summaryrefslogtreecommitdiff
path: root/src/intel/compiler/brw_fs_scoreboard.cpp
blob: 8702e12e5167921f45aa003b3c12c3ff3c5cc4f1 (plain)
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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
/*
 * Copyright © 2019 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */

/** @file brw_fs_scoreboard.cpp
 *
 * Gfx12+ hardware lacks the register scoreboard logic that used to guarantee
 * data coherency between register reads and writes in previous generations.
 * This lowering pass runs after register allocation in order to make up for
 * it.
 *
 * It works by performing global dataflow analysis in order to determine the
 * set of potential dependencies of every instruction in the shader, and then
 * inserts any required SWSB annotations and additional SYNC instructions in
 * order to guarantee data coherency.
 *
 * WARNING - Access of the following (rarely used) ARF registers is not
 *           tracked here, and require the RegDist SWSB annotation to be set
 *           to 1 by the generator in order to avoid data races:
 *
 *  - sp stack pointer
 *  - sr0 state register
 *  - cr0 control register
 *  - ip instruction pointer
 *  - tm0 timestamp register
 *  - dbg0 debug register
 *  - acc2-9 special accumulator registers on TGL
 *  - mme0-7 math macro extended accumulator registers
 *
 * The following ARF registers don't need to be tracked here because data
 * coherency is still provided transparently by the hardware:
 *
 *  - f0-1 flag registers
 *  - n0 notification register
 *  - tdr0 thread dependency register
 */

#include "brw_fs.h"
#include "brw_cfg.h"

using namespace brw;

namespace {
   /**
    * In-order instruction accounting.
    * @{
    */

   /**
    * Return the RegDist pipeline the hardware will synchronize with if no
    * pipeline information is provided in the SWSB annotation of an
    * instruction (e.g. when TGL_PIPE_NONE is specified in tgl_swsb).
    */
   tgl_pipe
   inferred_sync_pipe(const struct intel_device_info *devinfo, const fs_inst *inst)
   {
      if (devinfo->verx10 >= 125) {
         bool has_int_src = false, has_long_src = false;

         if (is_send(inst))
            return TGL_PIPE_NONE;

         for (unsigned i = 0; i < inst->sources; i++) {
            if (inst->src[i].file != BAD_FILE &&
                !inst->is_control_source(i)) {
               const brw_reg_type t = inst->src[i].type;
               has_int_src |= !brw_reg_type_is_floating_point(t);
               has_long_src |= type_sz(t) >= 8;
            }
         }

         return has_long_src ? TGL_PIPE_LONG :
                has_int_src ? TGL_PIPE_INT :
                TGL_PIPE_FLOAT;

      } else {
         return TGL_PIPE_FLOAT;
      }
   }

   /**
    * Return the RegDist pipeline that will execute an instruction, or
    * TGL_PIPE_NONE if the instruction is out-of-order and doesn't use the
    * RegDist synchronization mechanism.
    */
   tgl_pipe
   inferred_exec_pipe(const struct intel_device_info *devinfo, const fs_inst *inst)
   {
      const brw_reg_type t = get_exec_type(inst);
      const bool is_dword_multiply = !brw_reg_type_is_floating_point(t) &&
         ((inst->opcode == BRW_OPCODE_MUL &&
           MIN2(type_sz(inst->src[0].type), type_sz(inst->src[1].type)) >= 4) ||
          (inst->opcode == BRW_OPCODE_MAD &&
           MIN2(type_sz(inst->src[1].type), type_sz(inst->src[2].type)) >= 4));

      if (is_unordered(inst))
         return TGL_PIPE_NONE;
      else if (devinfo->verx10 < 125)
         return TGL_PIPE_FLOAT;
      else if (inst->opcode == SHADER_OPCODE_MOV_INDIRECT &&
               type_sz(t) >= 8)
         return TGL_PIPE_INT;
      else if (inst->opcode == SHADER_OPCODE_BROADCAST &&
               !devinfo->has_64bit_float && type_sz(t) >= 8)
         return TGL_PIPE_INT;
      else if (type_sz(inst->dst.type) >= 8 || type_sz(t) >= 8 ||
               is_dword_multiply)
         return TGL_PIPE_LONG;
      else if (brw_reg_type_is_floating_point(inst->dst.type))
         return TGL_PIPE_FLOAT;
      else
         return TGL_PIPE_INT;
   }

   /**
    * Index of the \p p pipeline counter in the ordered_address vector defined
    * below.
    */
#define IDX(p) (p >= TGL_PIPE_FLOAT ? unsigned(p - TGL_PIPE_FLOAT) :    \
                (abort(), ~0u))

   /**
    * Number of in-order hardware instructions for pipeline index \p contained
    * in this IR instruction.  This determines the increment applied to the
    * RegDist counter calculated for any ordered dependency that crosses this
    * instruction.
    */
   unsigned
   ordered_unit(const struct intel_device_info *devinfo, const fs_inst *inst,
                unsigned p)
   {
      switch (inst->opcode) {
      case BRW_OPCODE_SYNC:
      case BRW_OPCODE_DO:
      case SHADER_OPCODE_UNDEF:
      case SHADER_OPCODE_HALT_TARGET:
      case FS_OPCODE_SCHEDULING_FENCE:
         return 0;
      default:
         /* Note that the following is inaccurate for virtual instructions
          * that expand to more in-order instructions than assumed here, but
          * that can only lead to suboptimal execution ordering, data
          * coherency won't be impacted.  Providing exact RegDist counts for
          * each virtual instruction would allow better ALU performance, but
          * it would require keeping this switch statement in perfect sync
          * with the generator in order to avoid data corruption.  Lesson is
          * (again) don't use virtual instructions if you want optimal
          * scheduling.
          */
         if (!is_unordered(inst) && (p == IDX(inferred_exec_pipe(devinfo, inst)) ||
                                     p == IDX(TGL_PIPE_ALL)))
            return 1;
         else
            return 0;
      }
   }

