-
Notifications
You must be signed in to change notification settings - Fork 5.4k
/
parse2.cpp
2827 lines (2540 loc) · 88.7 KB
/
parse2.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) 1998, 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 "jvm_io.h"
#include "ci/ciMethodData.hpp"
#include "classfile/vmSymbols.hpp"
#include "compiler/compileLog.hpp"
#include "interpreter/linkResolver.hpp"
#include "memory/resourceArea.hpp"
#include "memory/universe.hpp"
#include "oops/oop.inline.hpp"
#include "opto/addnode.hpp"
#include "opto/castnode.hpp"
#include "opto/convertnode.hpp"
#include "opto/divnode.hpp"
#include "opto/idealGraphPrinter.hpp"
#include "opto/matcher.hpp"
#include "opto/memnode.hpp"
#include "opto/mulnode.hpp"
#include "opto/opaquenode.hpp"
#include "opto/parse.hpp"
#include "opto/runtime.hpp"
#include "runtime/deoptimization.hpp"
#include "runtime/sharedRuntime.hpp"
#ifndef PRODUCT
extern int explicit_null_checks_inserted,
explicit_null_checks_elided;
#endif
//---------------------------------array_load----------------------------------
void Parse::array_load(BasicType bt) {
const Type* elemtype = Type::TOP;
bool big_val = bt == T_DOUBLE || bt == T_LONG;
Node* adr = array_addressing(bt, 0, elemtype);
if (stopped()) return; // guaranteed null or range check
pop(); // index (already used)
Node* array = pop(); // the array itself
if (elemtype == TypeInt::BOOL) {
bt = T_BOOLEAN;
}
const TypeAryPtr* adr_type = TypeAryPtr::get_array_body_type(bt);
Node* ld = access_load_at(array, adr, adr_type, elemtype, bt,
IN_HEAP | IS_ARRAY | C2_CONTROL_DEPENDENT_LOAD);
if (big_val) {
push_pair(ld);
} else {
push(ld);
}
}
//--------------------------------array_store----------------------------------
void Parse::array_store(BasicType bt) {
const Type* elemtype = Type::TOP;
bool big_val = bt == T_DOUBLE || bt == T_LONG;
Node* adr = array_addressing(bt, big_val ? 2 : 1, elemtype);
if (stopped()) return; // guaranteed null or range check
if (bt == T_OBJECT) {
array_store_check();
if (stopped()) {
return;
}
}
Node* val; // Oop to store
if (big_val) {
val = pop_pair();
} else {
val = pop();
}
pop(); // index (already used)
Node* array = pop(); // the array itself
if (elemtype == TypeInt::BOOL) {
bt = T_BOOLEAN;
}
const TypeAryPtr* adr_type = TypeAryPtr::get_array_body_type(bt);
access_store_at(array, adr, adr_type, val, elemtype, bt, MO_UNORDERED | IN_HEAP | IS_ARRAY);
}
//------------------------------array_addressing-------------------------------
// Pull array and index from the stack. Compute pointer-to-element.
Node* Parse::array_addressing(BasicType type, int vals, const Type*& elemtype) {
Node *idx = peek(0+vals); // Get from stack without popping
Node *ary = peek(1+vals); // in case of exception
// Null check the array base, with correct stack contents
ary = null_check(ary, T_ARRAY);
// Compile-time detect of null-exception?
if (stopped()) return top();
const TypeAryPtr* arytype = _gvn.type(ary)->is_aryptr();
const TypeInt* sizetype = arytype->size();
elemtype = arytype->elem();
if (UseUniqueSubclasses) {
const Type* el = elemtype->make_ptr();
if (el && el->isa_instptr()) {
const TypeInstPtr* toop = el->is_instptr();
if (toop->klass()->as_instance_klass()->unique_concrete_subklass()) {
// If we load from "AbstractClass[]" we must see "ConcreteSubClass".
const Type* subklass = Type::get_const_type(toop->klass());
elemtype = subklass->join_speculative(el);
}
}
}
// Check for big class initializers with all constant offsets
// feeding into a known-size array.
