forked from hail-is/hail
-
Notifications
You must be signed in to change notification settings - Fork 0
/
IR.scala
972 lines (788 loc) · 37.7 KB
/
IR.scala
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
package is.hail.expr.ir
import is.hail.annotations.{Annotation, Region}
import is.hail.asm4s.Value
import is.hail.backend.ExecuteContext
import is.hail.expr.ir.ArrayZipBehavior.ArrayZipBehavior
import is.hail.expr.ir.agg.{AggStateSig, PhysicalAggSig}
import is.hail.expr.ir.functions._
import is.hail.expr.ir.lowering.TableStageDependency
import is.hail.expr.ir.streams.StreamProducer
import is.hail.io.avro.{AvroPartitionReader, AvroSchemaSerializer}
import is.hail.io.bgen.BgenPartitionReader
import is.hail.io.{AbstractTypedCodecSpec, BufferSpec, TypedCodecSpec}
import is.hail.rvd.RVDSpecMaker
import is.hail.types.encoded._
import is.hail.types.physical._
import is.hail.types.physical.stypes._
import is.hail.types.physical.stypes.concrete.SJavaString
import is.hail.types.physical.stypes.interfaces._
import is.hail.types.virtual._
import is.hail.types.{RIterable, RStruct, TypeWithRequiredness, tcoerce}
import is.hail.utils.{FastIndexedSeq, _}
import org.json4s.{DefaultFormats, Extraction, Formats, JValue, ShortTypeHints}
import java.io.OutputStream
import scala.language.existentials
sealed trait IR extends BaseIR {
private var _typ: Type = null
def typ: Type = {
if (_typ == null)
try {
_typ = InferType(this)
} catch {
case e: Throwable => throw new RuntimeException(s"typ: inference failure:", e)
}
_typ
}
lazy val children: IndexedSeq[BaseIR] =
Children(this)
override def copy(newChildren: IndexedSeq[BaseIR]): IR =
Copy(this, newChildren)
override def deepCopy(): this.type = {
val cp = super.deepCopy()
if (_typ != null)
cp._typ = _typ
cp
}
lazy val size: Int = 1 + children.map {
case x: IR => x.size
case _ => 0
}.sum
private[this] def _unwrap: IR => IR = {
case node: ApplyIR => MapIR(_unwrap)(node.explicitNode)
case node => MapIR(_unwrap)(node)
}
def unwrap: IR = _unwrap(this)
}
sealed trait TypedIR[T <: Type] extends IR {
override def typ: T = tcoerce[T](super.typ)
}
// Mark Refs and constants as IRs that are safe to duplicate
sealed trait TrivialIR extends IR
object Literal {
def coerce(t: Type, x: Any): IR = {
if (x == null)
return NA(t)
t match {
case TInt32 => I32(x.asInstanceOf[Number].intValue())
case TInt64 => I64(x.asInstanceOf[Number].longValue())
case TFloat32 => F32(x.asInstanceOf[Number].floatValue())
case TFloat64 => F64(x.asInstanceOf[Number].doubleValue())
case TBoolean => if (x.asInstanceOf[Boolean]) True() else False()
case TString => Str(x.asInstanceOf[String])
case _ => Literal(t, x)
}
}
}
final case class Literal(_typ: Type, value: Annotation) extends IR {
require(!CanEmit(_typ))
require(value != null)
// expensive, for debugging
// require(SafeRow.isSafe(value))
// assert(_typ.typeCheck(value), s"literal invalid:\n ${_typ}\n $value")
}
object EncodedLiteral {
def apply(codec: AbstractTypedCodecSpec, value: Array[Array[Byte]]): EncodedLiteral = {
EncodedLiteral(codec, new WrappedByteArrays(value))
}
def fromPTypeAndAddress(pt: PType, addr: Long, ctx: ExecuteContext): IR = {
pt match {
case _: PInt32 => I32(Region.loadInt(addr))
case _: PInt64 => I64(Region.loadLong(addr))
case _: PFloat32 => F32(Region.loadFloat(addr))
case _: PFloat64 => F64(Region.loadDouble(addr))
case _: PBoolean => if (Region.loadBoolean(addr)) True() else False()
case ts: PString => Str(ts.loadString(addr))
case _ =>
val etype = EType.defaultFromPType(pt)
val codec = TypedCodecSpec(etype, pt.virtualType, BufferSpec.defaultUncompressed)
val bytes = codec.encodeArrays(ctx, pt, addr)
EncodedLiteral(codec, bytes)
}
}
}
final case class EncodedLiteral(codec: AbstractTypedCodecSpec, value: WrappedByteArrays) extends IR {
require(!CanEmit(codec.encodedVirtualType))
require(value != null)
}
class WrappedByteArrays(val ba: Array[Array[Byte]]) {
override def hashCode(): Int = {
ba.foldLeft(31) { (h, b) => 37 * h + java.util.Arrays.hashCode(b) }
}
override def equals(obj: Any): Boolean = {
this.eq(obj.asInstanceOf[AnyRef]) || {
if (!obj.isInstanceOf[WrappedByteArrays]) {
false
}
else {
val other = obj.asInstanceOf[WrappedByteArrays]
ba.length == other.ba.length && (ba, other.ba).zipped.forall(java.util.Arrays.equals)
}
}
}
}
final case class I32(x: Int) extends IR with TrivialIR
final case class I64(x: Long) extends IR with TrivialIR
final case class F32(x: Float) extends IR with TrivialIR
final case class F64(x: Double) extends IR with TrivialIR
final case class Str(x: String) extends IR with TrivialIR {
override def toString(): String = s"""Str("${StringEscapeUtils.escapeString(x)}")"""
}
final case class True() extends IR with TrivialIR
final case class False() extends IR with TrivialIR
final case class Void() extends IR with TrivialIR
object UUID4 {
def apply(): UUID4 = UUID4(genUID())
}
// WARNING! This node can only be used when trying to append a one-off,
// random string that will not be reused elsewhere in the pipeline.
