-
Notifications
You must be signed in to change notification settings - Fork 224
/
Copy pathvectorIntrinsics.cpp
1910 lines (1711 loc) · 77.8 KB
/
vectorIntrinsics.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*
* Copyright (c) 2020, 2021, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License version 2 only, as
* published by the Free Software Foundation.
*
* This code is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
* version 2 for more details (a copy is included in the LICENSE file that
* accompanied this code).
*
* You should have received a copy of the GNU General Public License version
* 2 along with this work; if not, write to the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
*
* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
* or visit www.oracle.com if you need additional information or have any
* questions.
*
*/
#include "precompiled.hpp"
#include "ci/ciSymbols.hpp"
#include "classfile/vmSymbols.hpp"
#include "opto/library_call.hpp"
#include "opto/runtime.hpp"
#include "opto/vectornode.hpp"
#include "prims/vectorSupport.hpp"
#include "runtime/stubRoutines.hpp"
#ifdef ASSERT
static bool is_vector(ciKlass* klass) {
return klass->is_subclass_of(ciEnv::current()->vector_VectorPayload_klass());
}
static bool check_vbox(const TypeInstPtr* vbox_type) {
assert(vbox_type->klass_is_exact(), "");
ciInstanceKlass* ik = vbox_type->klass()->as_instance_klass();
assert(is_vector(ik), "not a vector");
ciField* fd1 = ik->get_field_by_name(ciSymbols::ETYPE_name(), ciSymbols::class_signature(), /* is_static */ true);
assert(fd1 != nullptr, "element type info is missing");
ciConstant val1 = fd1->constant_value();
BasicType elem_bt = val1.as_object()->as_instance()->java_mirror_type()->basic_type();
assert(is_java_primitive(elem_bt), "element type info is missing");
ciField* fd2 = ik->get_field_by_name(ciSymbols::VLENGTH_name(), ciSymbols::int_signature(), /* is_static */ true);
assert(fd2 != nullptr, "vector length info is missing");
ciConstant val2 = fd2->constant_value();
assert(val2.as_int() > 0, "vector length info is missing");
return true;
}
#endif
Node* GraphKit::box_vector(Node* vector, const TypeInstPtr* vbox_type, BasicType elem_bt, int num_elem, bool deoptimize_on_exception) {
assert(EnableVectorSupport, "");
PreserveReexecuteState preexecs(this);
jvms()->set_should_reexecute(true);
VectorBoxAllocateNode* alloc = new VectorBoxAllocateNode(C, vbox_type);
set_edges_for_java_call(alloc, /*must_throw=*/false, /*separate_io_proj=*/true);
make_slow_call_ex(alloc, env()->Throwable_klass(), /*separate_io_proj=*/true, deoptimize_on_exception);
set_i_o(gvn().transform( new ProjNode(alloc, TypeFunc::I_O) ));
set_all_memory(gvn().transform( new ProjNode(alloc, TypeFunc::Memory) ));
Node* ret = gvn().transform(new ProjNode(alloc, TypeFunc::Parms));
assert(check_vbox(vbox_type), "");
const TypeVect* vt = TypeVect::make(elem_bt, num_elem);
VectorBoxNode* vbox = new VectorBoxNode(C, ret, vector, vbox_type, vt);
return gvn().transform(vbox);
}
Node* GraphKit::unbox_vector(Node* v, const TypeInstPtr* vbox_type, BasicType elem_bt, int num_elem, bool shuffle_to_vector) {
assert(EnableVectorSupport, "");
const TypeInstPtr* vbox_type_v = gvn().type(v)->is_instptr();
if (vbox_type->klass() != vbox_type_v->klass()) {
return nullptr; // arguments don't agree on vector shapes
}
if (vbox_type_v->maybe_null()) {
return nullptr; // no nulls are allowed
}
assert(check_vbox(vbox_type), "");
const TypeVect* vt = TypeVect::make(elem_bt, num_elem);
Node* unbox = gvn().transform(new VectorUnboxNode(C, vt, v, merged_memory(), shuffle_to_vector));
return unbox;
}
Node* GraphKit::vector_shift_count(Node* cnt, int shift_op, BasicType bt, int num_elem) {
assert(bt == T_INT || bt == T_LONG || bt == T_SHORT || bt == T_BYTE, "byte, short, long and int are supported");
juint mask = (type2aelembytes(bt) * BitsPerByte - 1);
Node* nmask = gvn().transform(ConNode::make(TypeInt::make(mask)));
Node* mcnt = gvn().transform(new AndINode(cnt, nmask));
return gvn().transform(VectorNode::shift_count(shift_op, mcnt, num_elem, bt));
}
bool LibraryCallKit::arch_supports_vector(int sopc, int num_elem, BasicType type, VectorMaskUseType mask_use_type, bool has_scalar_args) {
// Check that the operation is valid.
if (sopc <= 0) {
#ifndef PRODUCT
if (C->print_intrinsics()) {
tty->print_cr(" ** Rejected intrinsification because no valid vector op could be extracted");
}
#endif
return false;
}
// Check that architecture supports this op-size-type combination.
