
765 * machmode.def (VECTOR_BOOL_MODE): Document.
766 * genmodes.c (VECTOR_BOOL_MODE): New macro.
767 (make_vector_bool_mode): New function.
768 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
770 * lto-streamer-in.c (lto_input_mode_table): Likewise.
771 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
773 * stor-layout.c (int_mode_for_mode): Likewise.
774 * tree.c (build_vector_type_for_mode): Likewise.
775 * varasm.c (output_constant_pool_2): Likewise.
776 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
777 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
778 for MODE_VECTOR_BOOL.
779 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
780 of mode class checks.
781 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
782 instead of a list of mode class checks.
783 (expand_vector_scalar_condition): Likewise.
784 (type_for_widest_vector_mode): Handle BImode as an inner mode.
786 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
787 Alan Hayward <alan.hayward@arm.com>
788 David Sherwood <david.sherwood@arm.com>
790 * machmode.h (mode_size): Change from unsigned short to
792 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
793 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
794 or if measurement_type is not polynomial.
795 (fixed_size_mode::includes_p): Check for constant-sized modes.
796 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
797 return a poly_uint16 rather than an unsigned short.
798 (emit_mode_size): Change the type of mode_size from unsigned short
799 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
800 (emit_mode_adjustments): Cope with polynomial vector sizes.
801 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
803 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
805 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
806 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
807 * caller-save.c (setup_save_areas): Likewise.
808 (replace_reg_with_saved_mem): Likewise.
809 * calls.c (emit_library_call_value_1): Likewise.
810 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
811 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
812 (gen_lowpart_for_combine): Likewise.
813 * convert.c (convert_to_integer_1): Likewise.
814 * cse.c (equiv_constant, cse_insn): Likewise.
815 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
816 (cselib_subst_to_values): Likewise.
817 * dce.c (word_dce_process_block): Likewise.
818 * df-problems.c (df_word_lr_mark_ref): Likewise.
819 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
820 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
821 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
822 (rtl_for_decl_location): Likewise.
823 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
824 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
825 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
826 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
827 (expand_expr_real_1): Likewise.
828 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
829 (pad_below): Likewise.
830 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
831 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
832 * ira.c (get_subreg_tracking_sizes): Likewise.
833 * ira-build.c (ira_create_allocno_objects): Likewise.
834 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
835 (ira_sort_regnos_for_alter_reg): Likewise.
836 * ira-costs.c (record_operand_costs): Likewise.
837 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
838 (resolve_simple_move): Likewise.
839 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
840 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
841 (lra_constraints): Likewise.
842 (CONST_POOL_OK_P): Reject variable-sized modes.
843 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
844 (add_pseudo_to_slot, lra_spill): Likewise.
845 * omp-low.c (omp_clause_aligned_alignment): Likewise.
846 * optabs-query.c (get_best_extraction_insn): Likewise.
847 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
848 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
849 (expand_mult_highpart, valid_multiword_target_p): Likewise.
850 * recog.c (offsettable_address_addr_space_p): Likewise.
851 * regcprop.c (maybe_mode_change): Likewise.
852 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
853 * regrename.c (build_def_use): Likewise.
854 * regstat.c (dump_reg_info): Likewise.
855 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
856 (find_reloads, find_reloads_subreg_address): Likewise.
857 * reload1.c (eliminate_regs_1): Likewise.
858 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
859 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
860 (simplify_binary_operation_1, simplify_subreg): Likewise.
861 * targhooks.c (default_function_arg_padding): Likewise.
862 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
863 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
864 (verify_gimple_assign_ternary): Likewise.
865 * tree-inline.c (estimate_move_cost): Likewise.
866 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
867 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
868 (get_address_cost_ainc): Likewise.
869 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
870 (vect_supportable_dr_alignment): Likewise.
871 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
872 (vectorizable_reduction): Likewise.
873 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
874 (vectorizable_operation, vectorizable_load): Likewise.
875 * tree.c (build_same_sized_truth_vector_type): Likewise.
876 * valtrack.c (cleanup_auto_inc_dec): Likewise.
877 * var-tracking.c (emit_note_insn_var_location): Likewise.
878 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
879 (ADDR_VEC_ALIGN): Likewise.
881 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
882 Alan Hayward <alan.hayward@arm.com>
883 David Sherwood <david.sherwood@arm.com>
885 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
887 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
888 or if measurement_type is polynomial.
889 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
890 * combine.c (make_extraction): Likewise.
891 * dse.c (find_shift_sequence): Likewise.
892 * dwarf2out.c (mem_loc_descriptor): Likewise.
893 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
894 (extract_bit_field, extract_low_bits): Likewise.
895 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
896 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
897 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
898 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
899 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
900 * reload.c (find_reloads): Likewise.
901 * reload1.c (alter_reg): Likewise.
902 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
903 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
904 * tree-if-conv.c (predicate_mem_writes): Likewise.
905 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
906 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
907 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
908 * valtrack.c (dead_debug_insert_temp): Likewise.
909 * varasm.c (mergeable_constant_section): Likewise.
910 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
912 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
913 Alan Hayward <alan.hayward@arm.com>
914 David Sherwood <david.sherwood@arm.com>
916 * expr.c (expand_assignment): Cope with polynomial mode sizes
917 when assigning to a CONCAT.
919 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
920 Alan Hayward <alan.hayward@arm.com>
921 David Sherwood <david.sherwood@arm.com>
923 * machmode.h (mode_precision): Change from unsigned short to
925 (mode_to_precision): Return a poly_uint16 rather than an unsigned
927 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
928 or if measurement_type is not polynomial.
929 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
930 in which the mode is already known to be a scalar_int_mode.
931 * genmodes.c (emit_mode_precision): Change the type of mode_precision
932 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
934 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
935 for GET_MODE_PRECISION.
936 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
937 for GET_MODE_PRECISION.
938 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
940 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
941 (expand_field_assignment, make_extraction): Likewise.
942 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
943 (get_last_value): Likewise.
944 * convert.c (convert_to_integer_1): Likewise.
945 * cse.c (cse_insn): Likewise.
946 * expr.c (expand_expr_real_1): Likewise.
947 * lra-constraints.c (simplify_operand_subreg): Likewise.
948 * optabs-query.c (can_atomic_load_p): Likewise.