// Any other uses will need to write and then read again; this node is
// non-deterministic and will not e.g. exhibit the correct semantics when
// self-joining on streams.
final case class UUID4(id: String) extends IR
final case class Cast(v: IR, _typ: Type) extends IR
final case class CastRename(v: IR, _typ: Type) extends IR
final case class NA(_typ: Type) extends IR with TrivialIR
final case class IsNA(value: IR) extends IR
final case class Coalesce(values: Seq[IR]) extends IR {
require(values.nonEmpty)
}
final case class Consume(value: IR) extends IR
final case class If(cond: IR, cnsq: IR, altr: IR) extends IR
final case class AggLet(name: String, value: IR, body: IR, isScan: Boolean) extends IR
final case class Let(name: String, value: IR, body: IR) extends IR
sealed abstract class BaseRef extends IR with TrivialIR {
def name: String
def _typ: Type
}
final case class Ref(name: String, var _typ: Type) extends BaseRef
// Recur can't exist outside of loop
// Loops can be nested, but we can't call outer loops in terms of inner loops so there can only be one loop "active" in a given context
final case class TailLoop(name: String, params: IndexedSeq[(String, IR)], body: IR) extends IR {
lazy val paramIdx: Map[String, Int] = params.map(_._1).zipWithIndex.toMap
}
final case class Recur(name: String, args: IndexedSeq[IR], _typ: Type) extends BaseRef
final case class RelationalLet(name: String, value: IR, body: IR) extends IR
final case class RelationalRef(name: String, _typ: Type) extends BaseRef
final case class ApplyBinaryPrimOp(op: BinaryOp, l: IR, r: IR) extends IR
final case class ApplyUnaryPrimOp(op: UnaryOp, x: IR) extends IR
final case class ApplyComparisonOp(op: ComparisonOp[_], l: IR, r: IR) extends IR
object MakeArray {
def apply(args: IR*): MakeArray = {
assert(args.nonEmpty)
MakeArray(args.toArray, TArray(args.head.typ))
}
def unify(ctx: ExecuteContext, args: IndexedSeq[IR], requestedType: TArray = null): MakeArray = {
assert(requestedType != null || args.nonEmpty)
if(args.nonEmpty)
if (args.forall(_.typ == args.head.typ))
return MakeArray(args, TArray(args.head.typ))
MakeArray(args.map { arg =>
val upcast = PruneDeadFields.upcast(ctx, arg, requestedType.elementType)
assert(upcast.typ == requestedType.elementType)
upcast
}, requestedType)
}
}
final case class MakeArray(args: IndexedSeq[IR], _typ: TArray) extends IR
object MakeStream {
def unify(ctx: ExecuteContext, args: IndexedSeq[IR], requiresMemoryManagementPerElement: Boolean = false, requestedType: TStream = null): MakeStream = {
assert(requestedType != null || args.nonEmpty)
if (args.nonEmpty)
if (args.forall(_.typ == args.head.typ))
return MakeStream(args, TStream(args.head.typ), requiresMemoryManagementPerElement)
MakeStream(args.map { arg =>
val upcast = PruneDeadFields.upcast(ctx, arg, requestedType.elementType)
assert(upcast.typ == requestedType.elementType)
upcast
}, requestedType, requiresMemoryManagementPerElement)
}
}
final case class MakeStream(args: IndexedSeq[IR], _typ: TStream, requiresMemoryManagementPerElement: Boolean = false) extends IR
object ArrayRef {
def apply(a: IR, i: IR): ArrayRef = ArrayRef(a, i, ErrorIDs.NO_ERROR)
}
final case class ArrayRef(a: IR, i: IR, errorID: Int) extends IR
final case class ArraySlice(a: IR, start: IR, stop: Option[IR], step:IR = I32(1), errorID: Int = ErrorIDs.NO_ERROR) extends IR
final case class ArrayLen(a: IR) extends IR
final case class ArrayZeros(length: IR) extends IR
final case class ArrayMaximalIndependentSet(edges: IR, tieBreaker: Option[(String, String, IR)]) extends IR
/**
* [[StreamIota]] is an infinite stream producer, whose element is an integer starting at `start`, updated by
* `step` at each iteration. The name comes from APL:
* [[https://stackoverflow.com/questions/9244879/what-does-iota-of-stdiota-stand-for]]
*/
final case class StreamIota(start: IR, step: IR, requiresMemoryManagementPerElement: Boolean = false) extends IR
final case class StreamRange(start: IR, stop: IR, step: IR, requiresMemoryManagementPerElement: Boolean = false,
errorID: Int = ErrorIDs.NO_ERROR) extends IR
object ArraySort {
def apply(a: IR, ascending: IR = True(), onKey: Boolean = false): ArraySort = {
val l = genUID()
val r = genUID()
val atyp = tcoerce[TStream](a.typ)