const TypeInt* idxtype = _gvn.type(idx)->is_int();
// See if the highest idx value is less than the lowest array bound,
// and if the idx value cannot be negative:
bool need_range_check = true;
if (idxtype->_hi < sizetype->_lo && idxtype->_lo >= 0) {
need_range_check = false;
if (C->log() != NULL) C->log()->elem("observe that='!need_range_check'");
}
ciKlass * arytype_klass = arytype->klass();
if ((arytype_klass != NULL) && (!arytype_klass->is_loaded())) {
// Only fails for some -Xcomp runs
// The class is unloaded. We have to run this bytecode in the interpreter.
uncommon_trap(Deoptimization::Reason_unloaded,
Deoptimization::Action_reinterpret,
arytype->klass(), "!loaded array");
return top();
}
// Do the range check
if (GenerateRangeChecks && need_range_check) {
Node* tst;
if (sizetype->_hi <= 0) {
// The greatest array bound is negative, so we can conclude that we're
// compiling unreachable code, but the unsigned compare trick used below
// only works with non-negative lengths. Instead, hack "tst" to be zero so
// the uncommon_trap path will always be taken.
tst = _gvn.intcon(0);
} else {
// Range is constant in array-oop, so we can use the original state of mem
Node* len = load_array_length(ary);
// Test length vs index (standard trick using unsigned compare)
Node* chk = _gvn.transform( new CmpUNode(idx, len) );
BoolTest::mask btest = BoolTest::lt;
tst = _gvn.transform( new BoolNode(chk, btest) );
}
RangeCheckNode* rc = new RangeCheckNode(control(), tst, PROB_MAX, COUNT_UNKNOWN);
_gvn.set_type(rc, rc->Value(&_gvn));
if (!tst->is_Con()) {
record_for_igvn(rc);
}
set_control(_gvn.transform(new IfTrueNode(rc)));
// Branch to failure if out of bounds
{
PreserveJVMState pjvms(this);
set_control(_gvn.transform(new IfFalseNode(rc)));
if (C->allow_range_check_smearing()) {
// Do not use builtin_throw, since range checks are sometimes
// made more stringent by an optimistic transformation.
// This creates "tentative" range checks at this point,
// which are not guaranteed to throw exceptions.
// See IfNode::Ideal, is_range_check, adjust_check.
uncommon_trap(Deoptimization::Reason_range_check,
Deoptimization::Action_make_not_entrant,
NULL, "range_check");
} else {
// If we have already recompiled with the range-check-widening
// heroic optimization turned off, then we must really be throwing
// range check exceptions.
builtin_throw(Deoptimization::Reason_range_check, idx);
}
}
}
// Check for always knowing you are throwing a range-check exception
if (stopped()) return top();
// Make array address computation control dependent to prevent it
// from floating above the range check during loop optimizations.
Node* ptr = array_element_address(ary, idx, type, sizetype, control());
assert(ptr != top(), "top should go hand-in-hand with stopped");
return ptr;
}
// returns IfNode
IfNode* Parse::jump_if_fork_int(Node* a, Node* b, BoolTest::mask mask, float prob, float cnt) {
Node *cmp = _gvn.transform(new CmpINode(a, b)); // two cases: shiftcount > 32 and shiftcount <= 32
Node *tst = _gvn.transform(new BoolNode(cmp, mask));
IfNode *iff = create_and_map_if(control(), tst, prob, cnt);
return iff;
}
// return Region node
Node* Parse::jump_if_join(Node* iffalse, Node* iftrue) {
Node *region = new RegionNode(3); // 2 results
record_for_igvn(region);
region->init_req(1, iffalse);
region->init_req(2, iftrue );
_gvn.set_type(region, Type::CONTROL);
region = _gvn.transform(region);
set_control (region);
return region;
}
// sentinel value for the target bci to mark never taken branches
// (according to profiling)
static const int never_reached = INT_MAX;
//------------------------------helper for tableswitch-------------------------
void Parse::jump_if_true_fork(IfNode *iff, int dest_bci_if_true, bool unc) {