   /**
    * Type for an instruction counter that increments for in-order
    * instructions only, arbitrarily denoted 'jp' throughout this lowering
    * pass in order to distinguish it from the regular instruction counter.
    * This is represented as a vector with an independent counter for each
    * asynchronous ALU pipeline in the EU.
    */
   struct ordered_address {
      /**
       * Construct the ordered address of a dependency known to execute on a
       * single specified pipeline \p p (unless TGL_PIPE_NONE or TGL_PIPE_ALL
       * is provided), in which case the vector counter will be initialized
       * with all components equal to INT_MIN (always satisfied) except for
       * component IDX(p).
       */
      ordered_address(tgl_pipe p = TGL_PIPE_NONE, int jp0 = INT_MIN) {
         for (unsigned q = 0; q < IDX(TGL_PIPE_ALL); q++)
            jp[q] = (p == TGL_PIPE_NONE || (IDX(p) != q && p != TGL_PIPE_ALL) ?
                     INT_MIN : jp0);
      }

      int jp[IDX(TGL_PIPE_ALL)];

      friend bool
      operator==(const ordered_address &jp0, const ordered_address &jp1)
      {
         for (unsigned p = 0; p < IDX(TGL_PIPE_ALL); p++) {
            if (jp0.jp[p] != jp1.jp[p])
               return false;
         }

         return true;
      }
   };

   /**
    * Return true if the specified ordered address is trivially satisfied for
    * all pipelines except potentially for the specified pipeline \p p.
    */
   bool
   is_single_pipe(const ordered_address &jp, tgl_pipe p)
   {
      for (unsigned q = 0; q < IDX(TGL_PIPE_ALL); q++) {
         if ((p == TGL_PIPE_NONE || IDX(p) != q) && jp.jp[q] > INT_MIN)
            return false;
      }

      return true;
   }

   /**
    * Return the number of instructions in the program.
    */
   unsigned
   num_instructions(const backend_shader *shader)
   {
      return shader->cfg->blocks[shader->cfg->num_blocks - 1]->end_ip + 1;
   }

   /**
    * Calculate the local ordered_address instruction counter at every
    * instruction of the shader for subsequent constant-time look-up.
    */
   ordered_address *
   ordered_inst_addresses(const fs_visitor *shader)
   {
      ordered_address *jps = new ordered_address[num_instructions(shader)];
      ordered_address jp(TGL_PIPE_ALL, 0);
      unsigned ip = 0;

      foreach_block_and_inst(block, fs_inst, inst, shader->cfg) {
         jps[ip] = jp;
         for (unsigned p = 0; p < IDX(TGL_PIPE_ALL); p++)
            jp.jp[p] += ordered_unit(shader->devinfo, inst, p);
         ip++;
      }

      return jps;
   }

   /**
    * Synchronization mode required for data manipulated by in-order
    * instructions.
    *
    * Similar to tgl_sbid_mode, but without SET mode.  Defined as a separate
    * enum for additional type safety.  The hardware doesn't provide control
    * over the synchronization mode for RegDist annotations, this is only used
    * internally in this pass in order to optimize out redundant read
    * dependencies where possible.
    */
   enum tgl_regdist_mode {
      TGL_REGDIST_NULL = 0,
      TGL_REGDIST_SRC = 1,
      TGL_REGDIST_DST = 2
   };

   /**
    * Allow bitwise arithmetic of tgl_regdist_mode enums.
    */
   tgl_regdist_mode
   operator|(tgl_regdist_mode x, tgl_regdist_mode y)
   {
      return tgl_regdist_mode(unsigned(x) | unsigned(y));
   }

   tgl_regdist_mode
   operator&(tgl_regdist_mode x, tgl_regdist_mode y)
   {
      return tgl_regdist_mode(unsigned(x) & unsigned(y));
   }

   tgl_regdist_mode &
   operator|=(tgl_regdist_mode &x, tgl_regdist_mode y)
   {
      return x = x | y;
   }

   tgl_regdist_mode &
   operator&=(tgl_regdist_mode &x, tgl_regdist_mode y)
   {
      return x = x & y;
   }

   /** @} */

   /**
    * Representation of an equivalence relation among the set of unsigned
    * integers.
    *
    * Its initial state is the identity relation '~' such that i ~ j if and
    * only if i == j for every pair of unsigned integers i and j.
    */
   struct equivalence_relation {
      equivalence_relation(unsigned n) : is(new unsigned[n]), n(n)
      {
         for (unsigned i = 0; i < n; i++)
            is[i] = i;
      }