val compare = if (onKey) {
val elementType = atyp.elementType.asInstanceOf[TBaseStruct]
elementType match {
case t: TStruct =>
val elt = tcoerce[TStruct](atyp.elementType)
ApplyComparisonOp(Compare(elt.types(0)), GetField(Ref(l, elt), elt.fieldNames(0)), GetField(Ref(r, atyp.elementType), elt.fieldNames(0)))
case t: TTuple =>
val elt = tcoerce[TTuple](atyp.elementType)
ApplyComparisonOp(Compare(elt.types(0)), GetTupleElement(Ref(l, elt), elt.fields(0).index), GetTupleElement(Ref(r, atyp.elementType), elt.fields(0).index))
}
} else {
ApplyComparisonOp(Compare(atyp.elementType), Ref(l, atyp.elementType), Ref(r, atyp.elementType))
}
ArraySort(a, l, r, If(ascending, compare < 0, compare > 0))
}
}
final case class ArraySort(a: IR, left: String, right: String, lessThan: IR) extends IR
final case class ToSet(a: IR) extends IR
final case class ToDict(a: IR) extends IR
final case class ToArray(a: IR) extends IR
final case class CastToArray(a: IR) extends IR
final case class ToStream(a: IR, requiresMemoryManagementPerElement: Boolean = false) extends IR
final case class StreamBufferedAggregate(streamChild: IR, initAggs: IR, newKey: IR, seqOps: IR, name: String,
aggSignatures: IndexedSeq[PhysicalAggSig], bufferSize: Int) extends IR
final case class LowerBoundOnOrderedCollection(orderedCollection: IR, elem: IR, onKey: Boolean) extends IR
final case class GroupByKey(collection: IR) extends IR
final case class RNGStateLiteral() extends IR
final case class RNGSplit(state: IR, dynBitstring: IR) extends IR
final case class StreamLen(a: IR) extends IR
final case class StreamGrouped(a: IR, groupSize: IR) extends IR
final case class StreamGroupByKey(a: IR, key: IndexedSeq[String], missingEqual: Boolean) extends IR
final case class StreamMap(a: IR, name: String, body: IR) extends IR {
override def typ: TStream = tcoerce[TStream](super.typ)
def elementTyp: Type = typ.elementType
}
final case class StreamTakeWhile(a: IR, elementName: String, body: IR) extends IR
final case class StreamDropWhile(a: IR, elementName: String, body: IR) extends IR
final case class StreamTake(a: IR, num: IR) extends IR
final case class StreamDrop(a: IR, num: IR) extends IR
// Generate, in ascending order, a uniform random sample, without replacement, of numToSample integers in the range [0, totalRange)
final case class SeqSample(totalRange: IR, numToSample: IR, rngState: IR, requiresMemoryManagementPerElement: Boolean) extends IR
// Take the child stream and sort each element into buckets based on the provided pivots. The first and last elements of
// pivots are the endpoints of the first and last interval respectively, should not be contained in the dataset.
final case class StreamDistribute(child: IR, pivots: IR, path: IR, comparisonOp: ComparisonOp[_],spec: AbstractTypedCodecSpec) extends IR
// "Whiten" a stream of vectors by regressing out from each vector all components
// in the direction of vectors in the preceding window. For efficiency, takes
// a stream of "chunks" of vectors.
// Takes a stream of structs, with two designated fields: `prevWindow` is the
// previous window (e.g. from the previous partition), if there is one, and
// `newChunk` is the new chunk to whiten.
final case class StreamWhiten(stream: IR, newChunk: String, prevWindow: String, vecSize: Int, windowSize: Int, chunkSize: Int, blockSize: Int, normalizeAfterWhiten: Boolean) extends IR
object ArrayZipBehavior extends Enumeration {
type ArrayZipBehavior = Value
val AssumeSameLength: Value = Value(0)
val AssertSameLength: Value = Value(1)
val TakeMinLength: Value = Value(2)
val ExtendNA: Value = Value(3)
}
final case class StreamZip(as: IndexedSeq[IR], names: IndexedSeq[String], body: IR, behavior: ArrayZipBehavior,
errorID: Int = ErrorIDs.NO_ERROR) extends IR {
lazy val nameIdx: Map[String, Int] = names.zipWithIndex.toMap
override def typ: TStream = tcoerce[TStream](super.typ)
}
final case class StreamMultiMerge(as: IndexedSeq[IR], key: IndexedSeq[String]) extends IR {
override def typ: TStream = tcoerce[TStream](super.typ)
}
/**
* The StreamZipJoin node assumes that input streams have distinct keys. If input streams
* do not have distinct keys, the key that is included in the result is undefined, but
* is likely the last.