// True branch, use existing map info
{ PreserveJVMState pjvms(this);
Node *iftrue = _gvn.transform( new IfTrueNode (iff) );
set_control( iftrue );
if (unc) {
repush_if_args();
uncommon_trap(Deoptimization::Reason_unstable_if,
Deoptimization::Action_reinterpret,
NULL,
"taken always");
} else {
assert(dest_bci_if_true != never_reached, "inconsistent dest");
merge_new_path(dest_bci_if_true);
}
}
// False branch
Node *iffalse = _gvn.transform( new IfFalseNode(iff) );
set_control( iffalse );
}
void Parse::jump_if_false_fork(IfNode *iff, int dest_bci_if_true, bool unc) {
// True branch, use existing map info
{ PreserveJVMState pjvms(this);
Node *iffalse = _gvn.transform( new IfFalseNode (iff) );
set_control( iffalse );
if (unc) {
repush_if_args();
uncommon_trap(Deoptimization::Reason_unstable_if,
Deoptimization::Action_reinterpret,
NULL,
"taken never");
} else {
assert(dest_bci_if_true != never_reached, "inconsistent dest");
merge_new_path(dest_bci_if_true);
}
}
// False branch
Node *iftrue = _gvn.transform( new IfTrueNode(iff) );
set_control( iftrue );
}
void Parse::jump_if_always_fork(int dest_bci, bool unc) {
// False branch, use existing map and control()
if (unc) {
repush_if_args();
uncommon_trap(Deoptimization::Reason_unstable_if,
Deoptimization::Action_reinterpret,
NULL,
"taken never");
} else {
assert(dest_bci != never_reached, "inconsistent dest");
merge_new_path(dest_bci);
}
}
extern "C" {
static int jint_cmp(const void *i, const void *j) {
int a = *(jint *)i;
int b = *(jint *)j;
return a > b ? 1 : a < b ? -1 : 0;
}
}
class SwitchRange : public StackObj {
// a range of integers coupled with a bci destination
jint _lo; // inclusive lower limit
jint _hi; // inclusive upper limit
int _dest;
float _cnt; // how many times this range was hit according to profiling
public:
jint lo() const { return _lo; }
jint hi() const { return _hi; }
int dest() const { return _dest; }
bool is_singleton() const { return _lo == _hi; }
float cnt() const { return _cnt; }
void setRange(jint lo, jint hi, int dest, float cnt) {
assert(lo <= hi, "must be a non-empty range");
_lo = lo, _hi = hi; _dest = dest; _cnt = cnt;
assert(_cnt >= 0, "");
}
bool adjoinRange(jint lo, jint hi, int dest, float cnt, bool trim_ranges) {
assert(lo <= hi, "must be a non-empty range");
if (lo == _hi+1) {
// see merge_ranges() comment below
if (trim_ranges) {
if (cnt == 0) {
if (_cnt != 0) {
return false;
}
if (dest != _dest) {
_dest = never_reached;
}
} else {
if (_cnt == 0) {
return false;
}
if (dest != _dest) {
return false;
}
}
} else {
if (dest != _dest) {
return false;
}
}
_hi = hi;
_cnt += cnt;
return true;
}
return false;
}
void set (jint value, int dest, float cnt) {
setRange(value, value, dest, cnt);
}
bool adjoin(jint value, int dest, float cnt, bool trim_ranges) {
return adjoinRange(value, value, dest, cnt, trim_ranges);
}
bool adjoin(SwitchRange& other) {
return adjoinRange(other._lo, other._hi, other._dest, other._cnt, false);
}
void print() {
if (is_singleton())
tty->print(" {%d}=>%d (cnt=%f)", lo(), dest(), cnt());
else if (lo() == min_jint)
tty->print(" {..%d}=>%d (cnt=%f)", hi(), dest(), cnt());
else if (hi() == max_jint)
tty->print(" {%d..}=>%d (cnt=%f)", lo(), dest(), cnt());
else
tty->print(" {%d..%d}=>%d (cnt=%f)", lo(), hi(), dest(), cnt());
}
};
// We try to minimize the number of ranges and the size of the taken
// ones using profiling data. When ranges are created,
// SwitchRange::adjoinRange() only allows 2 adjoining ranges to merge
// if both were never hit or both were hit to build longer unreached
// ranges. Here, we now merge adjoining ranges with the same
// destination and finally set destination of unreached ranges to the
// special value never_reached because it can help minimize the number
// of tests that are necessary.