      ~equivalence_relation()
      {
         delete[] is;
      }

      /**
       * Return equivalence class index of the specified element.  Effectively
       * this is the numeric value of an arbitrary representative from the
       * equivalence class.
       *
       * Allows the evaluation of the equivalence relation according to the
       * rule that i ~ j if and only if lookup(i) == lookup(j).
       */
      unsigned
      lookup(unsigned i) const
      {
         if (i < n && is[i] != i)
            return lookup(is[i]);
         else
            return i;
      }

      /**
       * Create an array with the results of the lookup() method for
       * constant-time evaluation.
       */
      unsigned *
      flatten() const
      {
         unsigned *ids = new unsigned[n];

         for (unsigned i = 0; i < n; i++)
            ids[i] = lookup(i);

         return ids;
      }

      /**
       * Mutate the existing equivalence relation minimally by imposing the
       * additional requirement that i ~ j.
       *
       * The algorithm updates the internal representation recursively in
       * order to guarantee transitivity while preserving the previously
       * specified equivalence requirements.
       */
      unsigned
      link(unsigned i, unsigned j)
      {
         const unsigned k = lookup(i);
         assign(i, k);
         assign(j, k);
         return k;
      }

   private:
      equivalence_relation(const equivalence_relation &);

      equivalence_relation &
      operator=(const equivalence_relation &);

      /**
       * Assign the representative of \p from to be equivalent to \p to.
       *
       * At the same time the data structure is partially flattened as much as
       * it's possible without increasing the number of recursive calls.
       */
      void
      assign(unsigned from, unsigned to)
      {
         if (from != to) {
            assert(from < n);

            if (is[from] != from)
               assign(is[from], to);

            is[from] = to;
         }
      }

      unsigned *is;
      unsigned n;
   };

   /**
    * Representation of a data dependency between two instructions in the
    * program.
    * @{
    */
   struct dependency {
      /**
       * No dependency information.
       */
      dependency() : ordered(TGL_REGDIST_NULL), jp(),
                     unordered(TGL_SBID_NULL), id(0),
                     exec_all(false) {}

      /**
       * Construct a dependency on the in-order instruction with the provided
       * ordered_address instruction counter.
       */
      dependency(tgl_regdist_mode mode, const ordered_address &jp,
                 bool exec_all) :
         ordered(mode), jp(jp), unordered(TGL_SBID_NULL), id(0),
         exec_all(exec_all) {}

      /**
       * Construct a dependency on the out-of-order instruction with the
       * specified synchronization token.
       */
      dependency(tgl_sbid_mode mode, unsigned id, bool exec_all) :
         ordered(TGL_REGDIST_NULL), jp(), unordered(mode), id(id),
         exec_all(exec_all) {}

      /**
       * Synchronization mode of in-order dependency, or zero if no in-order
       * dependency is present.
       */
      tgl_regdist_mode ordered;

      /**
       * Instruction counter of in-order dependency.
       *
       * For a dependency part of a different block in the program, this is
       * relative to the specific control flow path taken between the
       * dependency and the current block: It is the ordered_address such that
       * the difference between it and the ordered_address of the first
       * instruction of the current block is exactly the number of in-order
       * instructions across that control flow path.  It is not guaranteed to
       * be equal to the local ordered_address of the generating instruction
       * [as returned by ordered_inst_addresses()], except for block-local
       * dependencies.
       */
      ordered_address jp;

      /**
       * Synchronization mode of unordered dependency, or zero if no unordered
       * dependency is present.
       */
      tgl_sbid_mode unordered;

      /** Synchronization token of out-of-order dependency. */
      unsigned id;

      /**
       * Whether the dependency could be run with execution masking disabled,
       * which might lead to the unwanted execution of the generating
       * instruction in cases where a BB is executed with all channels
       * disabled due to hardware bug Wa_1407528679.
       */
      bool exec_all;

      /**
       * Trivial in-order dependency that's always satisfied.
       *
       * Note that unlike a default-constructed dependency() which is also
       * trivially satisfied, this is considered to provide dependency
       * information and can be used to clear a previously pending dependency
       * via shadow().
       */
      static const dependency done;

      friend bool
      operator==(const dependency &dep0, const dependency &dep1)
      {
         return dep0.ordered == dep1.ordered &&
                dep0.jp == dep1.jp &&
                dep0.unordered == dep1.unordered &&
                dep0.id == dep1.id &&
                dep0.exec_all == dep1.exec_all;
      }

      friend bool
      operator!=(const dependency &dep0, const dependency &dep1)
      {
         return !(dep0 == dep1);
      }
   };

   const dependency dependency::done =
        dependency(TGL_REGDIST_SRC, ordered_address(), false);

   /**
    * Return whether \p dep contains any dependency information.
    */
   bool
   is_valid(const dependency &dep)
   {
      return dep.ordered || dep.unordered;
   }

   /**
    * Combine \p dep0 and \p dep1 into a single dependency object that is only
    * satisfied when both original dependencies are satisfied.  This might
    * involve updating the equivalence relation \p eq in order to make sure
    * that both out-of-order dependencies are assigned the same hardware SBID
    * as synchronization token.
    */
   dependency
   merge(equivalence_relation &eq,
         const dependency &dep0, const dependency &dep1)
   {
      dependency dep;

      if (dep0.ordered || dep1.ordered) {
         dep.ordered = dep0.ordered | dep1.ordered;
         for (unsigned p = 0; p < IDX(TGL_PIPE_ALL); p++)
            dep.jp.jp[p] = MAX2(dep0.jp.jp[p], dep1.jp.jp[p]);
      }

      if (dep0.unordered || dep1.unordered) {
         dep.unordered = dep0.unordered | dep1.unordered;
         dep.id = eq.link(dep0.unordered ? dep0.id : dep1.id,
                          dep1.unordered ? dep1.id : dep0.id);
      }

      dep.exec_all = dep0.exec_all || dep1.exec_all;

      return dep;
   }

   /**
    * Override dependency information of \p dep0 with that of \p dep1.
    */
   dependency
   shadow(const dependency &dep0, const dependency &dep1)
   {
      return is_valid(dep1) ? dep1 : dep0;
   }