949 * optabs.c (expand_atomic_load): Likewise.
950 (expand_atomic_store): Likewise.
951 * ree.c (combine_reaching_defs): Likewise.
952 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
953 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
954 * tree.h (type_has_mode_precision_p): Likewise.
955 * ubsan.c (instrument_si_overflow): Likewise.
957 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
958 Alan Hayward <alan.hayward@arm.com>
959 David Sherwood <david.sherwood@arm.com>
961 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
962 polynomial numbers of units.
963 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
964 (valid_vector_subparts_p): New function.
965 (build_vector_type): Remove temporary shim and take the number
966 of units as a poly_uint64 rather than an int.
967 (build_opaque_vector_type): Take the number of units as a
968 poly_uint64 rather than an int.
969 * tree.c (build_vector_from_ctor): Handle polynomial
970 TYPE_VECTOR_SUBPARTS.
971 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
972 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
973 (build_vector_from_val): If the number of units is variable,
974 use build_vec_duplicate_cst for constant operands and
975 VEC_DUPLICATE_EXPR otherwise.
976 (make_vector_type): Remove temporary is_constant ().
977 (build_vector_type, build_opaque_vector_type): Take the number of
978 units as a poly_uint64 rather than an int.
979 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
981 * cfgexpand.c (expand_debug_expr): Likewise.
982 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
983 (store_constructor, expand_expr_real_1): Likewise.
984 (const_scalar_mask_from_tree): Likewise.
985 * fold-const-call.c (fold_const_reduction): Likewise.
986 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
987 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
988 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
989 (fold_relational_const): Likewise.
990 (native_interpret_vector): Likewise. Change the size from an
991 int to an unsigned int.
992 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
993 TYPE_VECTOR_SUBPARTS.
994 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
995 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
996 duplicating a non-constant operand into a variable-length vector.
997 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
998 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
999 * ipa-icf.c (sem_variable::equals): Likewise.
1000 * match.pd: Likewise.
1001 * omp-simd-clone.c (simd_clone_subparts): Likewise.
1002 * print-tree.c (print_node): Likewise.
1003 * stor-layout.c (layout_type): Likewise.
1004 * targhooks.c (default_builtin_vectorization_cost): Likewise.
1005 * tree-cfg.c (verify_gimple_comparison): Likewise.
1006 (verify_gimple_assign_binary): Likewise.
1007 (verify_gimple_assign_ternary): Likewise.
1008 (verify_gimple_assign_single): Likewise.
1009 * tree-pretty-print.c (dump_generic_node): Likewise.
1010 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1011 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
1012 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
1013 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
1014 (vect_shift_permute_load_chain): Likewise.
1015 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
1016 (expand_vector_condition, optimize_vector_constructor): Likewise.
1017 (lower_vec_perm, get_compute_type): Likewise.
1018 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
1019 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
1020 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
1021 (vect_recog_mask_conversion_pattern): Likewise.
1022 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
1023 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
1024 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1025 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
1026 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
1027 (vectorizable_shift, vectorizable_operation, vectorizable_store)
1028 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
1029 (supportable_widening_operation): Likewise.
1030 (supportable_narrowing_operation): Likewise.
1031 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
1033 * varasm.c (output_constant): Likewise.
1035 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1036 Alan Hayward <alan.hayward@arm.com>
1037 David Sherwood <david.sherwood@arm.com>
1039 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
1040 so that both the length == 3 and length != 3 cases set up their
1041 own permute vectors. Add comments explaining why we know the
1042 number of elements is constant.
1043 (vect_permute_load_chain): Likewise.
1045 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1046 Alan Hayward <alan.hayward@arm.com>
1047 David Sherwood <david.sherwood@arm.com>
1049 * machmode.h (mode_nunits): Change from unsigned char to
1051 (ONLY_FIXED_SIZE_MODES): New macro.
1052 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
1053 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
1054 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
1056 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
1057 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
1058 or if measurement_type is not polynomial.
1059 * genmodes.c (ZERO_COEFFS): New macro.
1060 (emit_mode_nunits_inline): Make mode_nunits_inline return a
1062 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
1063 Use ZERO_COEFFS when emitting initializers.
1064 * data-streamer.h (bp_pack_poly_value): New function.
1065 (bp_unpack_poly_value): Likewise.
1066 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
1067 for GET_MODE_NUNITS.
1068 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
1069 for GET_MODE_NUNITS.
1070 * tree.c (make_vector_type): Remove temporary shim and make
1071 the real function take the number of units as a poly_uint64
1073 (build_vector_type_for_mode): Handle polynomial nunits.
1074 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
1075 * emit-rtl.c (const_vec_series_p_1): Likewise.
1076 (gen_rtx_CONST_VECTOR): Likewise.
1077 * fold-const.c (test_vec_duplicate_folding): Likewise.
1078 * genrecog.c (validate_pattern): Likewise.
1079 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
1080 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
1081 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
1082 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
1083 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
1084 * rtlanal.c (subreg_get_info): Likewise.
1085 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1086 (vect_grouped_load_supported): Likewise.
1087 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
1088 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
1089 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
1090 (simplify_const_unary_operation, simplify_binary_operation_1)
1091 (simplify_const_binary_operation, simplify_ternary_operation)
1092 (test_vector_ops_duplicate, test_vector_ops): Likewise.
1093 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
1094 instead of CONST_VECTOR_NUNITS.
1095 * varasm.c (output_constant_pool_2): Likewise.
1096 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
1097 explicit-encoded elements in the XVEC for variable-length vectors.
1099 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1101 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
1103 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1104 Alan Hayward <alan.hayward@arm.com>
1105 David Sherwood <david.sherwood@arm.com>
1107 * coretypes.h (fixed_size_mode): Declare.
1108 (fixed_size_mode_pod): New typedef.
1109 * builtins.h (target_builtins::x_apply_args_mode)
1110 (target_builtins::x_apply_result_mode): Change type to
1111 fixed_size_mode_pod.
1112 * builtins.c (apply_args_size, apply_result_size, result_vector)
1113 (expand_builtin_apply_args_1, expand_builtin_apply)
1114 (expand_builtin_return): Update accordingly.
1116 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1118 * cse.c (hash_rtx_cb): Hash only the encoded elements.