if (!Matcher::match_rule_supported_vector(sopc, num_elem, type)) {
#ifndef PRODUCT
if (C->print_intrinsics()) {
tty->print_cr(" ** Rejected vector op (%s,%s,%d) because architecture does not support it",
NodeClassNames[sopc], type2name(type), num_elem);
}
#endif
return false;
} else {
assert(Matcher::match_rule_supported(sopc), "must be supported");
}
if (num_elem == 1) {
if (mask_use_type != VecMaskNotUsed) {
#ifndef PRODUCT
if (C->print_intrinsics()) {
tty->print_cr(" ** Rejected vector mask op (%s,%s,%d) because architecture does not support it",
NodeClassNames[sopc], type2name(type), num_elem);
}
#endif
return false;
}
if (sopc != 0) {
if (sopc != Op_LoadVector && sopc != Op_StoreVector) {
#ifndef PRODUCT
if (C->print_intrinsics()) {
tty->print_cr(" ** Not a svml call or load/store vector op (%s,%s,%d)",
NodeClassNames[sopc], type2name(type), num_elem);
}
#endif
return false;
}
}
}
if (!has_scalar_args && VectorNode::is_vector_shift(sopc) &&
Matcher::supports_vector_variable_shifts() == false) {
if (C->print_intrinsics()) {
tty->print_cr(" ** Rejected vector op (%s,%s,%d) because architecture does not support variable vector shifts",
NodeClassNames[sopc], type2name(type), num_elem);
}
return false;
}
// Check whether mask unboxing is supported.
if (mask_use_type == VecMaskUseAll || mask_use_type == VecMaskUseLoad) {
if (!Matcher::match_rule_supported_vector(Op_VectorLoadMask, num_elem, type)) {
#ifndef PRODUCT
if (C->print_intrinsics()) {
tty->print_cr(" ** Rejected vector mask loading (%s,%s,%d) because architecture does not support it",
NodeClassNames[Op_VectorLoadMask], type2name(type), num_elem);
}
#endif
return false;
}
}
// Check whether mask boxing is supported.
if (mask_use_type == VecMaskUseAll || mask_use_type == VecMaskUseStore) {
if (!Matcher::match_rule_supported_vector(Op_VectorStoreMask, num_elem, type)) {
#ifndef PRODUCT
if (C->print_intrinsics()) {
tty->print_cr("Rejected vector mask storing (%s,%s,%d) because architecture does not support it",
NodeClassNames[Op_VectorStoreMask], type2name(type), num_elem);
}
#endif
return false;
}
}
return true;
}
static bool is_vector_mask(ciKlass* klass) {
return klass->is_subclass_of(ciEnv::current()->vector_VectorMask_klass());
}
static bool is_vector_shuffle(ciKlass* klass) {
return klass->is_subclass_of(ciEnv::current()->vector_VectorShuffle_klass());
}
static bool is_klass_initialized(const TypeInstPtr* vec_klass) {
if (vec_klass->const_oop() == nullptr) {
return false; // uninitialized or some kind of unsafe access
}
assert(vec_klass->const_oop()->as_instance()->java_lang_Class_klass() != nullptr, "klass instance expected");
ciInstanceKlass* klass = vec_klass->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
return klass->is_initialized();
}
// public static
// <VM>
// VM unaryOp(int oprId, Class<? extends VM> vmClass, Class<?> elementType, int length,
// VM vm,
// Function<VM, VM> defaultImpl) {
//
// public static
// <VM>
// VM binaryOp(int oprId, Class<? extends VM> vmClass, Class<?> elementType, int length,
// VM vm1, VM vm2,
// BiFunction<VM, VM, VM> defaultImpl) {
//
// public static
// <VM>
// VM ternaryOp(int oprId, Class<? extends VM> vmClass, Class<?> elementType, int length,
// VM vm1, VM vm2, VM vm3,
// TernaryOperation<VM> defaultImpl) {
//
bool LibraryCallKit::inline_vector_nary_operation(int n) {
const TypeInt* opr = gvn().type(argument(0))->isa_int();
const TypeInstPtr* vector_klass = gvn().type(argument(1))->isa_instptr();
const TypeInstPtr* elem_klass = gvn().type(argument(2))->isa_instptr();
const TypeInt* vlen = gvn().type(argument(3))->isa_int();
if (opr == nullptr || vector_klass == nullptr || elem_klass == nullptr || vlen == nullptr ||
!opr->is_con() || vector_klass->const_oop() == nullptr || elem_klass->const_oop() == nullptr || !vlen->is_con()) {
if (C->print_intrinsics()) {
tty->print_cr(" ** missing constant: opr=%s vclass=%s etype=%s vlen=%s",
NodeClassNames[argument(0)->Opcode()],
NodeClassNames[argument(1)->Opcode()],
NodeClassNames[argument(2)->Opcode()],
NodeClassNames[argument(3)->Opcode()]);
}
return false; // not enough info for intrinsification
}
ciType* elem_type = elem_klass->const_oop()->as_instance()->java_mirror_type();
if (!elem_type->is_primitive_type()) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not a primitive bt=%d", elem_type->basic_type());
}
return false; // should be primitive type
}
if (!is_klass_initialized(vector_klass)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** klass argument not initialized");
}
return false;
}
BasicType elem_bt = elem_type->basic_type();
int num_elem = vlen->get_con();
int opc = VectorSupport::vop2ideal(opr->get_con(), elem_bt);
int sopc = VectorNode::opcode(opc, elem_bt);
if ((opc != Op_CallLeafVector) && (sopc == 0)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** operation not supported: opc=%s bt=%s", NodeClassNames[opc], type2name(elem_bt));
}
return false; // operation not supported
}
if (num_elem == 1) {
if (opc != Op_CallLeafVector || elem_bt != T_DOUBLE) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not a svml call: arity=%d opc=%d vlen=%d etype=%s",
n, opc, num_elem, type2name(elem_bt));
}
return false;
}
}
ciKlass* vbox_klass = vector_klass->const_oop()->as_instance()->java_lang_Class_klass();
const TypeInstPtr* vbox_type = TypeInstPtr::make_exact(TypePtr::NotNull, vbox_klass);
if (opc == Op_CallLeafVector) {
if (!UseVectorStubs) {
if (C->print_intrinsics()) {
tty->print_cr(" ** vector stubs support is disabled");
}
return false;
}
if (!Matcher::supports_vector_calling_convention()) {
if (C->print_intrinsics()) {
tty->print_cr(" ** no vector calling conventions supported");
}
return false;
}
if (!Matcher::vector_size_supported(elem_bt, num_elem)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** vector size (vlen=%d, etype=%s) is not supported",
num_elem, type2name(elem_bt));
}
return false;
}
}
// TODO When mask usage is supported, VecMaskNotUsed needs to be VecMaskUseLoad.