   /**
    * Translate dependency information across the program.
    *
    * This returns a dependency on the same instruction translated to the
    * ordered_address space of a different block.  The correct shift for
    * transporting a dependency across an edge of the CFG is the difference
    * between the local ordered_address of the first instruction of the target
    * block and the local ordered_address of the instruction immediately after
    * the end of the origin block.
    */
   dependency
   transport(dependency dep, int delta[IDX(TGL_PIPE_ALL)])
   {
      if (dep.ordered) {
         for (unsigned p = 0; p < IDX(TGL_PIPE_ALL); p++) {
            if (dep.jp.jp[p] > INT_MIN)
               dep.jp.jp[p] += delta[p];
         }
      }

      return dep;
   }

   /**
    * Return simplified dependency removing any synchronization modes not
    * applicable to an instruction reading the same register location.
    */
   dependency
   dependency_for_read(dependency dep)
   {
      dep.ordered &= TGL_REGDIST_DST;
      return dep;
   }

   /**
    * Return simplified dependency removing any synchronization modes not
    * applicable to an instruction \p inst writing the same register location.
    */
   dependency
   dependency_for_write(const fs_inst *inst, dependency dep)
   {
      if (!is_unordered(inst))
         dep.ordered &= TGL_REGDIST_DST;
      return dep;
   }

   /** @} */

   /**
    * Scoreboard representation.  This keeps track of the data dependencies of
    * registers with GRF granularity.
    */
   class scoreboard {
   public:
      /**
       * Look up the most current data dependency for register \p r.
       */
      dependency
      get(const fs_reg &r) const
      {
         if (const dependency *p = const_cast<scoreboard *>(this)->dep(r))
            return *p;
         else
            return dependency();
      }

      /**
       * Specify the most current data dependency for register \p r.
       */
      void
      set(const fs_reg &r, const dependency &d)
      {
         if (dependency *p = dep(r))
            *p = d;
      }

      /**
       * Component-wise merge() of corresponding dependencies from two
       * scoreboard objects.  \sa merge().
       */
      friend scoreboard
      merge(equivalence_relation &eq,
            const scoreboard &sb0, const scoreboard &sb1)
      {
         scoreboard sb;

         for (unsigned i = 0; i < ARRAY_SIZE(sb.grf_deps); i++)
            sb.grf_deps[i] = merge(eq, sb0.grf_deps[i], sb1.grf_deps[i]);

         sb.addr_dep = merge(eq, sb0.addr_dep, sb1.addr_dep);
         sb.accum_dep = merge(eq, sb0.accum_dep, sb1.accum_dep);

         return sb;
      }

      /**
       * Component-wise shadow() of corresponding dependencies from two
       * scoreboard objects.  \sa shadow().
       */
      friend scoreboard
      shadow(const scoreboard &sb0, const scoreboard &sb1)
      {
         scoreboard sb;

         for (unsigned i = 0; i < ARRAY_SIZE(sb.grf_deps); i++)
            sb.grf_deps[i] = shadow(sb0.grf_deps[i], sb1.grf_deps[i]);

         sb.addr_dep = shadow(sb0.addr_dep, sb1.addr_dep);
         sb.accum_dep = shadow(sb0.accum_dep, sb1.accum_dep);

         return sb;
      }

      /**
       * Component-wise transport() of dependencies from a scoreboard
       * object.  \sa transport().
       */
      friend scoreboard
      transport(const scoreboard &sb0, int delta[IDX(TGL_PIPE_ALL)])
      {
         scoreboard sb;

         for (unsigned i = 0; i < ARRAY_SIZE(sb.grf_deps); i++)
            sb.grf_deps[i] = transport(sb0.grf_deps[i], delta);

         sb.addr_dep = transport(sb0.addr_dep, delta);
         sb.accum_dep = transport(sb0.accum_dep, delta);

         return sb;
      }

      friend bool
      operator==(const scoreboard &sb0, const scoreboard &sb1)
      {
         for (unsigned i = 0; i < ARRAY_SIZE(sb0.grf_deps); i++) {
            if (sb0.grf_deps[i] != sb1.grf_deps[i])
               return false;
         }

         if (sb0.addr_dep != sb1.addr_dep)
            return false;

         if (sb0.accum_dep != sb1.accum_dep)
            return false;

         return true;
      }

      friend bool
      operator!=(const scoreboard &sb0, const scoreboard &sb1)
      {
         return !(sb0 == sb1);
      }

   private:
      dependency grf_deps[BRW_MAX_GRF];
      dependency addr_dep;
      dependency accum_dep;

      dependency *
      dep(const fs_reg &r)
      {
         const unsigned reg = (r.file == VGRF ? r.nr + r.offset / REG_SIZE :
                               reg_offset(r) / REG_SIZE);

         return (r.file == VGRF || r.file == FIXED_GRF ? &grf_deps[reg] :
                 r.file == MRF ? &grf_deps[GFX7_MRF_HACK_START + reg] :
                 r.file == ARF && reg >= BRW_ARF_ADDRESS &&
                                  reg < BRW_ARF_ACCUMULATOR ? &addr_dep :
                 r.file == ARF && reg >= BRW_ARF_ACCUMULATOR &&
                                  reg < BRW_ARF_FLAG ? &accum_dep :
                 NULL);
      }
   };

