Repository navigation
Expand file tree
/
Copy pathscatter.py
More file actions
1124 lines (941 loc) · 40.4 KB
/
Copy pathscatter.py
File metadata and controls
1124 lines (941 loc) · 40.4 KB
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) 2026, Modular Inc. All rights reserved.
#
# Licensed under the Apache License v2.0 with LLVM Exceptions:
# https://llvm.org/LICENSE.txt
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ===----------------------------------------------------------------------=== #
"""Op implementation for scatter."""
from max._core.dialects import builtin, kgen, rmo
from max.dtype import DType
from .. import dtype_promotion
from ..dim import DimLike
from ..graph import Graph
from ..type import DeviceRef
from ..value import TensorValue, TensorValueLike
from .constant import constant
from .nonzero import nonzero
from .validation import assert_same_device
def scatter(
input: TensorValueLike,
updates: TensorValueLike,
indices: TensorValueLike,
axis: int = -1,
) -> TensorValue:
"""Writes ``updates`` into a copy of ``input`` at positions given by ``indices``.
.. code-block:: python
from max.dtype import DType
from max.engine import InferenceSession
from max.graph import DeviceRef, Graph, ops
device = DeviceRef.CPU()
with Graph("scatter") as graph:
x = ops.constant([1, 2, 3, 4, 5], DType.int32, device=device)
updates = ops.constant([10, 20], DType.int32, device=device)
indices = ops.constant([0, 3], DType.int64, device=device)
# Overwrite positions 0 and 3, producing [10, 2, 3, 20, 5].
graph.output(ops.scatter(x, updates, indices, axis=0))
model = InferenceSession().load(graph)
result = model.execute()[0]
Args:
input: The input symbolic tensor to write elements to.
updates: A symbolic tensor of elements to write to ``input``.
indices: The positions in ``input`` to update.
axis: The axis along which ``indices`` indexes. Defaults to ``-1``.
Returns:
A ``TensorValue`` representing ``input`` with ``updates`` written at
``indices``. It has the same shape and dtype as ``input``.
Raises:
ValueError: If ``axis`` is out of range, if dtypes mismatch, if
``indices`` dtype is not int32/int64, or if ``input``, ``updates``,
and ``indices`` aren't on the same device.
Error: If ``input``, ``updates``, and ``indices`` don't have equal
rank, if ``updates.shape`` and ``indices.shape`` differ, or if any
``indices`` dimension exceeds the corresponding ``input``
dimension.
"""
input = TensorValue(input)
if not (-input.rank <= axis < input.rank):
raise ValueError(
f"Invalid axis value {axis}. Axis must be in range [-{input.rank},"
f" {input.rank - 1}]"
)
updates = TensorValue(updates)
if input.dtype != updates.dtype:
raise ValueError(
f"The input dtype '{input.dtype}' and updates dtype"
f" '{updates.dtype}' must match."
)
indices = TensorValue(indices)
if indices.dtype not in [DType.int32, DType.int64]:
raise ValueError(
f"Invalid indices dtype: '{indices.dtype}'. Indices must be of type"
" int32 or int64."
)
assert_same_device(input=input, updates=updates, indices=indices)
return Graph.current._add_op_generated(
rmo.MoScatterOp,
result=input.type,
input=input,
updates=updates,
indices=indices,
axis=builtin.IntegerAttr(builtin.IndexType(), axis),
output_param_decls=kgen.ParamDeclArrayAttr([]),
)[0].tensor
def scatter_nd(
input: TensorValueLike,
updates: TensorValueLike,
indices: TensorValueLike,
) -> TensorValue:
"""Scatters slices from ``updates`` into a copy of ``input`` at N-dimensional indices.
The last dimension of ``indices`` is the index vector. Its values select a
slice (or scalar) in ``input``. When the index vector length ``k`` is less
than ``input.rank``, each update writes a whole slice of the trailing
``input.rank - k`` dimensions.