1119 * cselib.c (cselib_hash_rtx): Likewise.
1120 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
1121 CONST_VECTOR encoding.
1123 2017-01-03 Jakub Jelinek <jakub@redhat.com>
1124 Jeff Law <law@redhat.com>
1127 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
1128 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
1129 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
1130 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
1133 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
1134 explicitly probe *sp in a noreturn function if there were any callee
1135 register saves or frame pointer is needed.
1137 2018-01-03 Jakub Jelinek <jakub@redhat.com>
1140 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
1141 BLKmode for ternary, binary or unary expressions.
1144 * var-tracking.c (delete_vta_debug_insn): New inline function.
1145 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
1146 insns from get_insns () to NULL instead of each bb separately.
1147 Use delete_vta_debug_insn. No longer static.
1148 (vt_debug_insns_local, variable_tracking_main_1): Adjust
1149 delete_vta_debug_insns callers.
1150 * rtl.h (delete_vta_debug_insns): Declare.
1151 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
1152 instead of variable_tracking_main.
1154 2018-01-03 Martin Sebor <msebor@redhat.com>
1156 PR tree-optimization/83603
1157 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
1158 arguments past the endof the argument list in functions declared
1159 without a prototype.
1160 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
1161 Avoid checking when arguments are null.
1163 2018-01-03 Martin Sebor <msebor@redhat.com>
1166 * doc/extend.texi (attribute const): Fix a typo.
1167 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
1168 issuing -Wsuggest-attribute for void functions.
1170 2018-01-03 Martin Sebor <msebor@redhat.com>
1172 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
1173 offset_int::from instead of wide_int::to_shwi.
1174 (maybe_diag_overlap): Remove assertion.
1175 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
1176 * gimple-ssa-sprintf.c (format_directive): Same.
1177 (parse_directive): Same.
1178 (sprintf_dom_walker::compute_format_length): Same.
1179 (try_substitute_return_value): Same.
1181 2017-01-03 Jeff Law <law@redhat.com>
1184 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
1185 non-constant residual for zero at runtime and avoid probing in
1186 that case. Reorganize code for trailing problem to mirror handling
1189 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
1191 PR tree-optimization/83501
1192 * tree-ssa-strlen.c (get_string_cst): New.
1193 (handle_char_store): Call get_string_cst.
1195 2018-01-03 Martin Liska <mliska@suse.cz>
1197 PR tree-optimization/83593
1198 * tree-ssa-strlen.c: Include tree-cfg.h.
1199 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
1200 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
1201 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
1203 (strlen_dom_walker::before_dom_children): Call
1204 gimple_purge_dead_eh_edges. Dump tranformation with details
1206 (strlen_dom_walker::before_dom_children): Update call by adding
1207 new argument cleanup_eh.
1208 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
1210 2018-01-03 Martin Liska <mliska@suse.cz>
1213 * cif-code.def (VARIADIC_THUNK): New enum value.
1214 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
1217 2018-01-03 Jan Beulich <jbeulich@suse.com>
1219 * sse.md (mov<mode>_internal): Tighten condition for when to use
1220 vmovdqu<ssescalarsize> for TI and OI modes.
1222 2018-01-03 Jakub Jelinek <jakub@redhat.com>
1224 Update copyright years.
1226 2018-01-03 Martin Liska <mliska@suse.cz>
1229 * ipa-visibility.c (function_and_variable_visibility): Skip
1230 functions with noipa attribure.
1232 2018-01-03 Jakub Jelinek <jakub@redhat.com>
1234 * gcc.c (process_command): Update copyright notice dates.
1235 * gcov-dump.c (print_version): Ditto.
1236 * gcov.c (print_version): Ditto.
1237 * gcov-tool.c (print_version): Ditto.
1238 * gengtype.c (create_file): Ditto.
1239 * doc/cpp.texi: Bump @copying's copyright year.
1240 * doc/cppinternals.texi: Ditto.
1241 * doc/gcc.texi: Ditto.
1242 * doc/gccint.texi: Ditto.
1243 * doc/gcov.texi: Ditto.
1244 * doc/install.texi: Ditto.
1245 * doc/invoke.texi: Ditto.
1247 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1249 * vector-builder.h (vector_builder::m_full_nelts): Change from
1250 unsigned int to poly_uint64.
1251 (vector_builder::full_nelts): Update prototype accordingly.
1252 (vector_builder::new_vector): Likewise.
1253 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
1254 (vector_builder::operator ==): Likewise.
1255 (vector_builder::finalize): Likewise.
1256 * int-vector-builder.h (int_vector_builder::int_vector_builder):
1257 Take the number of elements as a poly_uint64 rather than an
1259 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
1260 from unsigned int to poly_uint64.
1261 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
1262 (vec_perm_indices::new_vector): Likewise.
1263 (vec_perm_indices::length): Likewise.
1264 (vec_perm_indices::nelts_per_input): Likewise.
1265 (vec_perm_indices::input_nelts): Likewise.
1266 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
1267 number of elements per input as a poly_uint64 rather than an
1268 unsigned int. Use the original encoding for variable-length
1269 vectors, rather than clamping each individual element.
1270 For the second and subsequent elements in each pattern,
1271 clamp the step and base before clamping their sum.
1272 (vec_perm_indices::series_p): Handle polynomial element counts.
1273 (vec_perm_indices::all_in_range_p): Likewise.
1274 (vec_perm_indices_to_tree): Likewise.
1275 (vec_perm_indices_to_rtx): Likewise.
1276 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
1277 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
1278 (tree_vector_builder::new_binary_operation): Handle polynomial
1279 element counts. Return false if we need to know the number
1280 of elements at compile time.
1281 * fold-const.c (fold_vec_perm): Punt if the number of elements
1282 isn't known at compile time.
1284 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1286 * vec-perm-indices.h (vec_perm_builder): Change element type
1287 from HOST_WIDE_INT to poly_int64.
1288 (vec_perm_indices::element_type): Update accordingly.
1289 (vec_perm_indices::clamp): Handle polynomial element_types.
1290 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
1291 (vec_perm_indices::all_in_range_p): Likewise.
1292 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
1294 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
1295 polynomial vec_perm_indices element types.
1296 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
1297 * fold-const.c (fold_vec_perm): Likewise.