   /**
    * Dependency list handling.
    * @{
    */
   struct dependency_list {
      dependency_list() : deps(NULL), n(0) {}

      ~dependency_list()
      {
         free(deps);
      }

      void
      push_back(const dependency &dep)
      {
         deps = (dependency *)realloc(deps, (n + 1) * sizeof(*deps));
         deps[n++] = dep;
      }

      unsigned
      size() const
      {
         return n;
      }

      const dependency &
      operator[](unsigned i) const
      {
         assert(i < n);
         return deps[i];
      }

      dependency &
      operator[](unsigned i)
      {
         assert(i < n);
         return deps[i];
      }

   private:
      dependency_list(const dependency_list &);
      dependency_list &
      operator=(const dependency_list &);

      dependency *deps;
      unsigned n;
   };

   /**
    * Add dependency \p dep to the list of dependencies of an instruction
    * \p deps.
    */
   void
   add_dependency(const unsigned *ids, dependency_list &deps, dependency dep)
   {
      if (is_valid(dep)) {
         /* Translate the unordered dependency token first in order to keep
          * the list minimally redundant.
          */
         if (dep.unordered)
            dep.id = ids[dep.id];

         /* Try to combine the specified dependency with any existing ones. */
         for (unsigned i = 0; i < deps.size(); i++) {
            /* Don't combine otherwise matching dependencies if there is an
             * exec_all mismatch which would cause a SET dependency to gain an
             * exec_all flag, since that would prevent it from being baked
             * into the instruction we want to allocate an SBID for.
             */
            if (deps[i].exec_all != dep.exec_all &&
                (!deps[i].exec_all || (dep.unordered & TGL_SBID_SET)) &&
                (!dep.exec_all || (deps[i].unordered & TGL_SBID_SET)))
               continue;

            if (dep.ordered && deps[i].ordered) {
               for (unsigned p = 0; p < IDX(TGL_PIPE_ALL); p++)
                  deps[i].jp.jp[p] = MAX2(deps[i].jp.jp[p], dep.jp.jp[p]);

               deps[i].ordered |= dep.ordered;
               deps[i].exec_all |= dep.exec_all;
               dep.ordered = TGL_REGDIST_NULL;
            }

            if (dep.unordered && deps[i].unordered && deps[i].id == dep.id) {
               deps[i].unordered |= dep.unordered;
               deps[i].exec_all |= dep.exec_all;
               dep.unordered = TGL_SBID_NULL;
            }
         }

         /* Add it to the end of the list if necessary. */
         if (is_valid(dep))
            deps.push_back(dep);
      }
   }

   /**
    * Construct a tgl_swsb annotation encoding any ordered dependencies from
    * the dependency list \p deps of an instruction with ordered_address \p
    * jp.  If \p exec_all is false only dependencies known to be executed with
    * channel masking applied will be considered in the calculation.
    */
   tgl_swsb
   ordered_dependency_swsb(const dependency_list &deps,
                           const ordered_address &jp,
                           bool exec_all)
   {
      tgl_pipe p = TGL_PIPE_NONE;
      unsigned min_dist = ~0u;

      for (unsigned i = 0; i < deps.size(); i++) {
         if (deps[i].ordered && exec_all >= deps[i].exec_all) {
            for (unsigned q = 0; q < IDX(TGL_PIPE_ALL); q++) {
               const unsigned dist = jp.jp[q] - int64_t(deps[i].jp.jp[q]);
               const unsigned max_dist = (q == IDX(TGL_PIPE_LONG) ? 14 : 10);
               assert(jp.jp[q] > deps[i].jp.jp[q]);
               if (dist <= max_dist) {
                  p = (p && IDX(p) != q ? TGL_PIPE_ALL :
                       tgl_pipe(TGL_PIPE_FLOAT + q));
                  min_dist = MIN3(min_dist, dist, 7);
               }
            }
         }
      }

      return { p ? min_dist : 0, p };
   }

   /**
    * Return whether the dependency list \p deps of an instruction with
    * ordered_address \p jp has any non-trivial ordered dependencies.  If \p
    * exec_all is false only dependencies known to be executed with channel
    * masking applied will be considered in the calculation.
    */
   bool
   find_ordered_dependency(const dependency_list &deps,
                           const ordered_address &jp,
                           bool exec_all)
   {
      return ordered_dependency_swsb(deps, jp, exec_all).regdist;
   }