.. code-block:: python
from max.dtype import DType
from max.engine import InferenceSession
from max.graph import DeviceRef, Graph, ops
device = DeviceRef.CPU()
with Graph("scatter_nd") as graph:
x = ops.constant(
[[1, 2], [3, 4], [5, 6]], DType.int32, device=device
)
updates = ops.constant(
[[10, 20], [50, 60]], DType.int32, device=device
)
indices = ops.constant([[0], [2]], DType.int64, device=device)
# Overwrite rows 0 and 2, producing [[10, 20], [3, 4], [50, 60]].
graph.output(ops.scatter_nd(x, updates, indices))
model = InferenceSession().load(graph)
result = model.execute()[0]
Args:
input: The input symbolic tensor to write elements to.
updates: A symbolic tensor of elements to write to ``input``. Its shape
is ``[*Q, *input.shape[K:]]``, matching the batch dimensions of
``indices`` and the trailing ``input`` dimensions that ``K``
doesn't index.
indices: A symbolic tensor of indices specifying where to write
``updates``. Its shape is ``[*Q, K]`` for any number of leading
batch dimensions ``Q`` and an index-vector length ``K`` in the last
dimension, where ``K <= input.rank``.
Returns:
A ``TensorValue`` representing ``input`` with ``updates`` scattered in.
It has the same shape and dtype as ``input``.
Raises:
ValueError: If the dtypes of ``input`` and ``updates`` mismatch, if
``indices`` dtype is not int32/int64, or if ``input``, ``updates``,
and ``indices`` aren't on the same device.
"""
input = TensorValue(input)
updates = TensorValue(updates)
indices = TensorValue(indices)
if input.dtype != updates.dtype:
raise ValueError(
f"The input dtype ({input.dtype}) and updates dtype"
f" ({updates.dtype}) must match"
)
if indices.dtype not in (DType.int32, DType.int64):
raise ValueError(
f"Invalid indices dtype: '{indices.dtype}'. Indices must be of type"
" int32 or int64."
)
assert_same_device(input=input, updates=updates, indices=indices)
return Graph.current._add_op_generated(
rmo.MoScatterNdOp,
input.type,
input,
updates,
indices,
kgen.ParamDeclArrayAttr([]),
)[0].tensor
def scatter_nd_add(
input: TensorValueLike,
updates: TensorValueLike,
indices: TensorValueLike,
) -> TensorValue:
"""Creates a new symbolic tensor by accumulating updates into input at N-D indices.
Produces an output tensor by scattering slices from updates into a copy
of input according to N-dimensional index vectors, summing values at
duplicate index positions. Each index vector is the last dimension of
``indices`` and selects a slice (or scalar) in input, so for an
``input`` of shape ``(4, 2)`` and ``indices`` of shape ``(3, 1)`` each
row ``i`` adds ``updates[i, :]`` to ``output[indices[i, 0], :]``. When
two index vectors point at the same row, both updates accumulate there.
The following example writes three row updates, two of which target
row ``0``, so that row receives the sum of both updates:
.. code-block:: python
import numpy as np
from max.driver import CPU
from max.dtype import DType
from max.engine import InferenceSession
from max.graph import DeviceRef, Graph, TensorType, ops
device = DeviceRef.CPU()
input_type = TensorType(DType.float32, [4, 2], device=device)
indices_type = TensorType(DType.int64, [3, 1], device=device)
with Graph(
"scatter_nd_add", input_types=[input_type, indices_type]
) as graph:
x = graph.inputs[0].tensor
indices = graph.inputs[1].tensor
updates = ops.constant(
np.array([[10.0, 10.0], [20.0, 20.0], [100.0, 100.0]],
dtype=np.float32),
DType.float32,
device=device,
)
graph.output(ops.scatter_nd_add(x, updates, indices))
model = InferenceSession(devices=[CPU()]).load(graph)
result = model.execute(
np.array([[1.0, 1.0], [2.0, 2.0], [3.0, 3.0], [4.0, 4.0]],
dtype=np.float32),
np.array([[0], [2], [0]], dtype=np.int64),
)[0]
# result: [[111.0, 111.0], [2.0, 2.0], [23.0, 23.0], [4.0, 4.0]]
.. invisible-code-block: python
np.testing.assert_allclose(
result.to_numpy(),
[[111.0, 111.0], [2.0, 2.0], [23.0, 23.0], [4.0, 4.0]],
)
Args:
input: The input symbolic tensor to accumulate into.