if ((sopc != 0) &&
!arch_supports_vector(sopc, num_elem, elem_bt, is_vector_mask(vbox_klass) ? VecMaskUseAll : VecMaskNotUsed)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not supported: arity=%d opc=%d vlen=%d etype=%s ismask=%d",
n, sopc, num_elem, type2name(elem_bt),
is_vector_mask(vbox_klass) ? 1 : 0);
}
return false; // not supported
}
Node* opd1 = nullptr; Node* opd2 = nullptr; Node* opd3 = nullptr;
switch (n) {
case 3: {
opd3 = unbox_vector(argument(6), vbox_type, elem_bt, num_elem);
if (opd3 == nullptr) {
if (C->print_intrinsics()) {
tty->print_cr(" ** unbox failed v3=%s",
NodeClassNames[argument(6)->Opcode()]);
}
return false;
}
// fall-through
}
case 2: {
opd2 = unbox_vector(argument(5), vbox_type, elem_bt, num_elem);
if (opd2 == nullptr) {
if (C->print_intrinsics()) {
tty->print_cr(" ** unbox failed v2=%s",
NodeClassNames[argument(5)->Opcode()]);
}
return false;
}
// fall-through
}
case 1: {
opd1 = unbox_vector(argument(4), vbox_type, elem_bt, num_elem);
if (opd1 == nullptr) {
if (C->print_intrinsics()) {
tty->print_cr(" ** unbox failed v1=%s",
NodeClassNames[argument(4)->Opcode()]);
}
return false;
}
break;
}
default: fatal("unsupported arity: %d", n);
}
Node* operation = nullptr;
if (opc == Op_CallLeafVector) {
assert(UseVectorStubs, "sanity");
operation = gen_call_to_svml(opr->get_con(), elem_bt, num_elem, opd1, opd2);
if (operation == nullptr) {
if (C->print_intrinsics()) {
tty->print_cr(" ** svml call failed for %s_%s_%d",
(elem_bt == T_FLOAT)?"float":"double",
VectorSupport::svmlname[opr->get_con() - VectorSupport::VECTOR_OP_SVML_START],
num_elem * type2aelembytes(elem_bt));
}
return false;
}
} else {
const TypeVect* vt = TypeVect::make(elem_bt, num_elem);
switch (n) {
case 1:
case 2: {
operation = gvn().transform(VectorNode::make(sopc, opd1, opd2, vt));
break;
}
case 3: {
operation = gvn().transform(VectorNode::make(sopc, opd1, opd2, opd3, vt));
break;
}
default: fatal("unsupported arity: %d", n);
}
}
// Wrap it up in VectorBox to keep object type information.