1298 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
1299 * tree-vect-generic.c (lower_vec_perm): Likewise.
1300 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
1301 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
1302 element type to HOST_WIDE_INT.
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 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
1309 rather than an int. Use plus_constant.
1310 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
1311 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
1313 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1314 Alan Hayward <alan.hayward@arm.com>
1315 David Sherwood <david.sherwood@arm.com>
1317 * calls.c (emit_call_1, expand_call): Change struct_value_size from
1318 a HOST_WIDE_INT to a poly_int64.
1320 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1321 Alan Hayward <alan.hayward@arm.com>
1322 David Sherwood <david.sherwood@arm.com>
1324 * calls.c (load_register_parameters): Cope with polynomial
1325 mode sizes. Require a constant size for BLKmode parameters
1326 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
1327 forces a parameter to be padded at the lsb end in order to
1328 fill a complete number of words, require the parameter size
1329 to be ordered wrt UNITS_PER_WORD.
1331 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1332 Alan Hayward <alan.hayward@arm.com>
1333 David Sherwood <david.sherwood@arm.com>
1335 * reload1.c (spill_stack_slot_width): Change element type
1336 from unsigned int to poly_uint64_pod.
1337 (alter_reg): Treat mode sizes as polynomial.
1339 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1340 Alan Hayward <alan.hayward@arm.com>
1341 David Sherwood <david.sherwood@arm.com>
1343 * reload.c (complex_word_subreg_p): New function.
1344 (reload_inner_reg_of_subreg, push_reload): Use it.
1346 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1347 Alan Hayward <alan.hayward@arm.com>
1348 David Sherwood <david.sherwood@arm.com>
1350 * lra-constraints.c (process_alt_operands): Reject matched
1351 operands whose sizes aren't ordered.
1352 (match_reload): Refer to this check here.
1354 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1355 Alan Hayward <alan.hayward@arm.com>
1356 David Sherwood <david.sherwood@arm.com>
1358 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
1359 that the mode size is in the set {1, 2, 4, 8, 16}.
1361 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1362 Alan Hayward <alan.hayward@arm.com>
1363 David Sherwood <david.sherwood@arm.com>
1365 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
1366 Use plus_constant instead of gen_rtx_PLUS.
1368 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1369 Alan Hayward <alan.hayward@arm.com>
1370 David Sherwood <david.sherwood@arm.com>
1372 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
1373 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
1374 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
1375 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
1376 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
1377 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
1378 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
1379 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
1380 * config/i386/i386.c (ix86_push_rounding): ...this new function.
1381 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
1383 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
1384 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
1385 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
1386 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
1387 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
1388 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
1389 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
1390 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
1391 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
1392 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
1394 * expr.c (emit_move_resolve_push): Treat the input and result
1395 of PUSH_ROUNDING as a poly_int64.
1396 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
1397 (emit_push_insn): Likewise.
1398 * lra-eliminations.c (mark_not_eliminable): Likewise.
1399 * recog.c (push_operand): Likewise.
1400 * reload1.c (elimination_effects): Likewise.
1401 * rtlanal.c (nonzero_bits1): Likewise.
1402 * calls.c (store_one_arg): Likewise. Require the padding to be
1403 known at compile time.
1405 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1406 Alan Hayward <alan.hayward@arm.com>
1407 David Sherwood <david.sherwood@arm.com>
1409 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
1410 Use plus_constant instead of gen_rtx_PLUS.
1412 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1413 Alan Hayward <alan.hayward@arm.com>
1414 David Sherwood <david.sherwood@arm.com>
1416 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
1419 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1420 Alan Hayward <alan.hayward@arm.com>
1421 David Sherwood <david.sherwood@arm.com>
1423 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
1424 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
1425 via stack temporaries. Treat the mode size as polynomial too.
1427 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1428 Alan Hayward <alan.hayward@arm.com>
1429 David Sherwood <david.sherwood@arm.com>
1431 * expr.c (expand_expr_real_2): When handling conversions involving
1432 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
1433 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
1434 as a poly_uint64 too.
1436 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1437 Alan Hayward <alan.hayward@arm.com>
1438 David Sherwood <david.sherwood@arm.com>
1440 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
1442 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1443 Alan Hayward <alan.hayward@arm.com>
1444 David Sherwood <david.sherwood@arm.com>
1446 * combine.c (can_change_dest_mode): Handle polynomial
1447 REGMODE_NATURAL_SIZE.
1448 * expmed.c (store_bit_field_1): Likewise.
1449 * expr.c (store_constructor): Likewise.
1450 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
1451 and polynomial REGMODE_NATURAL_SIZE.
1452 (gen_lowpart_common): Likewise.
1453 * reginfo.c (record_subregs_of_mode): Likewise.
1454 * rtlanal.c (read_modify_subreg_p): Likewise.
1456 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1457 Alan Hayward <alan.hayward@arm.com>
1458 David Sherwood <david.sherwood@arm.com>
1460 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
1461 numbers of elements.
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 * match.pd: Cope with polynomial numbers of vector elements.
1469 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1470 Alan Hayward <alan.hayward@arm.com>
1471 David Sherwood <david.sherwood@arm.com>
1473 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
1474 in a POINTER_PLUS_EXPR.
1476 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1477 Alan Hayward <alan.hayward@arm.com>
1478 David Sherwood <david.sherwood@arm.com>
1480 * omp-simd-clone.c (simd_clone_subparts): New function.
1481 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
1482 (ipa_simd_modify_function_body): Likewise.
1484 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1485 Alan Hayward <alan.hayward@arm.com>
1486 David Sherwood <david.sherwood@arm.com>
1488 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
1489 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
1490 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
1491 (expand_vector_condition, vector_element): Likewise.
1492 (subparts_gt): New function.
1493 (get_compute_type): Use subparts_gt.
1494 (count_type_subparts): Delete.
1495 (expand_vector_operations_1): Use subparts_gt instead of
1496 count_type_subparts.
1498 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1499 Alan Hayward <alan.hayward@arm.com>
1500 David Sherwood <david.sherwood@arm.com>
1502 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
1503 (vect_compile_time_alias): ...this new function. Do the calculation
1504 on poly_ints rather than trees.