updates: A symbolic tensor of values to add.
indices: An index tensor whose last dimension is the index vector
length ``k`` (``k <= input.rank``).
Returns:
A ``TensorValue`` representing the updated tensor. It has the same
shape and dtype as ``input``.
Raises:
ValueError: If the dtypes of ``input`` and ``updates`` mismatch, if
``indices`` dtype is not int32/int64, or if ``input``, ``updates``,
and ``indices`` aren't on the same device.
"""
input = TensorValue(input)
updates = TensorValue(updates)
indices = TensorValue(indices)
if input.dtype != updates.dtype:
raise ValueError(
f"The input dtype ({input.dtype}) and updates dtype"
f" ({updates.dtype}) must match"
)
if indices.dtype not in (DType.int32, DType.int64):
raise ValueError(
f"Invalid indices dtype: '{indices.dtype}'."
" Indices must be of type int32 or int64."
)
assert_same_device(input=input, updates=updates, indices=indices)
return Graph.current._add_op_generated(
rmo.MoScatterNdAddOp,
input.type,
input,
updates,
indices,
kgen.ParamDeclArrayAttr([]),
)[0].tensor
def scatter_nd_max(
input: TensorValueLike,
updates: TensorValueLike,
indices: TensorValueLike,
) -> TensorValue:
"""Creates a new symbolic tensor by scattering the maximum of updates into input at N-D indices.
Produces an output tensor by scattering slices from updates into a copy
of input according to N-dimensional index vectors, keeping the maximum
at duplicate index positions. Each index vector is the last dimension of
``indices`` and selects a slice (or scalar) in input, so for an
``input`` of shape ``(4, 2)`` and ``indices`` of shape ``(3, 1)`` each
row ``i`` sets ``output[indices[i, 0], :]`` to the element-wise maximum
of its current value and ``updates[i, :]``. When two index vectors point
at the same row, that row keeps the maximum over both updates.
The following example writes three row updates, two of which target
row ``0``, so that row keeps the larger of the two updates:
.. code-block:: python
import numpy as np
from max.driver import CPU
from max.dtype import DType
from max.engine import InferenceSession
from max.graph import DeviceRef, Graph, TensorType, ops
device = DeviceRef.CPU()
input_type = TensorType(DType.float32, [4, 2], device=device)
indices_type = TensorType(DType.int64, [3, 1], device=device)
with Graph(
"scatter_nd_max", input_types=[input_type, indices_type]
) as graph:
x = graph.inputs[0].tensor
indices = graph.inputs[1].tensor
updates = ops.constant(
np.array([[10.0, 10.0], [20.0, 20.0], [100.0, 100.0]],
dtype=np.float32),
DType.float32,
device=device,
)
graph.output(ops.scatter_nd_max(x, updates, indices))
model = InferenceSession(devices=[CPU()]).load(graph)
result = model.execute(
np.array([[1.0, 1.0], [2.0, 2.0], [3.0, 3.0], [4.0, 4.0]],
dtype=np.float32),
np.array([[0], [2], [0]], dtype=np.int64),
)[0]
# result: [[100.0, 100.0], [2.0, 2.0], [20.0, 20.0], [4.0, 4.0]]
.. invisible-code-block: python
np.testing.assert_allclose(
result.to_numpy(),
[[100.0, 100.0], [2.0, 2.0], [20.0, 20.0], [4.0, 4.0]],
)
Args:
input: The input symbolic tensor to scatter into.
updates: A symbolic tensor of values to compare.
indices: An index tensor whose last dimension is the index vector
length ``k`` (``k <= input.rank``).