//
// For instance:
// [0, 1] to target1 sometimes taken
// [1, 2] to target1 never taken
// [2, 3] to target2 never taken
// would lead to:
// [0, 1] to target1 sometimes taken
// [1, 3] never taken
//
// (first 2 ranges to target1 are not merged)
static void merge_ranges(SwitchRange* ranges, int& rp) {
if (rp == 0) {
return;
}
int shift = 0;
for (int j = 0; j < rp; j++) {
SwitchRange& r1 = ranges[j-shift];
SwitchRange& r2 = ranges[j+1];
if (r1.adjoin(r2)) {
shift++;
} else if (shift > 0) {
ranges[j+1-shift] = r2;
}
}
rp -= shift;
for (int j = 0; j <= rp; j++) {
SwitchRange& r = ranges[j];
if (r.cnt() == 0 && r.dest() != never_reached) {
r.setRange(r.lo(), r.hi(), never_reached, r.cnt());
}
}
}
//-------------------------------do_tableswitch--------------------------------
void Parse::do_tableswitch() {
// Get information about tableswitch
int default_dest = iter().get_dest_table(0);
int lo_index = iter().get_int_table(1);
int hi_index = iter().get_int_table(2);
int len = hi_index - lo_index + 1;
if (len < 1) {
// If this is a backward branch, add safepoint
maybe_add_safepoint(default_dest);
pop(); // the effect of the instruction execution on the operand stack
merge(default_dest);
return;
}
ciMethodData* methodData = method()->method_data();
ciMultiBranchData* profile = NULL;
if (methodData->is_mature() && UseSwitchProfiling) {
ciProfileData* data = methodData->bci_to_data(bci());
if (data != NULL && data->is_MultiBranchData()) {
profile = (ciMultiBranchData*)data;
}
}
bool trim_ranges = !C->too_many_traps(method(), bci(), Deoptimization::Reason_unstable_if);
// generate decision tree, using trichotomy when possible
int rnum = len+2;
bool makes_backward_branch = false;
SwitchRange* ranges = NEW_RESOURCE_ARRAY(SwitchRange, rnum);
int rp = -1;
if (lo_index != min_jint) {
uint cnt = 1;
if (profile != NULL) {
cnt = profile->default_count() / (hi_index != max_jint ? 2 : 1);
}
ranges[++rp].setRange(min_jint, lo_index-1, default_dest, cnt);
}
for (int j = 0; j < len; j++) {
jint match_int = lo_index+j;
int dest = iter().get_dest_table(j+3);
makes_backward_branch |= (dest <= bci());
uint cnt = 1;
if (profile != NULL) {
cnt = profile->count_at(j);
}
if (rp < 0 || !ranges[rp].adjoin(match_int, dest, cnt, trim_ranges)) {
ranges[++rp].set(match_int, dest, cnt);
}
}
jint highest = lo_index+(len-1);
assert(ranges[rp].hi() == highest, "");
if (highest != max_jint) {
uint cnt = 1;
if (profile != NULL) {
cnt = profile->default_count() / (lo_index != min_jint ? 2 : 1);
}
if (!ranges[rp].adjoinRange(highest+1, max_jint, default_dest, cnt, trim_ranges)) {
ranges[++rp].setRange(highest+1, max_jint, default_dest, cnt);
}
}
assert(rp < len+2, "not too many ranges");
if (trim_ranges) {
merge_ranges(ranges, rp);
}
// Safepoint in case if backward branch observed
if (makes_backward_branch) {
add_safepoint();
}
Node* lookup = pop(); // lookup value
jump_switch_ranges(lookup, &ranges[0], &ranges[rp]);
}
//------------------------------do_lookupswitch--------------------------------
void Parse::do_lookupswitch() {
// Get information about lookupswitch
int default_dest = iter().get_dest_table(0);
int len = iter().get_int_table(1);
if (len < 1) { // If this is a backward branch, add safepoint
maybe_add_safepoint(default_dest);
pop(); // the effect of the instruction execution on the operand stack
merge(default_dest);
return;
}
ciMethodData* methodData = method()->method_data();
ciMultiBranchData* profile = NULL;
if (methodData->is_mature() && UseSwitchProfiling) {