1505 (vect_prune_runtime_alias_test_list): Update call accordingly.
1507 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1508 Alan Hayward <alan.hayward@arm.com>
1509 David Sherwood <david.sherwood@arm.com>
1511 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
1513 (vect_schedule_slp_instance): Likewise.
1515 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1516 Alan Hayward <alan.hayward@arm.com>
1517 David Sherwood <david.sherwood@arm.com>
1519 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
1520 constant and extern definitions for variable-length vectors.
1521 (vect_get_constant_vectors): Note that the number of units
1522 is known to be constant.
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-stmts.c (vectorizable_conversion): Treat the number
1529 of units as polynomial. Choose between WIDE and NARROW based
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-stmts.c (simd_clone_subparts): New function.
1537 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
1539 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1540 Alan Hayward <alan.hayward@arm.com>
1541 David Sherwood <david.sherwood@arm.com>
1543 * tree-vect-stmts.c (vectorizable_call): Treat the number of
1544 vectors as polynomial. Use build_index_vector for
1547 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1548 Alan Hayward <alan.hayward@arm.com>
1549 David Sherwood <david.sherwood@arm.com>
1551 * tree-vect-stmts.c (get_load_store_type): Treat the number of
1552 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
1553 for variable-length vectors.
1554 (vectorizable_mask_load_store): Treat the number of units as
1555 polynomial, asserting that it is constant if the condition has
1556 already been enforced.
1557 (vectorizable_store, vectorizable_load): Likewise.
1559 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1560 Alan Hayward <alan.hayward@arm.com>
1561 David Sherwood <david.sherwood@arm.com>
1563 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
1564 of units as polynomial. Punt if we can't tell at compile time
1565 which vector contains the final result.
1567 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1568 Alan Hayward <alan.hayward@arm.com>
1569 David Sherwood <david.sherwood@arm.com>
1571 * tree-vect-loop.c (vectorizable_induction): Treat the number
1572 of units as polynomial. Punt on SLP inductions. Use an integer
1573 VEC_SERIES_EXPR for variable-length integer reductions. Use a
1574 cast of such a series for variable-length floating-point
1577 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1578 Alan Hayward <alan.hayward@arm.com>
1579 David Sherwood <david.sherwood@arm.com>
1581 * tree.h (build_index_vector): Declare.
1582 * tree.c (build_index_vector): New function.
1583 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
1584 of units as polynomial, forcibly converting it to a constant if
1585 vectorizable_reduction has already enforced the condition.
1586 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
1587 to create a {1,2,3,...} vector.
1588 (vectorizable_reduction): Treat the number of units as polynomial.
1589 Choose vectype_in based on the largest scalar element size rather
1590 than the smallest number of units. Enforce the restrictions
1593 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1594 Alan Hayward <alan.hayward@arm.com>
1595 David Sherwood <david.sherwood@arm.com>
1597 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
1598 number of units as polynomial.
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.h (vector_sizes, auto_vector_sizes): New typedefs.
1605 * target.def (autovectorize_vector_sizes): Return the vector sizes
1606 by pointer, using vector_sizes rather than a bitmask.
1607 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
1608 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
1609 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
1611 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
1612 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
1613 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
1614 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
1615 * omp-general.c (omp_max_vf): Likewise.
1616 * omp-low.c (omp_clause_aligned_alignment): Likewise.
1617 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
1618 * tree-vect-loop.c (vect_analyze_loop): Likewise.
1619 * tree-vect-slp.c (vect_slp_bb): Likewise.
1620 * doc/tm.texi: Regenerate.
1621 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
1623 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
1624 the vector size as a poly_uint64 rather than an unsigned int.
1625 (current_vector_size): Change from an unsigned int to a poly_uint64.
1626 (get_vectype_for_scalar_type): Update accordingly.
1627 * tree.h (build_truth_vector_type): Take the size and number of
1628 units as a poly_uint64 rather than an unsigned int.
1629 (build_vector_type): Add a temporary overload that takes
1630 the number of units as a poly_uint64 rather than an unsigned int.
1631 * tree.c (make_vector_type): Likewise.
1632 (build_truth_vector_type): Take the number of units as a poly_uint64
1633 rather than an unsigned int.
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 * target.def (get_mask_mode): Take the number of units and length
1640 as poly_uint64s rather than unsigned ints.
1641 * targhooks.h (default_get_mask_mode): Update accordingly.
1642 * targhooks.c (default_get_mask_mode): Likewise.
1643 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
1644 * doc/tm.texi: Regenerate.
1646 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1647 Alan Hayward <alan.hayward@arm.com>
1648 David Sherwood <david.sherwood@arm.com>
1650 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
1651 * omp-general.c (omp_max_vf): Likewise.
1652 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
1653 (expand_omp_simd): Handle polynomial safelen.
1654 * omp-low.c (omplow_simd_context): Add a default constructor.
1655 (omplow_simd_context::max_vf): Change from int to poly_uint64.
1656 (lower_rec_simd_input_clauses): Update accordingly.
1657 (lower_rec_input_clauses): Likewise.
1659 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1660 Alan Hayward <alan.hayward@arm.com>
1661 David Sherwood <david.sherwood@arm.com>
1663 * tree-vectorizer.h (vect_nunits_for_cost): New function.
1664 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
1665 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
1666 (vect_analyze_slp_cost): Likewise.
1667 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
1668 (vect_model_load_cost): Likewise.
1670 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1671 Alan Hayward <alan.hayward@arm.com>
1672 David Sherwood <david.sherwood@arm.com>
1674 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
1675 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
1676 from an unsigned int * to a poly_uint64_pod *.
1677 (calculate_unrolling_factor): New function.
1678 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
1680 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1681 Alan Hayward <alan.hayward@arm.com>
1682 David Sherwood <david.sherwood@arm.com>
1684 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
1685 from an unsigned int to a poly_uint64.
1686 (_loop_vec_info::slp_unrolling_factor): Likewise.
1687 (_loop_vec_info::vectorization_factor): Change from an int
1689 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
1690 (vect_get_num_vectors): New function.
1691 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
1692 (vect_get_num_copies): Use vect_get_num_vectors.
1693 (vect_analyze_data_ref_dependences): Change max_vf from an int *
1694 to an unsigned int *.