*/
final case class StreamZipJoin(as: IndexedSeq[IR], key: IndexedSeq[String], curKey: String, curVals: String, joinF: IR) extends IR {
override def typ: TStream = tcoerce[TStream](super.typ)
}
final case class StreamFilter(a: IR, name: String, cond: IR) extends IR {
override def typ: TStream = tcoerce[TStream](super.typ)
}
final case class StreamFlatMap(a: IR, name: String, body: IR) extends IR {
override def typ: TStream = tcoerce[TStream](super.typ)
}
final case class StreamFold(a: IR, zero: IR, accumName: String, valueName: String, body: IR) extends IR
object StreamFold2 {
def apply(a: StreamFold): StreamFold2 = {
StreamFold2(a.a, FastIndexedSeq((a.accumName, a.zero)), a.valueName, FastSeq(a.body), Ref(a.accumName, a.zero.typ))
}
}
final case class StreamFold2(a: IR, accum: IndexedSeq[(String, IR)], valueName: String, seq: IndexedSeq[IR], result: IR) extends IR {
assert(accum.length == seq.length)
val nameIdx: Map[String, Int] = accum.map(_._1).zipWithIndex.toMap
}
final case class StreamScan(a: IR, zero: IR, accumName: String, valueName: String, body: IR) extends IR
final case class StreamFor(a: IR, valueName: String, body: IR) extends IR
final case class StreamAgg(a: IR, name: String, query: IR) extends IR
final case class StreamAggScan(a: IR, name: String, query: IR) extends IR
object StreamJoin {
def apply(
left: IR, right: IR,
lKey: IndexedSeq[String], rKey: IndexedSeq[String],
l: String, r: String,
joinF: IR,
joinType: String,
requiresMemoryManagement: Boolean,
rightKeyIsDistinct: Boolean = false
): IR = {
val lType = tcoerce[TStream](left.typ)
val rType = tcoerce[TStream](right.typ)
val lEltType = tcoerce[TStruct](lType.elementType)
val rEltType = tcoerce[TStruct](rType.elementType)
assert(lEltType.typeAfterSelectNames(lKey) isIsomorphicTo rEltType.typeAfterSelectNames(rKey))
if(!rightKeyIsDistinct) {
val rightGroupedStream = StreamGroupByKey(right, rKey, missingEqual = false)
val groupField = genUID()
// stream of {key, groupField}, where 'groupField' is an array of all rows
// in 'right' with key 'key'
val rightGrouped = mapIR(rightGroupedStream) { group =>
bindIR(ToArray(group)) { array =>
bindIR(ArrayRef(array, 0)) { head =>
MakeStruct(rKey.map { key => key -> GetField(head, key) } :+ groupField -> array)
}
}
}
val rElt = Ref(genUID(), tcoerce[TStream](rightGrouped.typ).elementType)
val lElt = Ref(genUID(), lEltType)
val makeTupleFromJoin = MakeStruct(FastSeq("left" -> lElt, "rightGroup" -> rElt))
val joined = StreamJoinRightDistinct(left, rightGrouped, lKey, rKey, lElt.name, rElt.name, makeTupleFromJoin, joinType)
// joined is a stream of {leftElement, rightGroup}
bindIR(MakeArray(NA(rEltType))) { missingSingleton =>
flatMapIR(joined) { x =>
Let(l, GetField(x, "left"), bindIR(GetField(GetField(x, "rightGroup"), groupField)) { rightElts =>
joinType match {
case "left" | "outer" => StreamMap(ToStream(If(IsNA(rightElts), missingSingleton, rightElts), requiresMemoryManagement), r, joinF)
case "right" | "inner" => StreamMap(ToStream(rightElts, requiresMemoryManagement), r, joinF)
}
})
}
}
}
else {
val rElt = Ref(r, rEltType)
val lElt = Ref(l, lEltType)
StreamJoinRightDistinct(left, right, lKey, rKey, lElt.name, rElt.name, joinF, joinType)
}
}
}
final case class StreamJoinRightDistinct(left: IR, right: IR, lKey: IndexedSeq[String], rKey: IndexedSeq[String], l: String, r: String, joinF: IR, joinType: String) extends IR {
def isIntervalJoin: Boolean = {
if (rKey.size != 1) return false
val lKeyTyp = tcoerce[TStruct](tcoerce[TStream](left.typ).elementType).fieldType(lKey(0))
val rKeyTyp = tcoerce[TStruct](tcoerce[TStream](right.typ).elementType).fieldType(rKey(0))
rKeyTyp.isInstanceOf[TInterval] && lKeyTyp != rKeyTyp
}
}
final case class StreamLocalLDPrune(child: IR, r2Threshold: IR, windowSize: IR, maxQueueSize: IR, nSamples: IR) extends IR
sealed trait NDArrayIR extends TypedIR[TNDArray] {
def elementTyp: Type = typ.elementType
}
object MakeNDArray {
def fill(elt: IR, shape: IndexedSeq[IR], rowMajor: IR): MakeNDArray = {
val flatSize: IR = if (shape.nonEmpty)
shape.reduce { (l, r) => l * r }
else
0L
MakeNDArray(
ToArray(mapIR(rangeIR(flatSize.toI))(_ => elt)),
MakeTuple.ordered(shape), rowMajor, ErrorIDs.NO_ERROR)
}
}
final case class MakeNDArray(data: IR, shape: IR, rowMajor: IR, errorId: Int) extends NDArrayIR
final case class NDArrayShape(nd: IR) extends IR