Returns:
A ``TensorValue`` representing the updated tensor. It has the same
shape and dtype as ``input``.
Raises:
ValueError: If the dtypes of ``input`` and ``updates`` mismatch, if
``indices`` dtype is not int32/int64, or if ``input``, ``updates``,
and ``indices`` aren't on the same device.
"""
input = TensorValue(input)
updates = TensorValue(updates)
indices = TensorValue(indices)
if input.dtype != updates.dtype:
raise ValueError(
f"The input dtype ({input.dtype}) and updates dtype"
f" ({updates.dtype}) must match"
)
if indices.dtype not in (DType.int32, DType.int64):
raise ValueError(
f"Invalid indices dtype: '{indices.dtype}'."
" Indices must be of type int32 or int64."
)
assert_same_device(input=input, updates=updates, indices=indices)
return Graph.current._add_op_generated(
rmo.MoScatterNdMaxOp,
input.type,
input,
updates,
indices,
kgen.ParamDeclArrayAttr([]),
)[0].tensor
def scatter_nd_min(
input: TensorValueLike,
updates: TensorValueLike,
indices: TensorValueLike,
) -> TensorValue:
"""Creates a new symbolic tensor by scattering the minimum of updates into input at N-D indices.
Produces an output tensor by scattering slices from updates into a copy
of input according to N-dimensional index vectors, keeping the minimum
at duplicate index positions. Each index vector is the last dimension of
``indices`` and selects a slice (or scalar) in input, so for an
``input`` of shape ``(4, 2)`` and ``indices`` of shape ``(3, 1)`` each
row ``i`` sets ``output[indices[i, 0], :]`` to the element-wise minimum
of its current value and ``updates[i, :]``. When two index vectors point
at the same row, that row keeps the minimum over both updates.
The following example writes updates whose values all exceed the input,
so every targeted row keeps its original smaller value:
.. code-block:: python
import numpy as np
from max.driver import CPU
from max.dtype import DType
from max.engine import InferenceSession
from max.graph import DeviceRef, Graph, TensorType, ops
device = DeviceRef.CPU()
input_type = TensorType(DType.float32, [4, 2], device=device)
indices_type = TensorType(DType.int64, [3, 1], device=device)
with Graph(
"scatter_nd_min", input_types=[input_type, indices_type]
) as graph:
x = graph.inputs[0].tensor
indices = graph.inputs[1].tensor
updates = ops.constant(
np.array([[10.0, 10.0], [20.0, 20.0], [100.0, 100.0]],
dtype=np.float32),
DType.float32,
device=device,
)
graph.output(ops.scatter_nd_min(x, updates, indices))
model = InferenceSession(devices=[CPU()]).load(graph)
result = model.execute(
np.array([[1.0, 1.0], [2.0, 2.0], [3.0, 3.0], [4.0, 4.0]],
dtype=np.float32),
np.array([[0], [2], [0]], dtype=np.int64),
)[0]
# result: [[1.0, 1.0], [2.0, 2.0], [3.0, 3.0], [4.0, 4.0]]
.. invisible-code-block: python
np.testing.assert_allclose(
result.to_numpy(),
[[1.0, 1.0], [2.0, 2.0], [3.0, 3.0], [4.0, 4.0]],
)
Args:
input: The input symbolic tensor to scatter into.
updates: A symbolic tensor of values to compare.
indices: An index tensor whose last dimension is the index vector
length ``k`` (``k <= input.rank``).
Returns:
A ``TensorValue`` representing the updated tensor. It has the same
shape and dtype as ``input``.
Raises:
ValueError: If the dtypes of ``input`` and ``updates`` mismatch, if
``indices`` dtype is not int32/int64, or if ``input``, ``updates``,
and ``indices`` aren't on the same device.