Node* vbox = box_vector(operation, vbox_type, elem_bt, num_elem);
set_result(vbox);
C->set_max_vector_size(MAX2(C->max_vector_size(), (uint)(num_elem * type2aelembytes(elem_bt))));
return true;
}
// <Sh extends VectorShuffle<E>, E>
// Sh ShuffleIota(Class<?> E, Class<?> ShuffleClass, Vector.Species<E> s, int length,
// int start, int step, int wrap, ShuffleIotaOperation<Sh, E> defaultImpl)
bool LibraryCallKit::inline_vector_shuffle_iota() {
const TypeInstPtr* shuffle_klass = gvn().type(argument(1))->isa_instptr();
const TypeInt* vlen = gvn().type(argument(3))->isa_int();
const TypeInt* start_val = gvn().type(argument(4))->isa_int();
const TypeInt* step_val = gvn().type(argument(5))->isa_int();
const TypeInt* wrap = gvn().type(argument(6))->isa_int();
Node* start = argument(4);
Node* step = argument(5);
if (shuffle_klass == nullptr || vlen == nullptr || start_val == nullptr || step_val == nullptr || wrap == nullptr) {
return false; // dead code
}
if (!vlen->is_con() || !is_power_of_2(vlen->get_con()) ||
shuffle_klass->const_oop() == nullptr || !wrap->is_con()) {
return false; // not enough info for intrinsification
}
if (!is_klass_initialized(shuffle_klass)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** klass argument not initialized");
}
return false;
}
int do_wrap = wrap->get_con();
int num_elem = vlen->get_con();
BasicType elem_bt = T_BYTE;
if (!arch_supports_vector(VectorNode::replicate_opcode(elem_bt), num_elem, elem_bt, VecMaskNotUsed)) {
return false;
}
if (!arch_supports_vector(Op_AddVB, num_elem, elem_bt, VecMaskNotUsed)) {
return false;
}
if (!arch_supports_vector(Op_AndV, num_elem, elem_bt, VecMaskNotUsed)) {
return false;
}
if (!arch_supports_vector(Op_VectorLoadConst, num_elem, elem_bt, VecMaskNotUsed)) {
return false;
}
if (!arch_supports_vector(Op_VectorBlend, num_elem, elem_bt, VecMaskUseLoad)) {
return false;
}
if (!arch_supports_vector(Op_VectorMaskCmp, num_elem, elem_bt, VecMaskUseStore)) {
return false;
}
const Type * type_bt = Type::get_const_basic_type(elem_bt);
const TypeVect * vt = TypeVect::make(type_bt, num_elem);
Node* res = gvn().transform(new VectorLoadConstNode(gvn().makecon(TypeInt::ZERO), vt));
if(!step_val->is_con() || !is_power_of_2(step_val->get_con())) {
Node* bcast_step = gvn().transform(VectorNode::scalar2vector(step, num_elem, type_bt));
res = gvn().transform(VectorNode::make(Op_MulI, res, bcast_step, num_elem, elem_bt));
} else if (step_val->get_con() > 1) {
Node* cnt = gvn().makecon(TypeInt::make(log2i_exact(step_val->get_con())));
Node* shift_cnt = vector_shift_count(cnt, Op_LShiftI, elem_bt, num_elem);
res = gvn().transform(VectorNode::make(Op_LShiftVB, res, shift_cnt, vt));
}
if (!start_val->is_con() || start_val->get_con() != 0) {
Node* bcast_start = gvn().transform(VectorNode::scalar2vector(start, num_elem, type_bt));
res = gvn().transform(VectorNode::make(Op_AddI, res, bcast_start, num_elem, elem_bt));
}
Node * mod_val = gvn().makecon(TypeInt::make(num_elem-1));
Node * bcast_mod = gvn().transform(VectorNode::scalar2vector(mod_val, num_elem, type_bt));
if(do_wrap) {
// Wrap the indices greater than lane count.
res = gvn().transform(VectorNode::make(Op_AndI, res, bcast_mod, num_elem, elem_bt));
} else {
ConINode* pred_node = (ConINode*)gvn().makecon(TypeInt::make(1));
Node * lane_cnt = gvn().makecon(TypeInt::make(num_elem));
Node * bcast_lane_cnt = gvn().transform(VectorNode::scalar2vector(lane_cnt, num_elem, type_bt));
Node* mask = gvn().transform(new VectorMaskCmpNode(BoolTest::ge, bcast_lane_cnt, res, pred_node, vt));
// Make the indices greater than lane count as -ve values. This matches the java side implementation.
res = gvn().transform(VectorNode::make(Op_AndI, res, bcast_mod, num_elem, elem_bt));
Node * biased_val = gvn().transform(VectorNode::make(Op_SubI, res, bcast_lane_cnt, num_elem, elem_bt));
res = gvn().transform(new VectorBlendNode(biased_val, res, mask));
}
ciKlass* sbox_klass = shuffle_klass->const_oop()->as_instance()->java_lang_Class_klass();
const TypeInstPtr* shuffle_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, sbox_klass);
// Wrap it up in VectorBox to keep object type information.