   /**
    * Return the full tgl_sbid_mode bitset for the first unordered dependency
    * on the list \p deps that matches the specified tgl_sbid_mode, or zero if
    * no such dependency is present.  If \p exec_all is false only
    * dependencies known to be executed with channel masking applied will be
    * considered in the calculation.
    */
   tgl_sbid_mode
   find_unordered_dependency(const dependency_list &deps,
                             tgl_sbid_mode unordered,
                             bool exec_all)
   {
      if (unordered) {
         for (unsigned i = 0; i < deps.size(); i++) {
            if ((unordered & deps[i].unordered) &&
                exec_all >= deps[i].exec_all)
               return deps[i].unordered;
         }
      }

      return TGL_SBID_NULL;
   }

   /**
    * Return the tgl_sbid_mode bitset of an unordered dependency from the list
    * \p deps that can be represented directly in the SWSB annotation of the
    * instruction without additional SYNC instructions, or zero if no such
    * dependency is present.
    */
   tgl_sbid_mode
   baked_unordered_dependency_mode(const struct intel_device_info *devinfo,
                                   const fs_inst *inst,
                                   const dependency_list &deps,
                                   const ordered_address &jp)
   {
      const bool exec_all = inst->force_writemask_all;
      const bool has_ordered = find_ordered_dependency(deps, jp, exec_all);
      const tgl_pipe ordered_pipe = ordered_dependency_swsb(deps, jp,
                                                            exec_all).pipe;

      if (find_unordered_dependency(deps, TGL_SBID_SET, exec_all))
         return find_unordered_dependency(deps, TGL_SBID_SET, exec_all);
      else if (has_ordered && is_unordered(inst))
         return TGL_SBID_NULL;
      else if (find_unordered_dependency(deps, TGL_SBID_DST, exec_all) &&
               (!has_ordered || ordered_pipe == inferred_sync_pipe(devinfo, inst)))
         return find_unordered_dependency(deps, TGL_SBID_DST, exec_all);
      else if (!has_ordered)
         return find_unordered_dependency(deps, TGL_SBID_SRC, exec_all);
      else
         return TGL_SBID_NULL;
   }

   /**
    * Return whether an ordered dependency from the list \p deps can be
    * represented directly in the SWSB annotation of the instruction without
    * additional SYNC instructions.
    */
   bool
   baked_ordered_dependency_mode(const struct intel_device_info *devinfo,
                                 const fs_inst *inst,
                                 const dependency_list &deps,
                                 const ordered_address &jp)
   {
      const bool exec_all = inst->force_writemask_all;
      const bool has_ordered = find_ordered_dependency(deps, jp, exec_all);
      const tgl_pipe ordered_pipe = ordered_dependency_swsb(deps, jp,
                                                            exec_all).pipe;
      const tgl_sbid_mode unordered_mode =
         baked_unordered_dependency_mode(devinfo, inst, deps, jp);

      if (!has_ordered)
         return false;
      else if (!unordered_mode)
         return true;
      else
         return ordered_pipe == inferred_sync_pipe(devinfo, inst) &&
                unordered_mode == (is_unordered(inst) ? TGL_SBID_SET :
                                   TGL_SBID_DST);
   }

   /** @} */

   /**
    * Shader instruction dependency calculation.
    * @{
    */

   /**
    * Update scoreboard object \p sb to account for the execution of
    * instruction \p inst.
    */
   void
   update_inst_scoreboard(const fs_visitor *shader, const ordered_address *jps,
                          const fs_inst *inst, unsigned ip, scoreboard &sb)
   {
      const bool exec_all = inst->force_writemask_all;
      const struct intel_device_info *devinfo = shader->devinfo;
      const tgl_pipe p = inferred_exec_pipe(devinfo, inst);
      const ordered_address jp = p ? ordered_address(p, jps[ip].jp[IDX(p)]) :
                                     ordered_address();

      /* Track any source registers that may be fetched asynchronously by this
       * instruction, otherwise clear the dependency in order to avoid
       * subsequent redundant synchronization.
       */
      for (unsigned i = 0; i < inst->sources; i++) {
         const dependency rd_dep =
            (inst->is_payload(i) ||
             inst->is_math()) ? dependency(TGL_SBID_SRC, ip, exec_all) :
            ordered_unit(devinfo, inst, IDX(TGL_PIPE_ALL)) ?
               dependency(TGL_REGDIST_SRC, jp, exec_all) :
            dependency::done;

         for (unsigned j = 0; j < regs_read(inst, i); j++)
            sb.set(byte_offset(inst->src[i], REG_SIZE * j), rd_dep);
      }

      if (inst->reads_accumulator_implicitly())
         sb.set(brw_acc_reg(8), dependency(TGL_REGDIST_SRC, jp, exec_all));

      if (is_send(inst) && inst->base_mrf != -1) {
         const dependency rd_dep = dependency(TGL_SBID_SRC, ip, exec_all);

         for (unsigned j = 0; j < inst->mlen; j++)
            sb.set(brw_uvec_mrf(8, inst->base_mrf + j, 0), rd_dep);
      }

      /* Track any destination registers of this instruction. */
      const dependency wr_dep =
         is_unordered(inst) ? dependency(TGL_SBID_DST, ip, exec_all) :
         ordered_unit(devinfo, inst, IDX(TGL_PIPE_ALL)) ?
            dependency(TGL_REGDIST_DST, jp, exec_all) :
         dependency();

      if (inst->writes_accumulator_implicitly(devinfo))
         sb.set(brw_acc_reg(8), wr_dep);

      if (is_valid(wr_dep) && inst->dst.file != BAD_FILE &&
          !inst->dst.is_null()) {
         for (unsigned j = 0; j < regs_written(inst); j++)
            sb.set(byte_offset(inst->dst, REG_SIZE * j), wr_dep);
      }
   }