"""
input = TensorValue(input)
updates = TensorValue(updates)
indices = TensorValue(indices)
if input.dtype != updates.dtype:
raise ValueError(
f"The input dtype ({input.dtype}) and updates dtype"
f" ({updates.dtype}) must match"
)
if indices.dtype not in (DType.int32, DType.int64):
raise ValueError(
f"Invalid indices dtype: '{indices.dtype}'."
" Indices must be of type int32 or int64."
)
assert_same_device(input=input, updates=updates, indices=indices)
return Graph.current._add_op_generated(
rmo.MoScatterNdMinOp,
input.type,
input,
updates,
indices,
kgen.ParamDeclArrayAttr([]),
)[0].tensor
def scatter_nd_mul(
input: TensorValueLike,
updates: TensorValueLike,
indices: TensorValueLike,
) -> TensorValue:
"""Creates a new symbolic tensor by scattering the product of updates into input at N-D indices.
Produces an output tensor by scattering slices from updates into a copy
of input according to N-dimensional index vectors, multiplying values
at duplicate index positions. Each index vector is the last dimension of
``indices`` and selects a slice (or scalar) in input, so for an
``input`` of shape ``(4, 2)`` and ``indices`` of shape ``(3, 1)`` each
row ``i`` multiplies ``output[indices[i, 0], :]`` by ``updates[i, :]``.
When two index vectors point at the same row, both factors multiply into
that row.
The following example writes three row updates, two of which target
row ``0``, so that row is multiplied by both factors:
.. code-block:: python
import numpy as np
from max.driver import CPU
from max.dtype import DType
from max.engine import InferenceSession
from max.graph import DeviceRef, Graph, TensorType, ops
device = DeviceRef.CPU()
input_type = TensorType(DType.float32, [4, 2], device=device)
indices_type = TensorType(DType.int64, [3, 1], device=device)
with Graph(
"scatter_nd_mul", input_types=[input_type, indices_type]
) as graph:
x = graph.inputs[0].tensor
indices = graph.inputs[1].tensor
updates = ops.constant(
np.array([[10.0, 10.0], [20.0, 20.0], [100.0, 100.0]],
dtype=np.float32),
DType.float32,
device=device,
)
graph.output(ops.scatter_nd_mul(x, updates, indices))
model = InferenceSession(devices=[CPU()]).load(graph)
result = model.execute(
np.array([[1.0, 1.0], [2.0, 2.0], [3.0, 3.0], [4.0, 4.0]],
dtype=np.float32),
np.array([[0], [2], [0]], dtype=np.int64),
)[0]
# result: [[1000.0, 1000.0], [2.0, 2.0], [60.0, 60.0], [4.0, 4.0]]
.. invisible-code-block: python
np.testing.assert_allclose(
result.to_numpy(),
[[1000.0, 1000.0], [2.0, 2.0], [60.0, 60.0], [4.0, 4.0]],
)
Args:
input: The input symbolic tensor to scatter into.
updates: A symbolic tensor of values to multiply.
indices: An index tensor whose last dimension is the index vector
length ``k`` (``k <= input.rank``).
Returns:
A ``TensorValue`` representing the updated tensor. It has the same
shape and dtype as ``input``.
Raises:
ValueError: If the dtypes of ``input`` and ``updates`` mismatch, if
``indices`` dtype is not int32/int64, or if ``input``, ``updates``,
and ``indices`` aren't on the same device.
"""
input = TensorValue(input)
updates = TensorValue(updates)
indices = TensorValue(indices)
if input.dtype != updates.dtype:
raise ValueError(
f"The input dtype ({input.dtype}) and updates dtype"
f" ({updates.dtype}) must match"
)
if indices.dtype not in (DType.int32, DType.int64):
raise ValueError(
f"Invalid indices dtype: '{indices.dtype}'."
" Indices must be of type int32 or int64."