res = box_vector(res, shuffle_box_type, elem_bt, num_elem);
set_result(res);
C->set_max_vector_size(MAX2(C->max_vector_size(), (uint)(num_elem * type2aelembytes(elem_bt))));
return true;
}
// <E, M>
// int maskReductionCoerced(int oper, Class<? extends M> maskClass, Class<?> elemClass,
// int length, M m, VectorMaskOp<M> defaultImpl)
bool LibraryCallKit::inline_vector_mask_operation() {
const TypeInt* oper = gvn().type(argument(0))->isa_int();
const TypeInstPtr* mask_klass = gvn().type(argument(1))->isa_instptr();
const TypeInstPtr* elem_klass = gvn().type(argument(2))->isa_instptr();
const TypeInt* vlen = gvn().type(argument(3))->isa_int();
Node* mask = argument(4);
if (mask_klass == nullptr || elem_klass == nullptr || mask->is_top() || vlen == nullptr) {
return false; // dead code
}
if (!is_klass_initialized(mask_klass)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** klass argument not initialized");
}
return false;
}
int num_elem = vlen->get_con();
ciType* elem_type = elem_klass->const_oop()->as_instance()->java_mirror_type();
BasicType elem_bt = elem_type->basic_type();
if (!arch_supports_vector(Op_LoadVector, num_elem, T_BOOLEAN, VecMaskNotUsed)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not supported: arity=1 op=cast#%d/3 vlen2=%d etype2=%s",
Op_LoadVector, num_elem, type2name(T_BOOLEAN));
}
return false; // not supported
}
int mopc = VectorSupport::vop2ideal(oper->get_con(), elem_bt);
if (!arch_supports_vector(mopc, num_elem, elem_bt, VecMaskNotUsed)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not supported: arity=1 op=cast#%d/3 vlen2=%d etype2=%s",
mopc, num_elem, type2name(elem_bt));
}
return false; // not supported
}
const Type* elem_ty = Type::get_const_basic_type(elem_bt);
ciKlass* mbox_klass = mask_klass->const_oop()->as_instance()->java_lang_Class_klass();
const TypeInstPtr* mask_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, mbox_klass);
Node* mask_vec = unbox_vector(mask, mask_box_type, elem_bt, num_elem, true);
Node* store_mask = gvn().transform(VectorStoreMaskNode::make(gvn(), mask_vec, elem_bt, num_elem));
Node* maskoper = gvn().transform(VectorMaskOpNode::make(store_mask, TypeInt::INT, mopc));
set_result(maskoper);
C->set_max_vector_size(MAX2(C->max_vector_size(), (uint)(num_elem * type2aelembytes(elem_bt))));
return true;
}
// <VM ,Sh extends VectorShuffle<E>, E>
// VM shuffleToVector(Class<VM> VecClass, Class<?>E , Class<?> ShuffleClass, Sh s, int length,
// ShuffleToVectorOperation<VM,Sh,E> defaultImpl)
bool LibraryCallKit::inline_vector_shuffle_to_vector() {
const TypeInstPtr* vector_klass = gvn().type(argument(0))->isa_instptr();
const TypeInstPtr* elem_klass = gvn().type(argument(1))->isa_instptr();
const TypeInstPtr* shuffle_klass = gvn().type(argument(2))->isa_instptr();
Node* shuffle = argument(3);
const TypeInt* vlen = gvn().type(argument(4))->isa_int();
if (vector_klass == nullptr || elem_klass == nullptr || shuffle_klass == nullptr || shuffle->is_top() || vlen == nullptr) {
return false; // dead code
}
if (!vlen->is_con() || vector_klass->const_oop() == nullptr || shuffle_klass->const_oop() == nullptr) {
return false; // not enough info for intrinsification
}
if (!is_klass_initialized(shuffle_klass) || !is_klass_initialized(vector_klass) ) {
if (C->print_intrinsics()) {
tty->print_cr(" ** klass argument not initialized");
}
return false;
}
int num_elem = vlen->get_con();
ciType* elem_type = elem_klass->const_oop()->as_instance()->java_mirror_type();
BasicType elem_bt = elem_type->basic_type();
if (num_elem < 4) {
return false;
}
int cast_vopc = VectorCastNode::opcode(T_BYTE); // from shuffle of type T_BYTE
// Make sure that cast is implemented to particular type/size combination.
if (!arch_supports_vector(cast_vopc, num_elem, elem_bt, VecMaskNotUsed)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not supported: arity=1 op=cast#%d/3 vlen2=%d etype2=%s",
cast_vopc, num_elem, type2name(elem_bt));
}
return false;
}
ciKlass* sbox_klass = shuffle_klass->const_oop()->as_instance()->java_lang_Class_klass();
const TypeInstPtr* shuffle_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, sbox_klass);
// Unbox shuffle with true flag to indicate its load shuffle to vector
// shuffle is a byte array
Node* shuffle_vec = unbox_vector(shuffle, shuffle_box_type, T_BYTE, num_elem, true);
// cast byte to target element type
shuffle_vec = gvn().transform(VectorCastNode::make(cast_vopc, shuffle_vec, elem_bt, num_elem));
ciKlass* vbox_klass = vector_klass->const_oop()->as_instance()->java_lang_Class_klass();
const TypeInstPtr* vec_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, vbox_klass);
// Box vector
Node* res = box_vector(shuffle_vec, vec_box_type, elem_bt, num_elem);
set_result(res);
C->set_max_vector_size(MAX2(C->max_vector_size(), (uint)(num_elem * type2aelembytes(elem_bt))));
return true;
}
// <V extends Vector<?,?>>
// V broadcastCoerced(Class<?> vectorClass, Class<?> elementType, int vlen,
// long bits,
// LongFunction<V> defaultImpl)
bool LibraryCallKit::inline_vector_broadcast_coerced() {
const TypeInstPtr* vector_klass = gvn().type(argument(0))->isa_instptr();
const TypeInstPtr* elem_klass = gvn().type(argument(1))->isa_instptr();
const TypeInt* vlen = gvn().type(argument(2))->isa_int();
if (vector_klass == nullptr || elem_klass == nullptr || vlen == nullptr ||
vector_klass->const_oop() == nullptr || elem_klass->const_oop() == nullptr || !vlen->is_con()) {
if (C->print_intrinsics()) {
tty->print_cr(" ** missing constant: vclass=%s etype=%s vlen=%s",
NodeClassNames[argument(0)->Opcode()],
NodeClassNames[argument(1)->Opcode()],
NodeClassNames[argument(2)->Opcode()]);
}
return false; // not enough info for intrinsification
}
if (!is_klass_initialized(vector_klass)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** klass argument not initialized");
}
return false;
}
ciType* elem_type = elem_klass->const_oop()->as_instance()->java_mirror_type();
if (!elem_type->is_primitive_type()) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not a primitive bt=%d", elem_type->basic_type());
}
return false; // should be primitive type
}
BasicType elem_bt = elem_type->basic_type();
int num_elem = vlen->get_con();
ciKlass* vbox_klass = vector_klass->const_oop()->as_instance()->java_lang_Class_klass();
const TypeInstPtr* vbox_type = TypeInstPtr::make_exact(TypePtr::NotNull, vbox_klass);
// TODO When mask usage is supported, VecMaskNotUsed needs to be VecMaskUseLoad.