final case class NDArrayReshape(nd: IR, shape: IR, errorID: Int) extends NDArrayIR
final case class NDArrayConcat(nds: IR, axis: Int) extends NDArrayIR
final case class NDArrayRef(nd: IR, idxs: IndexedSeq[IR], errorId: Int) extends IR
final case class NDArraySlice(nd: IR, slices: IR) extends NDArrayIR
final case class NDArrayFilter(nd: IR, keep: IndexedSeq[IR]) extends NDArrayIR
final case class NDArrayMap(nd: IR, valueName: String, body: IR) extends NDArrayIR
final case class NDArrayMap2(l: IR, r: IR, lName: String, rName: String, body: IR, errorID: Int) extends NDArrayIR
final case class NDArrayReindex(nd: IR, indexExpr: IndexedSeq[Int]) extends NDArrayIR
final case class NDArrayAgg(nd: IR, axes: IndexedSeq[Int]) extends IR
final case class NDArrayWrite(nd: IR, path: IR) extends IR
final case class NDArrayMatMul(l: IR, r: IR, errorID: Int) extends NDArrayIR
object NDArrayQR {
def pType(mode: String, req: Boolean): PType = {
mode match {
case "r" => PCanonicalNDArray(PFloat64Required, 2, req)
case "raw" => PCanonicalTuple(req, PCanonicalNDArray(PFloat64Required, 2, true), PCanonicalNDArray(PFloat64Required, 1, true))
case "reduced" => PCanonicalTuple(req, PCanonicalNDArray(PFloat64Required, 2, true), PCanonicalNDArray(PFloat64Required, 2, true))
case "complete" => PCanonicalTuple(req, PCanonicalNDArray(PFloat64Required, 2, true), PCanonicalNDArray(PFloat64Required, 2, true))
}
}
}
object NDArraySVD {
def pTypes(computeUV: Boolean, req: Boolean): PType = {
if (computeUV) {
PCanonicalTuple(req, PCanonicalNDArray(PFloat64Required, 2, true), PCanonicalNDArray(PFloat64Required, 1, true), PCanonicalNDArray(PFloat64Required, 2, true))
}
else {
PCanonicalNDArray(PFloat64Required, 1, req)
}
}
}
object NDArrayInv {
val pType = PCanonicalNDArray(PFloat64Required, 2)
}
final case class NDArrayQR(nd: IR, mode: String, errorID: Int) extends IR
final case class NDArraySVD(nd: IR, fullMatrices: Boolean, computeUV: Boolean, errorID: Int) extends IR
object NDArrayEigh {
def pTypes(eigvalsOnly: Boolean, req: Boolean): PType = {
if (eigvalsOnly) {
PCanonicalNDArray(PFloat64Required, 1, req)
} else {
PCanonicalTuple(req, PCanonicalNDArray(PFloat64Required, 1, true), PCanonicalNDArray(PFloat64Required, 2, true))
}
}
}
final case class NDArrayEigh(nd: IR, eigvalsOnly: Boolean, errorID: Int) extends IR
final case class NDArrayInv(nd: IR, errorID: Int) extends IR
final case class AggFilter(cond: IR, aggIR: IR, isScan: Boolean) extends IR
final case class AggExplode(array: IR, name: String, aggBody: IR, isScan: Boolean) extends IR
final case class AggGroupBy(key: IR, aggIR: IR, isScan: Boolean) extends IR
final case class AggArrayPerElement(a: IR, elementName: String, indexName: String, aggBody: IR, knownLength: Option[IR], isScan: Boolean) extends IR
object ApplyAggOp {
def apply(op: AggOp, initOpArgs: IR*)(seqOpArgs: IR*): ApplyAggOp =
ApplyAggOp(initOpArgs.toIndexedSeq, seqOpArgs.toIndexedSeq, AggSignature(op, initOpArgs.map(_.typ), seqOpArgs.map(_.typ)))
}
final case class ApplyAggOp(initOpArgs: IndexedSeq[IR], seqOpArgs: IndexedSeq[IR], aggSig: AggSignature) extends IR {
def nSeqOpArgs = seqOpArgs.length
def nInitArgs = initOpArgs.length
def op: AggOp = aggSig.op
}
object AggFold {
def min(element: IR, sortFields: IndexedSeq[SortField]): IR = {
val elementType = element.typ.asInstanceOf[TStruct]
val keyType = elementType.select(sortFields.map(_.field))._1
minAndMaxHelper(element, keyType, StructLT(keyType, sortFields))
}
def max(element: IR, sortFields: IndexedSeq[SortField]): IR = {
val elementType = element.typ.asInstanceOf[TStruct]
val keyType = elementType.select(sortFields.map(_.field))._1
minAndMaxHelper(element, keyType, StructGT(keyType, sortFields))
}
def all(element: IR): IR = {
aggFoldIR(True(), element) { case (accum, element) =>
ApplySpecial("land", Seq.empty[Type], Seq(accum, element), TBoolean, ErrorIDs.NO_ERROR)
} { case (accum1, accum2) => ApplySpecial("land", Seq.empty[Type], Seq(accum1, accum2), TBoolean, ErrorIDs.NO_ERROR) }
}
private def minAndMaxHelper(element: IR, keyType: TStruct, comp: ComparisonOp[Boolean]): IR = {
val keyFields = keyType.fields.map(_.name)
val minAndMaxZero = NA(keyType)
val aggFoldMinAccumName1 = genUID()
val aggFoldMinAccumName2 = genUID()
val aggFoldMinAccumRef1 = Ref(aggFoldMinAccumName1, keyType)
val aggFoldMinAccumRef2 = Ref(aggFoldMinAccumName2, keyType)