)
assert_same_device(input=input, updates=updates, indices=indices)
return Graph.current._add_op_generated(
rmo.MoScatterNdMulOp,
input.type,
input,
updates,
indices,
kgen.ParamDeclArrayAttr([]),
)[0].tensor
def scatter_add(
input: TensorValueLike,
updates: TensorValueLike,
indices: TensorValueLike,
axis: int = -1,
) -> TensorValue:
"""Creates a new symbolic tensor by accumulating updates into input at indices.
Produces an output tensor by scattering elements from updates into input
according to indices, summing values at duplicate indices. For a 2-D
input with ``axis=0``, each element adds as ``output[indices[i][j]][j]
+= updates[i][j]``; with ``axis=1`` the index selects the column, so
``output[i][indices[i][j]] += updates[i][j]``.
The following example adds the same updates along each axis of a
zero-filled input, where index collisions accumulate:
.. code-block:: python
import numpy as np
from max.driver import CPU
from max.dtype import DType
from max.engine import InferenceSession
from max.graph import DeviceRef, Graph, TensorType, ops
device = DeviceRef.CPU()
input_type = TensorType(DType.float32, [3, 3], device=device)
with Graph("scatter_add", input_types=[input_type]) as graph:
x = graph.inputs[0].tensor
updates = ops.constant(
np.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]],
dtype=np.float32),
DType.float32,
device=device,
)
indices = ops.constant(
np.array([[0, 1, 2], [0, 1, 0]], dtype=np.int64),
DType.int64,
device=device,
)
graph.output(
ops.scatter_add(x, updates, indices, axis=0),
ops.scatter_add(x, updates, indices, axis=1),
)
model = InferenceSession(devices=[CPU()]).load(graph)
along_rows, along_cols = model.execute(np.zeros((3, 3), np.float32))
# along_rows: [[5.0, 0.0, 6.0], [0.0, 7.0, 0.0], [0.0, 0.0, 3.0]]
# along_cols: [[1.0, 2.0, 3.0], [10.0, 5.0, 0.0], [0.0, 0.0, 0.0]]
.. invisible-code-block: python
np.testing.assert_allclose(
along_rows.to_numpy(),
[[5.0, 0.0, 6.0], [0.0, 7.0, 0.0], [0.0, 0.0, 3.0]],
)
np.testing.assert_allclose(
along_cols.to_numpy(),
[[1.0, 2.0, 3.0], [10.0, 5.0, 0.0], [0.0, 0.0, 0.0]],
)
Args:
input: The input symbolic tensor to accumulate into.
updates: A symbolic tensor of values to add.
indices: The positions in input to update.
axis: The axis along which indices indexes into.
Returns:
A ``TensorValue`` representing the updated tensor. It has the same
shape and dtype as ``input``.
Raises:
ValueError: If ``axis`` is out of range, if dtypes mismatch, if
``indices`` dtype is not int32/int64, or if ``input``, ``updates``,
and ``indices`` aren't on the same device.
Error: If ``input``, ``updates``, and ``indices`` don't have equal
rank, if ``updates.shape`` and ``indices.shape`` differ, or if any
``indices`` dimension exceeds the corresponding ``input``
dimension.
"""
input = TensorValue(input)
if not (-input.rank <= axis < input.rank):
raise ValueError(
f"Invalid axis value {axis}. Axis must be in range"
f" [-{input.rank}, {input.rank - 1}]"
)
updates = TensorValue(updates)
if input.dtype != updates.dtype:
raise ValueError(
f"The input dtype '{input.dtype}' and updates dtype"
f" '{updates.dtype}' must match."
)
indices = TensorValue(indices)
if indices.dtype not in [DType.int32, DType.int64]:
raise ValueError(
f"Invalid indices dtype: '{indices.dtype}'."
" Indices must be of type int32 or int64."