ciProfileData* data = methodData->bci_to_data(bci());
if (data != NULL && data->is_MultiBranchData()) {
profile = (ciMultiBranchData*)data;
}
}
bool trim_ranges = !C->too_many_traps(method(), bci(), Deoptimization::Reason_unstable_if);
// generate decision tree, using trichotomy when possible
jint* table = NEW_RESOURCE_ARRAY(jint, len*3);
{
for (int j = 0; j < len; j++) {
table[3*j+0] = iter().get_int_table(2+2*j);
table[3*j+1] = iter().get_dest_table(2+2*j+1);
// Handle overflow when converting from uint to jint
table[3*j+2] = (profile == NULL) ? 1 : MIN2<uint>(max_jint, profile->count_at(j));
}
qsort(table, len, 3*sizeof(table[0]), jint_cmp);
}
float defaults = 0;
jint prev = min_jint;
for (int j = 0; j < len; j++) {
jint match_int = table[3*j+0];
if (match_int != prev) {
defaults += (float)match_int - prev;
}
prev = match_int+1;
}
if (prev != min_jint) {
defaults += (float)max_jint - prev + 1;
}
float default_cnt = 1;
if (profile != NULL) {
default_cnt = profile->default_count()/defaults;
}
int rnum = len*2+1;
bool makes_backward_branch = false;
SwitchRange* ranges = NEW_RESOURCE_ARRAY(SwitchRange, rnum);
int rp = -1;
for (int j = 0; j < len; j++) {
jint match_int = table[3*j+0];
int dest = table[3*j+1];
int cnt = table[3*j+2];
int next_lo = rp < 0 ? min_jint : ranges[rp].hi()+1;
makes_backward_branch |= (dest <= bci());
float c = default_cnt * ((float)match_int - next_lo);
if (match_int != next_lo && (rp < 0 || !ranges[rp].adjoinRange(next_lo, match_int-1, default_dest, c, trim_ranges))) {
assert(default_dest != never_reached, "sentinel value for dead destinations");
ranges[++rp].setRange(next_lo, match_int-1, default_dest, c);
}
if (rp < 0 || !ranges[rp].adjoin(match_int, dest, cnt, trim_ranges)) {
assert(dest != never_reached, "sentinel value for dead destinations");
ranges[++rp].set(match_int, dest, cnt);
}
}
jint highest = table[3*(len-1)];
assert(ranges[rp].hi() == highest, "");
if (highest != max_jint &&
!ranges[rp].adjoinRange(highest+1, max_jint, default_dest, default_cnt * ((float)max_jint - highest), trim_ranges)) {
ranges[++rp].setRange(highest+1, max_jint, default_dest, default_cnt * ((float)max_jint - highest));
}
assert(rp < rnum, "not too many ranges");
if (trim_ranges) {
merge_ranges(ranges, rp);
}
// Safepoint in case backward branch observed
if (makes_backward_branch) {
add_safepoint();
}
Node *lookup = pop(); // lookup value
jump_switch_ranges(lookup, &ranges[0], &ranges[rp]);
}
static float if_prob(float taken_cnt, float total_cnt) {
assert(taken_cnt <= total_cnt, "");
if (total_cnt == 0) {
return PROB_FAIR;
}
float p = taken_cnt / total_cnt;
return clamp(p, PROB_MIN, PROB_MAX);
}
static float if_cnt(float cnt) {
if (cnt == 0) {
return COUNT_UNKNOWN;
}
return cnt;
}
static float sum_of_cnts(SwitchRange *lo, SwitchRange *hi) {
float total_cnt = 0;
for (SwitchRange* sr = lo; sr <= hi; sr++) {
total_cnt += sr->cnt();
}
return total_cnt;
}
class SwitchRanges : public ResourceObj {
public:
SwitchRange* _lo;
SwitchRange* _hi;
SwitchRange* _mid;
float _cost;
enum {
Start,
LeftDone,
RightDone,
Done
} _state;
SwitchRanges(SwitchRange *lo, SwitchRange *hi)
: _lo(lo), _hi(hi), _mid(NULL),
_cost(0), _state(Start) {
}
SwitchRanges()
: _lo(NULL), _hi(NULL), _mid(NULL),
_cost(0), _state(Start) {}
};
// Estimate cost of performing a binary search on lo..hi
static float compute_tree_cost(SwitchRange *lo, SwitchRange *hi, float total_cnt) {
GrowableArray<SwitchRanges> tree;
SwitchRanges root(lo, hi);
tree.push(root);
float cost = 0;