val minSeq = bindIR(SelectFields(element, keyFields)) { keyOfCurElementRef =>
If(IsNA(aggFoldMinAccumRef1),
keyOfCurElementRef,
If(ApplyComparisonOp(comp, aggFoldMinAccumRef1, keyOfCurElementRef), aggFoldMinAccumRef1, keyOfCurElementRef)
)
}
val minComb =
If(IsNA(aggFoldMinAccumRef1),
aggFoldMinAccumRef2,
If (ApplyComparisonOp(comp, aggFoldMinAccumRef1, aggFoldMinAccumRef2), aggFoldMinAccumRef1, aggFoldMinAccumRef2)
)
AggFold(minAndMaxZero, minSeq, minComb, aggFoldMinAccumName1, aggFoldMinAccumName2, false)
}
}
final case class AggFold(zero: IR, seqOp: IR, combOp: IR, accumName: String, otherAccumName: String, isScan: Boolean) extends IR
object ApplyScanOp {
def apply(op: AggOp, initOpArgs: IR*)(seqOpArgs: IR*): ApplyScanOp =
ApplyScanOp(initOpArgs.toIndexedSeq, seqOpArgs.toIndexedSeq, AggSignature(op, initOpArgs.map(_.typ), seqOpArgs.map(_.typ)))
}
final case class ApplyScanOp(initOpArgs: IndexedSeq[IR], seqOpArgs: IndexedSeq[IR], aggSig: AggSignature) extends IR {
def nSeqOpArgs = seqOpArgs.length
def nInitArgs = initOpArgs.length
def op: AggOp = aggSig.op
}
final case class InitOp(i: Int, args: IndexedSeq[IR], aggSig: PhysicalAggSig) extends IR
final case class SeqOp(i: Int, args: IndexedSeq[IR], aggSig: PhysicalAggSig) extends IR
final case class CombOp(i1: Int, i2: Int, aggSig: PhysicalAggSig) extends IR
object ResultOp {
def makeTuple(aggs: IndexedSeq[PhysicalAggSig]) = {
MakeTuple.ordered(aggs.zipWithIndex.map { case (aggSig, index) =>
ResultOp(index, aggSig)
})
}
}
final case class ResultOp(idx: Int, aggSig: PhysicalAggSig) extends IR
private final case class CombOpValue(i: Int, value: IR, aggSig: PhysicalAggSig) extends IR
final case class AggStateValue(i: Int, aggSig: AggStateSig) extends IR
final case class InitFromSerializedValue(i: Int, value: IR, aggSig: AggStateSig) extends IR
final case class SerializeAggs(startIdx: Int, serializedIdx: Int, spec: BufferSpec, aggSigs: IndexedSeq[AggStateSig]) extends IR
final case class DeserializeAggs(startIdx: Int, serializedIdx: Int, spec: BufferSpec, aggSigs: IndexedSeq[AggStateSig]) extends IR
final case class RunAgg(body: IR, result: IR, signature: IndexedSeq[AggStateSig]) extends IR
final case class RunAggScan(array: IR, name: String, init: IR, seqs: IR, result: IR, signature: IndexedSeq[AggStateSig]) extends IR
final case class Begin(xs: IndexedSeq[IR]) extends IR
final case class MakeStruct(fields: IndexedSeq[(String, IR)]) extends IR
final case class SelectFields(old: IR, fields: IndexedSeq[String]) extends IR
object InsertFields {
def apply(old: IR, fields: Seq[(String, IR)]): InsertFields = InsertFields(old, fields, None)
}
final case class InsertFields(old: IR, fields: Seq[(String, IR)], fieldOrder: Option[IndexedSeq[String]]) extends IR {
override def typ: TStruct = tcoerce[TStruct](super.typ)
}
object GetFieldByIdx {
def apply(s: IR, field: Int): IR = {
(s.typ: @unchecked) match {
case t: TStruct => GetField(s, t.fieldNames(field))
case _: TTuple => GetTupleElement(s, field)
}
}
}
final case class GetField(o: IR, name: String) extends IR
object MakeTuple {
def ordered(types: IndexedSeq[IR]): MakeTuple = MakeTuple(types.zipWithIndex.map { case (ir, i) => (i, ir) })
}
final case class MakeTuple(fields: IndexedSeq[(Int, IR)]) extends IR
final case class GetTupleElement(o: IR, idx: Int) extends IR
object In {
def apply(i: Int, typ: Type): In = In(i, SingleCodeEmitParamType(false, typ match {
case TInt32 => Int32SingleCodeType
case TInt64 => Int64SingleCodeType
case TFloat32 => Float32SingleCodeType
case TFloat64 => Float64SingleCodeType
case TBoolean => BooleanSingleCodeType
case ts: TStream => throw new UnsupportedOperationException
case t => PTypeReferenceSingleCodeType(PType.canonical(t))
}))
}
// Function Input
final case class In(i: Int, _typ: EmitParamType) extends IR
// FIXME: should be type any
object Die {
def apply(message: String, typ: Type): Die = Die(Str(message), typ, ErrorIDs.NO_ERROR)
def apply(message: String, typ: Type, errorId: Int): Die = Die(Str(message), typ, errorId)
}
/**
* the Trap node runs the `child` node with an exception handler. If the child
* throws a HailException (user exception), then we return the tuple ((msg, errorId), NA).
* If the child throws any other exception, we raise that exception. If the
* child does not throw, then we return the tuple (NA, child value).