if (!arch_supports_vector(VectorNode::replicate_opcode(elem_bt), num_elem, elem_bt,
(is_vector_mask(vbox_klass) ? VecMaskUseStore : VecMaskNotUsed), true /*has_scalar_args*/)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not supported: arity=0 op=broadcast vlen=%d etype=%s ismask=%d",
num_elem, type2name(elem_bt),
is_vector_mask(vbox_klass) ? 1 : 0);
}
return false; // not supported
}
Node* bits = argument(3); // long
Node* elem = nullptr;
switch (elem_bt) {
case T_BOOLEAN: // fall-through
case T_BYTE: // fall-through
case T_SHORT: // fall-through
case T_CHAR: // fall-through
case T_INT: {
elem = gvn().transform(new ConvL2INode(bits));
break;
}
case T_DOUBLE: {
elem = gvn().transform(new MoveL2DNode(bits));
break;
}
case T_FLOAT: {
bits = gvn().transform(new ConvL2INode(bits));
elem = gvn().transform(new MoveI2FNode(bits));
break;
}
case T_LONG: {
elem = bits; // no conversion needed
break;
}
default: fatal("%s", type2name(elem_bt));
}
Node* broadcast = VectorNode::scalar2vector(elem, num_elem, Type::get_const_basic_type(elem_bt));
broadcast = gvn().transform(broadcast);
Node* box = box_vector(broadcast, vbox_type, elem_bt, num_elem);
set_result(box);
C->set_max_vector_size(MAX2(C->max_vector_size(), (uint)(num_elem * type2aelembytes(elem_bt))));
return true;
}
static bool elem_consistent_with_arr(BasicType elem_bt, const TypeAryPtr* arr_type) {
assert(arr_type != nullptr, "unexpected");
BasicType arr_elem_bt = arr_type->elem()->array_element_basic_type();
if (elem_bt == arr_elem_bt) {
return true;
} else if (elem_bt == T_SHORT && arr_elem_bt == T_CHAR) {
// Load/store of short vector from/to char[] is supported
return true;
} else if (elem_bt == T_BYTE && arr_elem_bt == T_BOOLEAN) {
// Load/store of byte vector from/to boolean[] is supported
return true;
} else {
return false;
}
}
// <C, V extends Vector<?,?>>
// V load(Class<?> vectorClass, Class<?> elementType, int vlen,
// Object base, long offset,
// /* Vector.Mask<E,S> m*/
// Object container, int index,
// LoadOperation<C, VM> defaultImpl) {
//
// <C, V extends Vector<?,?>>
// void store(Class<?> vectorClass, Class<?> elementType, int vlen,
// Object base, long offset,
// V v, /*Vector.Mask<E,S> m*/
// Object container, int index,
// StoreVectorOperation<C, V> defaultImpl) {
bool LibraryCallKit::inline_vector_mem_operation(bool is_store) {
const TypeInstPtr* vector_klass = gvn().type(argument(0))->isa_instptr();
const TypeInstPtr* elem_klass = gvn().type(argument(1))->isa_instptr();
const TypeInt* vlen = gvn().type(argument(2))->isa_int();
if (vector_klass == nullptr || elem_klass == nullptr || vlen == nullptr ||
vector_klass->const_oop() == nullptr || elem_klass->const_oop() == nullptr || !vlen->is_con()) {
if (C->print_intrinsics()) {
tty->print_cr(" ** missing constant: vclass=%s etype=%s vlen=%s",
NodeClassNames[argument(0)->Opcode()],
NodeClassNames[argument(1)->Opcode()],
NodeClassNames[argument(2)->Opcode()]);
}
return false; // not enough info for intrinsification
}
if (!is_klass_initialized(vector_klass)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** klass argument not initialized");
}
return false;
}
ciType* elem_type = elem_klass->const_oop()->as_instance()->java_mirror_type();
if (!elem_type->is_primitive_type()) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not a primitive bt=%d", elem_type->basic_type());
}
return false; // should be primitive type
}
BasicType elem_bt = elem_type->basic_type();
int num_elem = vlen->get_con();
// TODO When mask usage is supported, VecMaskNotUsed needs to be VecMaskUseLoad.
if (!arch_supports_vector(is_store ? Op_StoreVector : Op_LoadVector, num_elem, elem_bt, VecMaskNotUsed)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not supported: arity=%d op=%s vlen=%d etype=%s ismask=no",
is_store, is_store ? "store" : "load",
num_elem, type2name(elem_bt));
}
return false; // not supported
}
ciKlass* vbox_klass = vector_klass->const_oop()->as_instance()->java_lang_Class_klass();
bool is_mask = is_vector_mask(vbox_klass);
Node* base = argument(3);
Node* offset = ConvL2X(argument(4));
// Save state and restore on bailout
uint old_sp = sp();
SafePointNode* old_map = clone_map();
Node* addr = make_unsafe_address(base, offset, (is_mask ? T_BOOLEAN : elem_bt), true);
// Can base be null? Otherwise, always on-heap access.
bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(gvn().type(base));
const TypePtr *addr_type = gvn().type(addr)->isa_ptr();
const TypeAryPtr* arr_type = addr_type->isa_aryptr();
// Now handle special case where load/store happens from/to byte array but element type is not byte.
bool using_byte_array = arr_type != nullptr && arr_type->elem()->array_element_basic_type() == T_BYTE && elem_bt != T_BYTE;
// Handle loading masks.