   /**
    * Calculate scoreboard objects locally that represent any pending (and
    * unconditionally resolved) dependencies at the end of each block of the
    * program.
    */
   scoreboard *
   gather_block_scoreboards(const fs_visitor *shader,
                            const ordered_address *jps)
   {
      scoreboard *sbs = new scoreboard[shader->cfg->num_blocks];
      unsigned ip = 0;

      foreach_block_and_inst(block, fs_inst, inst, shader->cfg)
         update_inst_scoreboard(shader, jps, inst, ip++, sbs[block->num]);

      return sbs;
   }

   /**
    * Propagate data dependencies globally through the control flow graph
    * until a fixed point is reached.
    *
    * Calculates the set of dependencies potentially pending at the beginning
    * of each block, and returns it as an array of scoreboard objects.
    */
   scoreboard *
   propagate_block_scoreboards(const fs_visitor *shader,
                               const ordered_address *jps,
                               equivalence_relation &eq)
   {
      const scoreboard *delta_sbs = gather_block_scoreboards(shader, jps);
      scoreboard *in_sbs = new scoreboard[shader->cfg->num_blocks];
      scoreboard *out_sbs = new scoreboard[shader->cfg->num_blocks];

      for (bool progress = true; progress;) {
         progress = false;

         foreach_block(block, shader->cfg) {
            const scoreboard sb = shadow(in_sbs[block->num],
                                         delta_sbs[block->num]);

            if (sb != out_sbs[block->num]) {
               foreach_list_typed(bblock_link, child_link, link,
                                  &block->children) {
                  scoreboard &in_sb = in_sbs[child_link->block->num];
                  int delta[IDX(TGL_PIPE_ALL)];

                  for (unsigned p = 0; p < IDX(TGL_PIPE_ALL); p++)
                     delta[p] = jps[child_link->block->start_ip].jp[p]
                        - jps[block->end_ip].jp[p]
                        - ordered_unit(shader->devinfo,
                                       static_cast<const fs_inst *>(block->end()), p);

                  in_sb = merge(eq, in_sb, transport(sb, delta));
               }

               out_sbs[block->num] = sb;
               progress = true;
            }
         }
      }

      delete[] delta_sbs;
      delete[] out_sbs;

      return in_sbs;
   }

   /**
    * Return the list of potential dependencies of each instruction in the
    * shader based on the result of global dependency analysis.
    */
   dependency_list *
   gather_inst_dependencies(const fs_visitor *shader,
                            const ordered_address *jps)
   {
      const struct intel_device_info *devinfo = shader->devinfo;
      equivalence_relation eq(num_instructions(shader));
      scoreboard *sbs = propagate_block_scoreboards(shader, jps, eq);
      const unsigned *ids = eq.flatten();
      dependency_list *deps = new dependency_list[num_instructions(shader)];
      unsigned ip = 0;

      foreach_block_and_inst(block, fs_inst, inst, shader->cfg) {
         const bool exec_all = inst->force_writemask_all;
         const tgl_pipe p = inferred_exec_pipe(devinfo, inst);
         scoreboard &sb = sbs[block->num];

         for (unsigned i = 0; i < inst->sources; i++) {
            for (unsigned j = 0; j < regs_read(inst, i); j++)
               add_dependency(ids, deps[ip], dependency_for_read(
                  sb.get(byte_offset(inst->src[i], REG_SIZE * j))));
         }

         if (inst->reads_accumulator_implicitly()) {
            /* Wa_22012725308:
             *
             * "When the accumulator registers are used as source and/or
             *  destination, hardware does not ensure prevention of write
             *  after read hazard across execution pipes."
             */
            const dependency dep = sb.get(brw_acc_reg(8));
            if (dep.ordered && !is_single_pipe(dep.jp, p))
               add_dependency(ids, deps[ip], dep);
         }

         if (is_send(inst) && inst->base_mrf != -1) {
            for (unsigned j = 0; j < inst->mlen; j++)
               add_dependency(ids, deps[ip], dependency_for_read(
                  sb.get(brw_uvec_mrf(8, inst->base_mrf + j, 0))));
         }

         if (is_unordered(inst))
            add_dependency(ids, deps[ip],
                           dependency(TGL_SBID_SET, ip, exec_all));

         if (!inst->no_dd_check) {
            if (inst->dst.file != BAD_FILE && !inst->dst.is_null() &&
                !inst->dst.is_accumulator()) {
               for (unsigned j = 0; j < regs_written(inst); j++) {
                  add_dependency(ids, deps[ip], dependency_for_write(inst,
                     sb.get(byte_offset(inst->dst, REG_SIZE * j))));
               }
            }

            if (inst->writes_accumulator_implicitly(devinfo) ||
                inst->dst.is_accumulator()) {
               /* Wa_22012725308:
                *
                * "When the accumulator registers are used as source and/or
                *  destination, hardware does not ensure prevention of write
                *  after read hazard across execution pipes."
                */
               const dependency dep = sb.get(brw_acc_reg(8));
               if (dep.ordered && !is_single_pipe(dep.jp, p))
                  add_dependency(ids, deps[ip], dep);
            }

            if (is_send(inst) && inst->base_mrf != -1) {
               for (unsigned j = 0; j < inst->implied_mrf_writes(); j++)
                  add_dependency(ids, deps[ip], dependency_for_write(inst,
                     sb.get(brw_uvec_mrf(8, inst->base_mrf + j, 0))));
            }
         }

         update_inst_scoreboard(shader, jps, inst, ip, sb);
         ip++;
      }

      delete[] sbs;
      delete[] ids;

      return deps;
   }

   /** @} */

   /**
    * Allocate SBID tokens to track the execution of every out-of-order
    * instruction of the shader.
    */
   dependency_list *
   allocate_inst_dependencies(const fs_visitor *shader,
                              const dependency_list *deps0)
   {
      /* XXX - Use bin-packing algorithm to assign hardware SBIDs optimally in
       *       shaders with a large number of SEND messages.
       */