1695 (vect_analyze_data_refs): Change min_vf from an int * to a
1697 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1698 than an unsigned HOST_WIDE_INT.
1699 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
1700 (vect_analyze_data_ref_dependence): Change max_vf from an int *
1701 to an unsigned int *.
1702 (vect_analyze_data_ref_dependences): Likewise.
1703 (vect_compute_data_ref_alignment): Handle polynomial vf.
1704 (vect_enhance_data_refs_alignment): Likewise.
1705 (vect_prune_runtime_alias_test_list): Likewise.
1706 (vect_shift_permute_load_chain): Likewise.
1707 (vect_supportable_dr_alignment): Likewise.
1708 (dependence_distance_ge_vf): Take the vectorization factor as a
1709 poly_uint64 rather than an unsigned HOST_WIDE_INT.
1710 (vect_analyze_data_refs): Change min_vf from an int * to a
1712 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
1713 vfm1 as a poly_uint64 rather than an int. Make the same change
1714 for the returned bound_scalar.
1715 (vect_gen_vector_loop_niters): Handle polynomial vf.
1716 (vect_do_peeling): Likewise. Update call to
1717 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
1718 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
1720 * tree-vect-loop.c (vect_determine_vectorization_factor)
1721 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
1722 (vect_get_known_peeling_cost): Likewise.
1723 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
1724 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
1725 (vect_transform_loop): Likewise. Use the lowest possible VF when
1726 updating the upper bounds of the loop.
1727 (vect_min_worthwhile_factor): Make static. Return an unsigned int
1729 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
1730 polynomial unroll factors.
1731 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
1732 (vect_make_slp_decision): Likewise.
1733 (vect_supported_load_permutation_p): Likewise, and polynomial
1735 (vect_analyze_slp_cost): Handle polynomial vf.
1736 (vect_slp_analyze_node_operations): Likewise.
1737 (vect_slp_analyze_bb_1): Likewise.
1738 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1739 than an unsigned HOST_WIDE_INT.
1740 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
1741 (vectorizable_load): Handle polynomial vf.
1742 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
1744 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
1746 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1747 Alan Hayward <alan.hayward@arm.com>
1748 David Sherwood <david.sherwood@arm.com>
1750 * match.pd: Handle bit operations involving three constants
1751 and try to fold one pair.
1753 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
1755 * tree-vect-loop-manip.c: Include gimple-fold.h.
1756 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
1757 niters_maybe_zero parameters. Handle other cases besides a step of 1.
1758 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
1759 Add a path that uses a step of VF instead of 1, but disable it
1761 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
1762 and niters_no_overflow parameters. Update calls to
1763 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
1764 Create a new SSA name if the latter choses to use a ste other
1765 than zero, and return it via niters_vector_mult_vf_var.
1766 * tree-vect-loop.c (vect_transform_loop): Update calls to
1767 vect_do_peeling, vect_gen_vector_loop_niters and
1768 slpeel_make_loop_iterate_ntimes.
1769 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
1770 (vect_gen_vector_loop_niters): Update declarations after above changes.
1772 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
1774 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
1775 128-bit round to integer instructions.
1776 (ceil<mode>2): Likewise.
1777 (btrunc<mode>2): Likewise.
1778 (round<mode>2): Likewise.
1780 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1782 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
1783 unaligned VSX load/store on P8/P9.
1784 (expand_block_clear): Allow the use of unaligned VSX
1785 load/store on P8/P9.
1787 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1789 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
1791 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
1792 swap associated with both a load and a store.
1794 2018-01-02 Andrew Waterman <andrew@sifive.com>
1796 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
1797 * config/riscv/riscv.md (clear_cache): Use it.
1799 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
1801 * web.c: Remove out-of-date comment.
1803 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1805 * expr.c (fixup_args_size_notes): Check that any existing
1806 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
1807 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
1808 (emit_single_push_insn): ...here.
1810 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1812 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
1813 (const_vector_encoded_nelts): New function.
1814 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
1815 (const_vector_int_elt, const_vector_elt): Declare.
1816 * emit-rtl.c (const_vector_int_elt_1): New function.
1817 (const_vector_elt): Likewise.
1818 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
1819 of CONST_VECTOR_ELT.
1821 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1823 * expr.c: Include rtx-vector-builder.h.
1824 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
1825 directly on the tree encoding.
1826 (const_vector_from_tree): Likewise.
1827 * optabs.c: Include rtx-vector-builder.h.
1828 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
1829 sequence of "u" values.
1830 * vec-perm-indices.c: Include rtx-vector-builder.h.
1831 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
1832 directly on the vec_perm_indices encoding.
1834 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1836 * doc/rtl.texi (const_vector): Describe new encoding scheme.
1837 * Makefile.in (OBJS): Add rtx-vector-builder.o.
1838 * rtx-vector-builder.h: New file.
1839 * rtx-vector-builder.c: Likewise.
1840 * rtl.h (rtx_def::u2): Add a const_vector field.
1841 (CONST_VECTOR_NPATTERNS): New macro.
1842 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
1843 (CONST_VECTOR_DUPLICATE_P): Likewise.
1844 (CONST_VECTOR_STEPPED_P): Likewise.
1845 (CONST_VECTOR_ENCODED_ELT): Likewise.
1846 (const_vec_duplicate_p): Check for a duplicated vector encoding.
1847 (unwrap_const_vec_duplicate): Likewise.
1848 (const_vec_series_p): Check for a non-duplicated vector encoding.
1849 Say that the function only returns true for integer vectors.
1850 * emit-rtl.c: Include rtx-vector-builder.h.
1851 (gen_const_vec_duplicate_1): Delete.
1852 (gen_const_vector): Call gen_const_vec_duplicate instead of
1853 gen_const_vec_duplicate_1.
1854 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
1855 (gen_const_vec_duplicate): Use rtx_vector_builder.
1856 (gen_const_vec_series): Likewise.
1857 (gen_rtx_CONST_VECTOR): Likewise.
1858 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
1859 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1860 Build a new vector rather than modifying a CONST_VECTOR in-place.
1861 (handle_special_swappables): Update call accordingly.
1862 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
1863 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1864 Build a new vector rather than modifying a CONST_VECTOR in-place.
1865 (handle_special_swappables): Update call accordingly.