// If there is no consistency between array and vector element types, it must be special byte array case or loading masks
if (arr_type != nullptr && !using_byte_array && !is_mask && !elem_consistent_with_arr(elem_bt, arr_type)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not supported: arity=%d op=%s vlen=%d etype=%s atype=%s ismask=no",
is_store, is_store ? "store" : "load",
num_elem, type2name(elem_bt), type2name(arr_type->elem()->array_element_basic_type()));
}
set_map(old_map);
set_sp(old_sp);
return false;
}
// Since we are using byte array, we need to double check that the byte operations are supported by backend.
if (using_byte_array) {
int byte_num_elem = num_elem * type2aelembytes(elem_bt);
if (!arch_supports_vector(is_store ? Op_StoreVector : Op_LoadVector, byte_num_elem, T_BYTE, VecMaskNotUsed)
|| !arch_supports_vector(Op_VectorReinterpret, byte_num_elem, T_BYTE, VecMaskNotUsed)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not supported: arity=%d op=%s vlen=%d*8 etype=%s/8 ismask=no",
is_store, is_store ? "store" : "load",
byte_num_elem, type2name(elem_bt));
}
set_map(old_map);
set_sp(old_sp);
return false; // not supported
}
}
if (is_mask) {
if (!arch_supports_vector(Op_LoadVector, num_elem, T_BOOLEAN, VecMaskNotUsed)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not supported: arity=%d op=%s/mask vlen=%d etype=bit ismask=no",
is_store, is_store ? "store" : "load",
num_elem);
}
set_map(old_map);
set_sp(old_sp);
return false; // not supported
}
if (!is_store) {
if (!arch_supports_vector(Op_LoadVector, num_elem, elem_bt, VecMaskUseLoad)) {
set_map(old_map);
set_sp(old_sp);
return false; // not supported
}
} else {
if (!arch_supports_vector(Op_StoreVector, num_elem, elem_bt, VecMaskUseStore)) {
set_map(old_map);
set_sp(old_sp);
return false; // not supported
}
}
}
const TypeInstPtr* vbox_type = TypeInstPtr::make_exact(TypePtr::NotNull, vbox_klass);
if (can_access_non_heap) {
insert_mem_bar(Op_MemBarCPUOrder);
}
if (is_store) {
Node* val = unbox_vector(argument(6), vbox_type, elem_bt, num_elem);
if (val == nullptr) {
set_map(old_map);
set_sp(old_sp);
return false; // operand unboxing failed
}
set_all_memory(reset_memory());
// In case the store needs to happen to byte array, reinterpret the incoming vector to byte vector.
int store_num_elem = num_elem;
if (using_byte_array) {
store_num_elem = num_elem * type2aelembytes(elem_bt);
const TypeVect* to_vect_type = TypeVect::make(T_BYTE, store_num_elem);
val = gvn().transform(new VectorReinterpretNode(val, val->bottom_type()->is_vect(), to_vect_type));
}
Node* vstore = gvn().transform(StoreVectorNode::make(0, control(), memory(addr), addr, addr_type, val, store_num_elem));
set_memory(vstore, addr_type);
} else {
// When using byte array, we need to load as byte then reinterpret the value. Otherwise, do a simple vector load.