      /* Allocate an unordered dependency ID to hardware SBID translation
       * table with as many entries as instructions there are in the shader,
       * which is the maximum number of unordered IDs we can find in the
       * program.
       */
      unsigned *ids = new unsigned[num_instructions(shader)];
      for (unsigned ip = 0; ip < num_instructions(shader); ip++)
         ids[ip] = ~0u;

      dependency_list *deps1 = new dependency_list[num_instructions(shader)];
      unsigned next_id = 0;

      for (unsigned ip = 0; ip < num_instructions(shader); ip++) {
         for (unsigned i = 0; i < deps0[ip].size(); i++) {
            const dependency &dep = deps0[ip][i];

            if (dep.unordered && ids[dep.id] == ~0u)
               ids[dep.id] = (next_id++) & 0xf;

            add_dependency(ids, deps1[ip], dep);
         }
      }

      delete[] ids;

      return deps1;
   }

   /**
    * Emit dependency information provided by \p deps into the shader,
    * inserting additional SYNC instructions for dependencies that can't be
    * represented directly by annotating existing instructions.
    */
   void
   emit_inst_dependencies(fs_visitor *shader,
                          const ordered_address *jps,
                          const dependency_list *deps)
   {
      const struct intel_device_info *devinfo = shader->devinfo;
      unsigned ip = 0;

      foreach_block_and_inst_safe(block, fs_inst, inst, shader->cfg) {
         const bool exec_all = inst->force_writemask_all;
         const bool ordered_mode =
            baked_ordered_dependency_mode(devinfo, inst, deps[ip], jps[ip]);
         const tgl_sbid_mode unordered_mode =
            baked_unordered_dependency_mode(devinfo, inst, deps[ip], jps[ip]);
         tgl_swsb swsb = !ordered_mode ? tgl_swsb() :
            ordered_dependency_swsb(deps[ip], jps[ip], exec_all);

         for (unsigned i = 0; i < deps[ip].size(); i++) {
            const dependency &dep = deps[ip][i];

            if (dep.unordered) {
               if (unordered_mode == dep.unordered &&
                   exec_all >= dep.exec_all && !swsb.mode) {
                  /* Bake unordered dependency into the instruction's SWSB if
                   * possible, except in cases where the current instruction
                   * isn't marked NoMask but the dependency is, since that
                   * might lead to data coherency issues due to
                   * Wa_1407528679.
                   */
                  swsb.sbid = dep.id;
                  swsb.mode = dep.unordered;
               } else {
                  /* Emit dependency into the SWSB of an extra SYNC
                   * instruction.
                   */
                  const fs_builder ibld = fs_builder(shader, block, inst)
                                          .exec_all().group(1, 0);
                  fs_inst *sync = ibld.emit(BRW_OPCODE_SYNC, ibld.null_reg_ud(),
                                            brw_imm_ud(TGL_SYNC_NOP));
                  sync->sched.sbid = dep.id;
                  sync->sched.mode = dep.unordered;
                  assert(!(sync->sched.mode & TGL_SBID_SET));
               }
            }
         }

         for (unsigned i = 0; i < deps[ip].size(); i++) {
            const dependency &dep = deps[ip][i];

            if (dep.ordered &&
                find_ordered_dependency(deps[ip], jps[ip], true) &&
                (!ordered_mode || dep.exec_all > exec_all)) {
               /* If the current instruction is not marked NoMask but an
                * ordered dependency is, perform the synchronization as a
                * separate NoMask SYNC instruction in order to avoid data
                * coherency issues due to Wa_1407528679.  The similar
                * scenario with unordered dependencies should have been
                * handled above.
                */
               const fs_builder ibld = fs_builder(shader, block, inst)
                                       .exec_all().group(1, 0);
               fs_inst *sync = ibld.emit(BRW_OPCODE_SYNC, ibld.null_reg_ud(),
                                         brw_imm_ud(TGL_SYNC_NOP));
               sync->sched = ordered_dependency_swsb(deps[ip], jps[ip], true);
               break;
            }
         }

         /* Update the IR. */
         inst->sched = swsb;
         inst->no_dd_check = inst->no_dd_clear = false;
         ip++;
      }
   }
}

bool
fs_visitor::lower_scoreboard()
{
   if (devinfo->ver >= 12) {
      const ordered_address *jps = ordered_inst_addresses(this);
      const dependency_list *deps0 = gather_inst_dependencies(this, jps);
      const dependency_list *deps1 = allocate_inst_dependencies(this, deps0);
      emit_inst_dependencies(this, jps, deps1);
      delete[] deps1;
      delete[] deps0;
      delete[] jps;
   }

   return true;
}