1867 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1869 * simplify-rtx.c (simplify_const_binary_operation): Use
1870 CONST_VECTOR_ELT instead of XVECEXP.
1872 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1874 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
1875 the selector elements to be different from the data elements
1876 if the selector is a VECTOR_CST.
1877 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
1878 ssizetype for the selector.
1880 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1882 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
1883 before testing each element individually.
1884 * tree-vect-generic.c (lower_vec_perm): Likewise.
1886 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1888 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
1889 * selftest-run-tests.c (selftest::run_tests): Call it.
1890 * vector-builder.h (vector_builder::operator ==): New function.
1891 (vector_builder::operator !=): Likewise.
1892 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
1893 (vec_perm_indices::all_from_input_p): New function.
1894 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
1895 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
1896 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
1897 instead of reading the VECTOR_CST directly. Detect whether both
1898 vector inputs are the same before constructing the vec_perm_indices,
1899 and update the number of inputs argument accordingly. Use the
1900 utility functions added above. Only construct sel2 if we need to.
1902 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1904 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
1905 the broadcast of the low byte.
1906 (expand_mult_highpart): Use an explicit encoding for the permutes.
1907 * optabs-query.c (can_mult_highpart_p): Likewise.
1908 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
1909 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1910 (vectorizable_bswap): Likewise.
1911 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
1912 explicit encoding for the power-of-2 permutes.
1913 (vect_permute_store_chain): Likewise.
1914 (vect_grouped_load_supported): Likewise.
1915 (vect_permute_load_chain): Likewise.
1917 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1919 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
1920 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
1921 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
1922 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
1923 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
1924 (vect_gen_perm_mask_any): Likewise.
1926 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1928 * int-vector-builder.h: New file.
1929 * vec-perm-indices.h: Include int-vector-builder.h.
1930 (vec_perm_indices): Redefine as an int_vector_builder.
1931 (auto_vec_perm_indices): Delete.
1932 (vec_perm_builder): Redefine as a stand-alone class.
1933 (vec_perm_indices::vec_perm_indices): New function.
1934 (vec_perm_indices::clamp): Likewise.
1935 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
1936 (vec_perm_indices::new_vector): New function.
1937 (vec_perm_indices::new_expanded_vector): Update for new
1938 vec_perm_indices class.
1939 (vec_perm_indices::rotate_inputs): New function.
1940 (vec_perm_indices::all_in_range_p): Operate directly on the
1941 encoded form, without computing elided elements.
1942 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
1943 encoding. Update for new vec_perm_indices class.
1944 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
1945 the given vec_perm_builder.
1946 (expand_vec_perm_var): Update vec_perm_builder constructor.
1947 (expand_mult_highpart): Use vec_perm_builder instead of
1948 auto_vec_perm_indices.
1949 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
1950 vec_perm_indices instead of auto_vec_perm_indices. Use a single
1951 or double series encoding as appropriate.
1952 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
1953 vec_perm_indices instead of auto_vec_perm_indices.
1954 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1955 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1956 (vect_permute_store_chain): Likewise.
1957 (vect_grouped_load_supported): Likewise.
1958 (vect_permute_load_chain): Likewise.
1959 (vect_shift_permute_load_chain): Likewise.
1960 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1961 (vect_transform_slp_perm_load): Likewise.
1962 (vect_schedule_slp_instance): Likewise.
1963 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1964 (vectorizable_mask_load_store): Likewise.
1965 (vectorizable_bswap): Likewise.
1966 (vectorizable_store): Likewise.
1967 (vectorizable_load): Likewise.
1968 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
1969 vec_perm_indices instead of auto_vec_perm_indices. Use
1970 tree_to_vec_perm_builder to read the vector from a tree.
1971 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
1972 vec_perm_builder instead of a vec_perm_indices.
1973 (have_whole_vector_shift): Use vec_perm_builder and
1974 vec_perm_indices instead of auto_vec_perm_indices. Leave the
1975 truncation to calc_vec_perm_mask_for_shift.
1976 (vect_create_epilog_for_reduction): Likewise.
1977 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
1978 from auto_vec_perm_indices to vec_perm_indices.
1979 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1980 instead of changing individual elements.
1981 (aarch64_vectorize_vec_perm_const): Use new_vector to install
1982 the vector in d.perm.
1983 * config/arm/arm.c (expand_vec_perm_d::perm): Change
1984 from auto_vec_perm_indices to vec_perm_indices.
1985 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1986 instead of changing individual elements.
1987 (arm_vectorize_vec_perm_const): Use new_vector to install
1988 the vector in d.perm.
1989 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
1990 Update vec_perm_builder constructor.
1991 (rs6000_expand_interleave): Likewise.
1992 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
1993 (rs6000_expand_interleave): Likewise.
1995 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1997 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
1998 to qimode could truncate the indices.
1999 * optabs.c (expand_vec_perm_var): Likewise.
2001 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2003 * Makefile.in (OBJS): Add vec-perm-indices.o.
2004 * vec-perm-indices.h: New file.
2005 * vec-perm-indices.c: Likewise.
2006 * target.h (vec_perm_indices): Replace with a forward class
2008 (auto_vec_perm_indices): Move to vec-perm-indices.h.
2009 * optabs.h: Include vec-perm-indices.h.
2010 (expand_vec_perm): Delete.
2011 (selector_fits_mode_p, expand_vec_perm_var): Declare.
2012 (expand_vec_perm_const): Declare.
2013 * target.def (vec_perm_const_ok): Replace with...
2014 (vec_perm_const): ...this new hook.
2015 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
2016 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
2017 * doc/tm.texi: Regenerate.
2018 * optabs.def (vec_perm_const): Delete.
2019 * doc/md.texi (vec_perm_const): Likewise.
2020 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
2021 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
2022 expand_vec_perm for constant permutation vectors. Assert that
2023 the mode of variable permutation vectors is the integer equivalent
2024 of the mode that is being permuted.
2025 * optabs-query.h (selector_fits_mode_p): Declare.
2026 * optabs-query.c: Include vec-perm-indices.h.
2027 (selector_fits_mode_p): New function.
2028 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
2029 is defined, instead of checking whether the vec_perm_const_optab
2030 exists. Use targetm.vectorize.vec_perm_const instead of
2031 targetm.vectorize.vec_perm_const_ok. Check whether the indices
2032 fit in the vector mode before using a variable permute.