*/
final case class Trap(child: IR) extends IR
final case class Die(message: IR, _typ: Type, errorId: Int) extends IR
final case class ConsoleLog(message: IR, result: IR) extends IR
final case class ApplyIR(function: String, typeArgs: Seq[Type], args: Seq[IR], errorID: Int) extends IR {
var conversion: (Seq[Type], Seq[IR], Int) => IR = _
var inline: Boolean = _
private lazy val refs = args.map(a => Ref(genUID(), a.typ)).toArray
lazy val body: IR = conversion(typeArgs, refs, errorID).deepCopy()
lazy val refIdx: Map[String, Int] = refs.map(_.name).zipWithIndex.toMap
lazy val explicitNode: IR = {
// foldRight because arg1 should be at the top so it is evaluated first
refs.zip(args).foldRight(body) { case ((ref, arg), bodyIR) => Let(ref.name, arg, bodyIR) }
}
}
sealed abstract class AbstractApplyNode[F <: JVMFunction] extends IR {
def function: String
def args: Seq[IR]
def returnType: Type
def typeArgs: Seq[Type]
def argTypes: Seq[Type] = args.map(_.typ)
lazy val implementation: F = IRFunctionRegistry.lookupFunctionOrFail(function, returnType, typeArgs, argTypes)
.asInstanceOf[F]
}
final case class Apply(function: String, typeArgs: Seq[Type], args: Seq[IR], returnType: Type, errorID: Int) extends AbstractApplyNode[UnseededMissingnessObliviousJVMFunction]
final case class ApplySeeded(function: String, _args: Seq[IR], rngState: IR, staticUID: Long, returnType: Type) extends AbstractApplyNode[UnseededMissingnessObliviousJVMFunction] {
val args = rngState +: _args
val typeArgs: Seq[Type] = Seq.empty[Type]
}
final case class ApplySpecial(function: String, typeArgs: Seq[Type], args: Seq[IR], returnType: Type, errorID: Int) extends AbstractApplyNode[UnseededMissingnessAwareJVMFunction]
final case class LiftMeOut(child: IR) extends IR
final case class TableCount(child: TableIR) extends IR
final case class MatrixCount(child: MatrixIR) extends IR
final case class TableAggregate(child: TableIR, query: IR) extends IR
final case class MatrixAggregate(child: MatrixIR, query: IR) extends IR
final case class TableWrite(child: TableIR, writer: TableWriter) extends IR
final case class TableMultiWrite(_children: IndexedSeq[TableIR], writer: WrappedMatrixNativeMultiWriter) extends IR {
private val t = _children.head.typ
require(_children.forall(_.typ == t))
}
final case class TableGetGlobals(child: TableIR) extends IR
final case class TableCollect(child: TableIR) extends IR
final case class MatrixWrite(child: MatrixIR, writer: MatrixWriter) extends IR
final case class MatrixMultiWrite(_children: IndexedSeq[MatrixIR], writer: MatrixNativeMultiWriter) extends IR {
private val t = _children.head.typ
assert(!t.rowType.hasField(MatrixReader.rowUIDFieldName) &&
!t.colType.hasField(MatrixReader.colUIDFieldName), t)
require(_children.forall(_.typ == t))
}
final case class TableToValueApply(child: TableIR, function: TableToValueFunction) extends IR
final case class MatrixToValueApply(child: MatrixIR, function: MatrixToValueFunction) extends IR
final case class BlockMatrixToValueApply(child: BlockMatrixIR, function: BlockMatrixToValueFunction) extends IR
final case class BlockMatrixCollect(child: BlockMatrixIR) extends NDArrayIR
final case class BlockMatrixWrite(child: BlockMatrixIR, writer: BlockMatrixWriter) extends IR
final case class BlockMatrixMultiWrite(blockMatrices: IndexedSeq[BlockMatrixIR], writer: BlockMatrixMultiWriter) extends IR
final case class CollectDistributedArray(contexts: IR, globals: IR, cname: String, gname: String, body: IR, dynamicID: IR, staticID: String, tsd: Option[TableStageDependency] = None) extends IR
object PartitionReader {
implicit val formats: Formats = new DefaultFormats() {
override val typeHints = ShortTypeHints(List(
classOf[PartitionRVDReader],
classOf[PartitionNativeReader],
classOf[PartitionNativeReaderIndexed],
classOf[PartitionNativeIntervalReader],
classOf[PartitionZippedNativeReader],
classOf[PartitionZippedIndexedNativeReader],
classOf[BgenPartitionReader],
classOf[AbstractTypedCodecSpec],
classOf[TypedCodecSpec],
classOf[AvroPartitionReader]),
typeHintFieldName = "name") + BufferSpec.shortTypeHints
} +
new TStructSerializer +
new TypeSerializer +
new PTypeSerializer +
new ETypeSerializer +
new AvroSchemaSerializer
def extract(ctx: ExecuteContext, jv: JValue): PartitionReader = {
(jv \ "name").extract[String] match {
case "PartitionNativeIntervalReader" =>
val path = (jv \ "path").extract[String]
val spec = TableNativeReader.read(ctx.fs, path, None).spec
PartitionNativeIntervalReader(ctx.stateManager, path, spec, (jv \ "uidFieldName").extract[String])
case _ => jv.extract[PartitionReader]
}
}
}
object PartitionWriter {
implicit val formats: Formats = new DefaultFormats() {
override val typeHints = ShortTypeHints(List(
classOf[PartitionNativeWriter],
classOf[TableTextPartitionWriter],
classOf[VCFPartitionWriter],
classOf[GenSampleWriter],
classOf[GenVariantWriter],
classOf[AbstractTypedCodecSpec],
classOf[TypedCodecSpec]), typeHintFieldName = "name"
) + BufferSpec.shortTypeHints
} +
new TStructSerializer +
new TypeSerializer +