Node* vload = nullptr;
if (using_byte_array) {
int load_num_elem = num_elem * type2aelembytes(elem_bt);
vload = gvn().transform(LoadVectorNode::make(0, control(), memory(addr), addr, addr_type, load_num_elem, T_BYTE));
const TypeVect* to_vect_type = TypeVect::make(elem_bt, num_elem);
vload = gvn().transform(new VectorReinterpretNode(vload, vload->bottom_type()->is_vect(), to_vect_type));
} else {
// Special handle for masks
if (is_mask) {
vload = gvn().transform(LoadVectorNode::make(0, control(), memory(addr), addr, addr_type, num_elem, T_BOOLEAN));
const TypeVect* to_vect_type = TypeVect::make(elem_bt, num_elem);
vload = gvn().transform(new VectorLoadMaskNode(vload, to_vect_type));
} else {
vload = gvn().transform(LoadVectorNode::make(0, control(), memory(addr), addr, addr_type, num_elem, elem_bt));
}
}
Node* box = box_vector(vload, vbox_type, elem_bt, num_elem);
set_result(box);
}
destruct_map_clone(old_map);
if (can_access_non_heap) {
insert_mem_bar(Op_MemBarCPUOrder);
}
C->set_max_vector_size(MAX2(C->max_vector_size(), (uint)(num_elem * type2aelembytes(elem_bt))));
return true;
}
// <C, V extends Vector<?>, W extends IntVector, E, S extends VectorSpecies<E>>
// void loadWithMap(Class<?> vectorClass, Class<E> E, int length, Class<?> vectorIndexClass,
// Object base, long offset, // Unsafe addressing
// W index_vector,
// C container, int index, int[] indexMap, int indexM, S s, // Arguments for default implementation
// LoadVectorOperationWithMap<C, V, E, S> defaultImpl)
//
// <C, V extends Vector<?>, W extends IntVector>
// void storeWithMap(Class<?> vectorClass, Class<?> elementType, int length, Class<?> vectorIndexClass,
// Object base, long offset, // Unsafe addressing
// W index_vector, V v,
// C container, int index, int[] indexMap, int indexM, // Arguments for default implementation
// StoreVectorOperationWithMap<C, V> defaultImpl) {
//
bool LibraryCallKit::inline_vector_gather_scatter(bool is_scatter) {
const TypeInstPtr* vector_klass = gvn().type(argument(0))->isa_instptr();
const TypeInstPtr* elem_klass = gvn().type(argument(1))->isa_instptr();
const TypeInt* vlen = gvn().type(argument(2))->isa_int();
const TypeInstPtr* vector_idx_klass = gvn().type(argument(3))->isa_instptr();
if (vector_klass == nullptr || elem_klass == nullptr || vector_idx_klass == nullptr || vlen == nullptr ||
vector_klass->const_oop() == nullptr || elem_klass->const_oop() == nullptr || vector_idx_klass->const_oop() == nullptr || !vlen->is_con()) {
if (C->print_intrinsics()) {
tty->print_cr(" ** missing constant: vclass=%s etype=%s vlen=%s viclass=%s",
NodeClassNames[argument(0)->Opcode()],
NodeClassNames[argument(1)->Opcode()],
NodeClassNames[argument(2)->Opcode()],
NodeClassNames[argument(3)->Opcode()]);
}
return false; // not enough info for intrinsification
}
if (!is_klass_initialized(vector_klass) || !is_klass_initialized(vector_idx_klass)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** klass argument not initialized");
}
return false;
}
ciType* elem_type = elem_klass->const_oop()->as_instance()->java_mirror_type();
if (!elem_type->is_primitive_type()) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not a primitive bt=%d", elem_type->basic_type());
}
return false; // should be primitive type
}
BasicType elem_bt = elem_type->basic_type();
int num_elem = vlen->get_con();
if (!arch_supports_vector(is_scatter ? Op_StoreVectorScatter : Op_LoadVectorGather, num_elem, elem_bt, VecMaskNotUsed)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not supported: arity=%d op=%s vlen=%d etype=%s ismask=no",
is_scatter, is_scatter ? "scatter" : "gather",
num_elem, type2name(elem_bt));
}
return false; // not supported
}
// Check that the vector holding indices is supported by architecture
if (!arch_supports_vector(Op_LoadVector, num_elem, T_INT, VecMaskNotUsed)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not supported: arity=%d op=%s/loadindex vlen=%d etype=int ismask=no",
is_scatter, is_scatter ? "scatter" : "gather",
num_elem);
}
return false; // not supported
}
Node* base = argument(4);
Node* offset = ConvL2X(argument(5));
// Save state and restore on bailout
uint old_sp = sp();
SafePointNode* old_map = clone_map();
Node* addr = make_unsafe_address(base, offset, elem_bt, true);
const TypePtr *addr_type = gvn().type(addr)->isa_ptr();
const TypeAryPtr* arr_type = addr_type->isa_aryptr();
// The array must be consistent with vector type
if (arr_type == nullptr || (arr_type != nullptr && !elem_consistent_with_arr(elem_bt, arr_type))) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not supported: arity=%d op=%s vlen=%d etype=%s atype=%s ismask=no",
is_scatter, is_scatter ? "scatter" : "gather",
num_elem, type2name(elem_bt), type2name(arr_type->elem()->array_element_basic_type()));
}
set_map(old_map);
set_sp(old_sp);
return false;
}
ciKlass* vbox_klass = vector_klass->const_oop()->as_instance()->java_lang_Class_klass();
const TypeInstPtr* vbox_type = TypeInstPtr::make_exact(TypePtr::NotNull, vbox_klass);
ciKlass* vbox_idx_klass = vector_idx_klass->const_oop()->as_instance()->java_lang_Class_klass();
if (vbox_idx_klass == nullptr) {
set_map(old_map);
set_sp(old_sp);
return false;
}
const TypeInstPtr* vbox_idx_type = TypeInstPtr::make_exact(TypePtr::NotNull, vbox_idx_klass);
Node* index_vect = unbox_vector(argument(7), vbox_idx_type, T_INT, num_elem);
if (index_vect == nullptr) {
set_map(old_map);
set_sp(old_sp);
return false;
}
const TypeVect* vector_type = TypeVect::make(elem_bt, num_elem);
if (is_scatter) {
Node* val = unbox_vector(argument(8), vbox_type, elem_bt, num_elem);
if (val == nullptr) {
set_map(old_map);
set_sp(old_sp);
return false; // operand unboxing failed
}
set_all_memory(reset_memory());
Node* vstore = gvn().transform(new StoreVectorScatterNode(control(), memory(addr), addr, addr_type, val, index_vect));