)
assert_same_device(input=input, updates=updates, indices=indices)
axis_constant = constant(axis, DType.int64, DeviceRef.CPU())
return Graph.current._add_op_generated(
rmo.MoScatterAddOp,
input.type,
input,
updates,
indices,
axis_constant,
kgen.ParamDeclArrayAttr([]),
)[0].tensor
def scatter_max(
input: TensorValueLike,
updates: TensorValueLike,
indices: TensorValueLike,
axis: int = -1,
) -> TensorValue:
"""Creates a new symbolic tensor by scattering the maximum of updates into input.
Produces an output tensor by scattering elements from updates into input
according to indices, keeping the maximum at duplicate indices. For a 2-D
input with ``axis=0``, each element updates as ``output[indices[i][j]][j]
= max(output[indices[i][j]][j], updates[i][j])``; with ``axis=1`` the
index selects the column, so ``output[i][indices[i][j]] =
max(output[i][indices[i][j]], updates[i][j])``.
The following example scatters the maximum along each axis of a
one-filled input, so a cell changes only where an update exceeds ``1``:
.. code-block:: python
import numpy as np
from max.driver import CPU
from max.dtype import DType
from max.engine import InferenceSession
from max.graph import DeviceRef, Graph, TensorType, ops
device = DeviceRef.CPU()
input_type = TensorType(DType.float32, [3, 3], device=device)
with Graph("scatter_max", input_types=[input_type]) as graph:
x = graph.inputs[0].tensor
updates = ops.constant(
np.array([[5.0, 2.0, 3.0], [4.0, 7.0, 6.0]],
dtype=np.float32),
DType.float32,
device=device,
)
indices = ops.constant(
np.array([[0, 1, 2], [0, 1, 0]], dtype=np.int64),
DType.int64,
device=device,
)
graph.output(
ops.scatter_max(x, updates, indices, axis=0),
ops.scatter_max(x, updates, indices, axis=1),
)
model = InferenceSession(devices=[CPU()]).load(graph)
along_rows, along_cols = model.execute(np.ones((3, 3), np.float32))
# along_rows: [[5.0, 1.0, 6.0], [1.0, 7.0, 1.0], [1.0, 1.0, 3.0]]
# along_cols: [[5.0, 2.0, 3.0], [6.0, 7.0, 1.0], [1.0, 1.0, 1.0]]
.. invisible-code-block: python
np.testing.assert_allclose(
along_rows.to_numpy(),
[[5.0, 1.0, 6.0], [1.0, 7.0, 1.0], [1.0, 1.0, 3.0]],
)
np.testing.assert_allclose(
along_cols.to_numpy(),
[[5.0, 2.0, 3.0], [6.0, 7.0, 1.0], [1.0, 1.0, 1.0]],
)
Args:
input: The input symbolic tensor to scatter into.
updates: A symbolic tensor of values to compare.
indices: The positions in input to update.
axis: The axis along which indices indexes into.
Returns:
A ``TensorValue`` representing the updated tensor. It has the same
shape and dtype as ``input``.
Raises:
ValueError: If ``axis`` is out of range, if dtypes mismatch, if
``indices`` dtype is not int32/int64, or if ``input``, ``updates``,
and ``indices`` aren't on the same device.
Error: If ``input``, ``updates``, and ``indices`` don't have equal
rank, if ``updates.shape`` and ``indices.shape`` differ, or if any
``indices`` dimension exceeds the corresponding ``input``
dimension.
"""
input = TensorValue(input)
if not (-input.rank <= axis < input.rank):
raise ValueError(
f"Invalid axis value {axis}. Axis must be in range"
f" [-{input.rank}, {input.rank - 1}]"
)
updates = TensorValue(updates)
if input.dtype != updates.dtype:
raise ValueError(
f"The input dtype '{input.dtype}' and updates dtype"
f" '{updates.dtype}' must match."
)
indices = TensorValue(indices)
if indices.dtype not in [DType.int32, DType.int64]:
raise ValueError(
f"Invalid indices dtype: '{indices.dtype}'."
" Indices must be of type int32 or int64."