new PTypeSerializer +
new PStructSerializer +
new ETypeSerializer
}
object MetadataWriter {
implicit val formats: Formats = new DefaultFormats() {
override val typeHints = ShortTypeHints(List(
classOf[RVDSpecWriter],
classOf[TableSpecWriter],
classOf[RelationalWriter],
classOf[TableTextFinalizer],
classOf[VCFExportFinalizer],
classOf[SimpleMetadataWriter],
classOf[RVDSpecMaker],
classOf[AbstractTypedCodecSpec],
classOf[TypedCodecSpec]),
typeHintFieldName = "name"
) + BufferSpec.shortTypeHints
} +
new TStructSerializer +
new TypeSerializer +
new PTypeSerializer +
new ETypeSerializer
}
abstract class PartitionReader {
assert(fullRowType.hasField(uidFieldName))
def contextType: Type
def fullRowType: TStruct
def uidFieldName: String
def rowRequiredness(requestedType: TStruct): RStruct
def emitStream(
ctx: ExecuteContext,
cb: EmitCodeBuilder,
mb: EmitMethodBuilder[_],
context: EmitCode,
requestedType: TStruct
): IEmitCode
def toJValue: JValue
}
abstract class PartitionWriter {
def consumeStream(
ctx: ExecuteContext,
cb: EmitCodeBuilder,
stream: StreamProducer,
context: EmitCode,
region: Value[Region]): IEmitCode
def ctxType: Type
def returnType: Type
def unionTypeRequiredness(r: TypeWithRequiredness, ctxType: TypeWithRequiredness, streamType: RIterable): Unit
def toJValue: JValue = Extraction.decompose(this)(PartitionWriter.formats)
}
abstract class SimplePartitionWriter extends PartitionWriter {
def ctxType: Type = TString
def returnType: Type = TString
def unionTypeRequiredness(r: TypeWithRequiredness, ctxType: TypeWithRequiredness, streamType: RIterable): Unit = {
r.union(ctxType.required)
r.union(streamType.required)
}
def consumeElement(cb: EmitCodeBuilder, element: EmitCode, os: Value[OutputStream], region: Value[Region]): Unit
def preConsume(cb: EmitCodeBuilder, os: Value[OutputStream]): Unit = ()
def postConsume(cb: EmitCodeBuilder, os: Value[OutputStream]): Unit = ()
final def consumeStream(ctx: ExecuteContext, cb: EmitCodeBuilder, stream: StreamProducer,
context: EmitCode, region: Value[Region]): IEmitCode = {
context.toI(cb).map(cb) { case ctx: SStringValue =>
val filename = ctx.loadString(cb)
val os = cb.memoize(cb.emb.create(filename))
preConsume(cb, os)
stream.memoryManagedConsume(region, cb) { cb =>
consumeElement(cb, stream.element, os, stream.elementRegion)
}
postConsume(cb, os)
cb += os.invoke[Unit]("flush")
cb += os.invoke[Unit]("close")
SJavaString.construct(cb, filename)
}
}
}
abstract class MetadataWriter {
def annotationType: Type
def writeMetadata(
writeAnnotations: => IEmitCode,
cb: EmitCodeBuilder,
region: Value[Region]): Unit
def toJValue: JValue = Extraction.decompose(this)(MetadataWriter.formats)
}
final case class SimpleMetadataWriter(val annotationType: Type) extends MetadataWriter {
def writeMetadata(writeAnnotations: => IEmitCode, cb: EmitCodeBuilder, region: Value[Region]): Unit =
writeAnnotations.consume(cb, {}, {_ => ()})
}
final case class ReadPartition(context: IR, rowType: TStruct, reader: PartitionReader) extends IR {
assert(context.typ == reader.contextType, s"context: ${context.typ}, expected: ${reader.contextType}")
assert(PruneDeadFields.isSupertype(rowType, reader.fullRowType), s"requested type: $rowType, full type: ${reader.fullRowType}")
}
final case class WritePartition(value: IR, writeCtx: IR, writer: PartitionWriter) extends IR
final case class WriteMetadata(writeAnnotations: IR, writer: MetadataWriter) extends IR
final case class ReadValue(path: IR, reader: ValueReader, requestedType: Type) extends IR
final case class WriteValue(value: IR, path: IR, writer: ValueWriter, stagingFile: Option[IR] = None) extends IR
class PrimitiveIR(val self: IR) extends AnyVal {
def +(other: IR): IR = {
assert(self.typ == other.typ)
if (self.typ == TString)
invoke("concat", TString, self, other)
else
ApplyBinaryPrimOp(Add(), self, other)
}
def -(other: IR): IR = ApplyBinaryPrimOp(Subtract(), self, other)
def *(other: IR): IR = ApplyBinaryPrimOp(Multiply(), self, other)
def /(other: IR): IR = ApplyBinaryPrimOp(FloatingPointDivide(), self, other)
def floorDiv(other: IR): IR = ApplyBinaryPrimOp(RoundToNegInfDivide(), self, other)
def &&(other: IR): IR = invoke("land", TBoolean, self, other)
def ||(other: IR): IR = invoke("lor", TBoolean, self, other)
def toI: IR = Cast(self, TInt32)
def toL: IR = Cast(self, TInt64)
def toF: IR = Cast(self, TFloat32)
def toD: IR = Cast(self, TFloat64)
def unary_-(): IR = ApplyUnaryPrimOp(Negate(), self)
def unary_!(): IR = ApplyUnaryPrimOp(Bang(), self)
def ceq(other: IR): IR = ApplyComparisonOp(EQWithNA(self.typ, other.typ), self, other)
def cne(other: IR): IR = ApplyComparisonOp(NEQWithNA(self.typ, other.typ), self, other)
def <(other: IR): IR = ApplyComparisonOp(LT(self.typ, other.typ), self, other)
def >(other: IR): IR = ApplyComparisonOp(GT(self.typ, other.typ), self, other)
def <=(other: IR): IR = ApplyComparisonOp(LTEQ(self.typ, other.typ), self, other)
def >=(other: IR): IR = ApplyComparisonOp(GTEQ(self.typ, other.typ), self, other)
}
object ErrorIDs {
val NO_ERROR = -1
}