do {
SwitchRanges& r = *tree.adr_at(tree.length()-1);
if (r._hi != r._lo) {
if (r._mid == NULL) {
float r_cnt = sum_of_cnts(r._lo, r._hi);
if (r_cnt == 0) {
tree.pop();
cost = 0;
continue;
}
SwitchRange* mid = NULL;
mid = r._lo;
for (float cnt = 0; ; ) {
assert(mid <= r._hi, "out of bounds");
cnt += mid->cnt();
if (cnt > r_cnt / 2) {
break;
}
mid++;
}
assert(mid <= r._hi, "out of bounds");
r._mid = mid;
r._cost = r_cnt / total_cnt;
}
r._cost += cost;
if (r._state < SwitchRanges::LeftDone && r._mid > r._lo) {
cost = 0;
r._state = SwitchRanges::LeftDone;
tree.push(SwitchRanges(r._lo, r._mid-1));
} else if (r._state < SwitchRanges::RightDone) {
cost = 0;
r._state = SwitchRanges::RightDone;
tree.push(SwitchRanges(r._mid == r._lo ? r._mid+1 : r._mid, r._hi));
} else {
tree.pop();
cost = r._cost;
}
} else {
tree.pop();
cost = r._cost;
}
} while (tree.length() > 0);
return cost;
}
// It sometimes pays off to test most common ranges before the binary search
void Parse::linear_search_switch_ranges(Node* key_val, SwitchRange*& lo, SwitchRange*& hi) {
uint nr = hi - lo + 1;
float total_cnt = sum_of_cnts(lo, hi);
float min = compute_tree_cost(lo, hi, total_cnt);
float extra = 1;
float sub = 0;
SwitchRange* array1 = lo;
SwitchRange* array2 = NEW_RESOURCE_ARRAY(SwitchRange, nr);
SwitchRange* ranges = NULL;
while (nr >= 2) {
assert(lo == array1 || lo == array2, "one the 2 already allocated arrays");
ranges = (lo == array1) ? array2 : array1;
// Find highest frequency range
SwitchRange* candidate = lo;
for (SwitchRange* sr = lo+1; sr <= hi; sr++) {
if (sr->cnt() > candidate->cnt()) {
candidate = sr;
}
}
SwitchRange most_freq = *candidate;
if (most_freq.cnt() == 0) {
break;
}
// Copy remaining ranges into another array
int shift = 0;
for (uint i = 0; i < nr; i++) {
SwitchRange* sr = &lo[i];
if (sr != candidate) {
ranges[i-shift] = *sr;
} else {
shift++;
if (i > 0 && i < nr-1) {
SwitchRange prev = lo[i-1];
prev.setRange(prev.lo(), sr->hi(), prev.dest(), prev.cnt());
if (prev.adjoin(lo[i+1])) {
shift++;
i++;
}
ranges[i-shift] = prev;
}
}
}
nr -= shift;
// Evaluate cost of testing the most common range and performing a
// binary search on the other ranges
float cost = extra + compute_tree_cost(&ranges[0], &ranges[nr-1], total_cnt);
if (cost >= min) {
break;
}
// swap arrays
lo = &ranges[0];
hi = &ranges[nr-1];
// It pays off: emit the test for the most common range
assert(most_freq.cnt() > 0, "must be taken");
Node* val = _gvn.transform(new SubINode(key_val, _gvn.intcon(most_freq.lo())));
Node* cmp = _gvn.transform(new CmpUNode(val, _gvn.intcon(most_freq.hi() - most_freq.lo())));
Node* tst = _gvn.transform(new BoolNode(cmp, BoolTest::le));
IfNode* iff = create_and_map_if(control(), tst, if_prob(most_freq.cnt(), total_cnt), if_cnt(most_freq.cnt()));
jump_if_true_fork(iff, most_freq.dest(), false);
sub += most_freq.cnt() / total_cnt;
extra += 1 - sub;
min = cost;
}
}
//----------------------------create_jump_tables-------------------------------
bool Parse::create_jump_tables(Node* key_val, SwitchRange* lo, SwitchRange* hi) {
// Are jumptables enabled
if (!UseJumpTables) return false;
// Are jumptables supported
if (!Matcher::has_match_rule(Op_Jump)) return false;
bool trim_ranges = !C->too_many_traps(method(), bci(), Deoptimization::Reason_unstable_if);
// Decide if a guard is needed to lop off big ranges at either (or
// both) end(s) of the input set. We'll call this the default target
// even though we can't be sure that it is the true "default".