2033 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
2034 vec_perm_indices instead of an rtx.
2035 (expand_vec_perm): Replace with...
2036 (expand_vec_perm_const): ...this new function. Take the selector
2037 as a vec_perm_indices rather than an rtx. Also take the mode of
2038 the selector. Update call to shift_amt_for_vec_perm_mask.
2039 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
2040 Use vec_perm_indices::new_expanded_vector to expand the original
2041 selector into bytes. Check whether the indices fit in the vector
2042 mode before using a variable permute.
2043 (expand_vec_perm_var): Make global.
2044 (expand_mult_highpart): Use expand_vec_perm_const.
2045 * fold-const.c: Includes vec-perm-indices.h.
2046 * tree-ssa-forwprop.c: Likewise.
2047 * tree-vect-data-refs.c: Likewise.
2048 * tree-vect-generic.c: Likewise.
2049 * tree-vect-loop.c: Likewise.
2050 * tree-vect-slp.c: Likewise.
2051 * tree-vect-stmts.c: Likewise.
2052 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
2054 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
2055 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
2056 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
2057 (aarch64_vectorize_vec_perm_const): ...this new function.
2058 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
2059 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2060 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
2061 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
2062 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
2063 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2064 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
2066 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
2067 check for NEON modes.
2068 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
2069 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
2070 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
2071 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
2073 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
2074 the old VEC_PERM_CONST conditions.
2075 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
2076 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
2077 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
2078 (ia64_vectorize_vec_perm_const_ok): Merge into...
2079 (ia64_vectorize_vec_perm_const): ...this new function.
2080 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
2081 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
2082 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
2083 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
2084 * config/mips/mips.c (mips_expand_vec_perm_const)
2085 (mips_vectorize_vec_perm_const_ok): Merge into...
2086 (mips_vectorize_vec_perm_const): ...this new function.
2087 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
2088 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
2089 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
2090 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
2091 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
2092 (rs6000_expand_vec_perm_const): Delete.
2093 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
2095 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2096 (altivec_expand_vec_perm_const_le): Take each operand individually.
2097 Operate on constant selectors rather than rtxes.
2098 (altivec_expand_vec_perm_const): Likewise. Update call to
2099 altivec_expand_vec_perm_const_le.
2100 (rs6000_expand_vec_perm_const): Delete.
2101 (rs6000_vectorize_vec_perm_const_ok): Delete.
2102 (rs6000_vectorize_vec_perm_const): New function.
2103 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
2104 an element count and rtx array.
2105 (rs6000_expand_extract_even): Update call accordingly.
2106 (rs6000_expand_interleave): Likewise.
2107 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
2108 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
2109 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
2110 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
2111 (rs6000_expand_vec_perm_const): Delete.
2112 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
2113 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2114 (altivec_expand_vec_perm_const_le): Take each operand individually.
2115 Operate on constant selectors rather than rtxes.
2116 (altivec_expand_vec_perm_const): Likewise. Update call to
2117 altivec_expand_vec_perm_const_le.
2118 (rs6000_expand_vec_perm_const): Delete.
2119 (rs6000_vectorize_vec_perm_const_ok): Delete.
2120 (rs6000_vectorize_vec_perm_const): New function. Remove stray
2121 reference to the SPE evmerge intructions.
2122 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
2123 an element count and rtx array.
2124 (rs6000_expand_extract_even): Update call accordingly.
2125 (rs6000_expand_interleave): Likewise.
2126 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
2127 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
2129 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
2131 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2133 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
2134 vector mode and that that mode matches the mode of the data
2136 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
2137 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
2138 directly using expand_vec_perm_1 when forcing selectors into
2140 (expand_vec_perm_var): New function, split out from expand_vec_perm.
2142 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2144 * optabs-query.h (can_vec_perm_p): Delete.
2145 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
2146 * optabs-query.c (can_vec_perm_p): Split into...
2147 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
2148 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
2149 particular selector is valid.
2150 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
2151 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
2152 (vect_grouped_load_supported): Likewise.
2153 (vect_shift_permute_load_chain): Likewise.
2154 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
2155 (vect_transform_slp_perm_load): Likewise.
2156 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
2157 (vectorizable_bswap): Likewise.
2158 (vect_gen_perm_mask_checked): Likewise.
2159 * fold-const.c (fold_ternary_loc): Likewise. Don't take
2160 implementations of variable permutation vectors into account
2161 when deciding which selector to use.
2162 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
2163 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
2164 with a false third argument.
2165 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
2166 to test whether the constant selector is valid and can_vec_perm_var_p
2167 to test whether a variable selector is valid.
2169 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2171 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
2172 * optabs-query.c (can_vec_perm_p): Likewise.
2173 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
2174 instead of vec_perm_indices.
2175 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
2176 (vect_gen_perm_mask_checked): Likewise,
2177 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
2178 (vect_gen_perm_mask_checked): Likewise,
2180 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
2182 * optabs-query.h (qimode_for_vec_perm): Declare.
2183 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
2184 (qimode_for_vec_perm): ...this new function.
2185 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
2187 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2189 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
2190 does not have a conditional at the top.
2192 2018-01-02 Richard Biener <rguenther@suse.de>
2194 * ipa-inline.c (big_speedup_p): Fix expression.
2196 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
2199 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
2202 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
2206 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
2207 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
2208 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
2209 cond_taken_branch_cost 3->4.
2211 2018-01-01 Jakub Jelinek <jakub@redhat.com>
2213 PR tree-optimization/83581
2214 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
2215 TODO_cleanup_cfg if any changes have been made.
2218 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
2219 convert_modes if target mode has the right side, but different mode
2223 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
2224 last argument when extracting from CONCAT. If either from_real or
2225 from_imag is NULL, use expansion through memory. If result is not
2226 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
2227 the parts directly to inner mode, if even that fails, use expansion
2231 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
2232 check for bswap in mode rather than HImode and use that in expand_unop
2235 Copyright (C) 2018 Free Software Foundation, Inc.
2237 Copying and distribution of this file, with or without modification,
2238 are permitted in any medium without royalty provided the copyright
2239 notice and this notice are preserved.