bool needs_guard = false;
int default_dest;
int64_t total_outlier_size = 0;
int64_t hi_size = ((int64_t)hi->hi()) - ((int64_t)hi->lo()) + 1;
int64_t lo_size = ((int64_t)lo->hi()) - ((int64_t)lo->lo()) + 1;
if (lo->dest() == hi->dest()) {
total_outlier_size = hi_size + lo_size;
default_dest = lo->dest();
} else if (lo_size > hi_size) {
total_outlier_size = lo_size;
default_dest = lo->dest();
} else {
total_outlier_size = hi_size;
default_dest = hi->dest();
}
float total = sum_of_cnts(lo, hi);
float cost = compute_tree_cost(lo, hi, total);
// If a guard test will eliminate very sparse end ranges, then
// it is worth the cost of an extra jump.
float trimmed_cnt = 0;
if (total_outlier_size > (MaxJumpTableSparseness * 4)) {
needs_guard = true;
if (default_dest == lo->dest()) {
trimmed_cnt += lo->cnt();
lo++;
}
if (default_dest == hi->dest()) {
trimmed_cnt += hi->cnt();
hi--;
}
}
// Find the total number of cases and ranges
int64_t num_cases = ((int64_t)hi->hi()) - ((int64_t)lo->lo()) + 1;
int num_range = hi - lo + 1;
// Don't create table if: too large, too small, or too sparse.
if (num_cases > MaxJumpTableSize)
return false;
if (UseSwitchProfiling) {
// MinJumpTableSize is set so with a well balanced binary tree,
// when the number of ranges is MinJumpTableSize, it's cheaper to
// go through a JumpNode that a tree of IfNodes. Average cost of a
// tree of IfNodes with MinJumpTableSize is
// log2f(MinJumpTableSize) comparisons. So if the cost computed
// from profile data is less than log2f(MinJumpTableSize) then
// going with the binary search is cheaper.
if (cost < log2f(MinJumpTableSize)) {
return false;
}
} else {
if (num_cases < MinJumpTableSize)
return false;
}
if (num_cases > (MaxJumpTableSparseness * num_range))
return false;
// Normalize table lookups to zero
int lowval = lo->lo();
key_val = _gvn.transform( new SubINode(key_val, _gvn.intcon(lowval)) );
// Generate a guard to protect against input keyvals that aren't
// in the switch domain.
if (needs_guard) {
Node* size = _gvn.intcon(num_cases);
Node* cmp = _gvn.transform(new CmpUNode(key_val, size));
Node* tst = _gvn.transform(new BoolNode(cmp, BoolTest::ge));
IfNode* iff = create_and_map_if(control(), tst, if_prob(trimmed_cnt, total), if_cnt(trimmed_cnt));
jump_if_true_fork(iff, default_dest, trim_ranges && trimmed_cnt == 0);
total -= trimmed_cnt;
}
// Create an ideal node JumpTable that has projections
// of all possible ranges for a switch statement
// The key_val input must be converted to a pointer offset and scaled.
// Compare Parse::array_addressing above.
// Clean the 32-bit int into a real 64-bit offset.
// Otherwise, the jint value 0 might turn into an offset of 0x0800000000.
// Make I2L conversion control dependent to prevent it from
// floating above the range check during loop optimizations.
// Do not use a narrow int type here to prevent the data path from dying
// while the control path is not removed. This can happen if the type of key_val
// is later known to be out of bounds of [0, num_cases] and therefore a narrow cast
// would be replaced by TOP while C2 is not able to fold the corresponding range checks.
// Set _carry_dependency for the cast to avoid being removed by IGVN.
#ifdef _LP64
key_val = C->constrained_convI2L(&_gvn, key_val, TypeInt::INT, control(), true /* carry_dependency */);
#endif
// Shift the value by wordsize so we have an index into the table, rather
// than a switch value
Node *shiftWord = _gvn.MakeConX(wordSize);
key_val = _gvn.transform( new MulXNode( key_val, shiftWord));
// Create the JumpNode
Arena* arena = C->comp_arena();
float* probs = (float*)arena->Amalloc(sizeof(float)*num_cases);
int i = 0;
if (total == 0) {
for (SwitchRange* r = lo; r <= hi; r++) {
for (int64_t j = r->lo(); j <= r->hi(); j++, i++) {
probs[i] = 1.0F / num_cases;
}
}
} else {
for (SwitchRange* r = lo; r <= hi; r++) {
float prob = r->cnt()/total;
for (int64_t j = r->lo(); j <= r->hi(); j++, i++) {
probs[i] = prob / (r->hi() - r->lo() + 1);
}
}
}
ciMethodData* methodData = method()->method_data();
ciMultiBranchData* profile = NULL;
if (methodData->is_mature()) {
ciProfileData* data = methodData->bci_to_data(bci());
if (data != NULL && data->is_MultiBranchData()) {
profile = (ciMultiBranchData*)data;
}
}
Node* jtn = _gvn.transform(new JumpNode(control(), key_val, num_cases, probs, profile == NULL ? COUNT_UNKNOWN : total));
// These are the switch destinations hanging off the jumpnode
i = 0;
for (SwitchRange* r = lo; r <= hi; r++) {
for (int64_t j = r->lo(); j <= r->hi(); j++, i++) {
Node* input = _gvn.transform(new JumpProjNode(jtn, i, r->dest(), (int)(j - lowval)));
{
PreserveJVMState pjvms(this);
set_control(input);
jump_if_always_fork(r->dest(), trim_ranges && r->cnt() == 0);
}
}
}
assert(i == num_cases, "miscount of cases");
stop_and_kill_map(); // no more uses for this JVMS
return true;
}
//----------------------------jump_switch_ranges-------------------------------
void Parse::jump_switch_ranges(Node* key_val, SwitchRange *lo, SwitchRange *hi, int switch_depth) {
Block* switch_block = block();
bool trim_ranges = !C->too_many_traps(method(), bci(), Deoptimization::Reason_unstable_if);
if (switch_depth == 0) {
// Do special processing for the top-level call.
assert(lo->lo() == min_jint, "initial range must exhaust Type::INT");
assert(hi->hi() == max_jint, "initial range must exhaust Type::INT");
// Decrement pred-numbers for the unique set of nodes.
#ifdef ASSERT
if (!trim_ranges) {
// Ensure that the block's successors are a (duplicate-free) set.
int successors_counted = 0; // block occurrences in [hi..lo]
int unique_successors = switch_block->num_successors();
for (int i = 0; i < unique_successors; i++) {
Block* target = switch_block->successor_at(i);
// Check that the set of successors is the same in both places.
int successors_found = 0;
for (SwitchRange* p = lo; p <= hi; p++) {
if (p->dest() == target->start()) successors_found++;
}
assert(successors_found > 0, "successor must be known");
successors_counted += successors_found;
}
assert(successors_counted == (hi-lo)+1, "no unexpected successors");
}
#endif
// Maybe prune the inputs, based on the type of key_val.
jint min_val = min_jint;
jint max_val = max_jint;
const TypeInt* ti = key_val->bottom_type()->isa_int();
if (ti != NULL) {
min_val = ti->_lo;
max_val = ti->_hi;
assert(min_val <= max_val, "invalid int type");
}
while (lo->hi() < min_val) {
lo++;
}
if (lo->lo() < min_val) {
lo->setRange(min_val, lo->hi(), lo->dest(), lo->cnt());
}
while (hi->lo() > max_val) {
hi--;
}
if (hi->hi() > max_val) {
hi->setRange(hi->lo(), max_val, hi->dest(), hi->cnt());
}
linear_search_switch_ranges(key_val, lo, hi);
}
#ifndef PRODUCT
if (switch_depth == 0) {
_max_switch_depth = 0;
_est_switch_depth = log2i_graceful((hi - lo + 1) - 1) + 1;
}
#endif
assert(lo <= hi, "must be a non-empty set of ranges");
if (lo == hi) {
jump_if_always_fork(lo->dest(), trim_ranges && lo->cnt() == 0);
} else {
assert(lo->hi() == (lo+1)->lo()-1, "contiguous ranges");
assert(hi->lo() == (hi-1)->hi()+1, "contiguous ranges");
if (create_jump_tables(key_val, lo, hi)) return;