)
assert_same_device(input=input, updates=updates, indices=indices)
axis_constant = constant(axis, DType.int64, DeviceRef.CPU())
return Graph.current._add_op_generated(
rmo.MoScatterMaxOp,
input.type,
input,
updates,
indices,
axis_constant,
kgen.ParamDeclArrayAttr([]),
)[0].tensor
def scatter_min(
input: TensorValueLike,
updates: TensorValueLike,
indices: TensorValueLike,
axis: int = -1,
) -> TensorValue:
"""Creates a new symbolic tensor by scattering the minimum of updates into input.
Produces an output tensor by scattering elements from updates into input
according to indices, keeping the minimum at duplicate indices. For a 2-D
input with ``axis=0``, each element updates as ``output[indices[i][j]][j]
= min(output[indices[i][j]][j], updates[i][j])``; with ``axis=1`` the
index selects the column, so ``output[i][indices[i][j]] =
min(output[i][indices[i][j]], updates[i][j])``.
The following example scatters the minimum along each axis of an input
filled with ``9``, so a cell changes only where an update is smaller:
.. code-block:: python
import numpy as np
from max.driver import CPU
from max.dtype import DType
from max.engine import InferenceSession
from max.graph import DeviceRef, Graph, TensorType, ops
device = DeviceRef.CPU()
input_type = TensorType(DType.float32, [3, 3], device=device)
with Graph("scatter_min", input_types=[input_type]) as graph:
x = graph.inputs[0].tensor
updates = ops.constant(
np.array([[5.0, 2.0, 3.0], [4.0, 7.0, 6.0]],
dtype=np.float32),
DType.float32,
device=device,
)
indices = ops.constant(
np.array([[0, 1, 2], [0, 1, 0]], dtype=np.int64),
DType.int64,
device=device,
)
graph.output(
ops.scatter_min(x, updates, indices, axis=0),
ops.scatter_min(x, updates, indices, axis=1),
)
model = InferenceSession(devices=[CPU()]).load(graph)
along_rows, along_cols = model.execute(
np.full((3, 3), 9.0, np.float32)
)
# along_rows: [[4.0, 9.0, 6.0], [9.0, 2.0, 9.0], [9.0, 9.0, 3.0]]
# along_cols: [[5.0, 2.0, 3.0], [4.0, 7.0, 9.0], [9.0, 9.0, 9.0]]
.. invisible-code-block: python
np.testing.assert_allclose(
along_rows.to_numpy(),
[[4.0, 9.0, 6.0], [9.0, 2.0, 9.0], [9.0, 9.0, 3.0]],
)
np.testing.assert_allclose(
along_cols.to_numpy(),
[[5.0, 2.0, 3.0], [4.0, 7.0, 9.0], [9.0, 9.0, 9.0]],
)
Args:
input: The input symbolic tensor to scatter into.
updates: A symbolic tensor of values to compare.
indices: The positions in input to update.
axis: The axis along which indices indexes into.
Returns:
A ``TensorValue`` representing the updated tensor. It has the same
shape and dtype as ``input``.
Raises:
ValueError: If ``axis`` is out of range, if dtypes mismatch, if
``indices`` dtype is not int32/int64, or if ``input``, ``updates``,
and ``indices`` aren't on the same device.
Error: If ``input``, ``updates``, and ``indices`` don't have equal
rank, if ``updates.shape`` and ``indices.shape`` differ, or if any
``indices`` dimension exceeds the corresponding ``input``
dimension.
"""
input = TensorValue(input)
if not (-input.rank <= axis < input.rank):
raise ValueError(
f"Invalid axis value {axis}. Axis must be in range"
f" [-{input.rank}, {input.rank - 1}]"
)
updates = TensorValue(updates)
if input.dtype != updates.dtype:
raise ValueError(
f"The input dtype '{input.dtype}' and updates dtype"
f" '{updates.dtype}' must match."
)
indices = TensorValue(indices)
if indices.dtype not in [DType.int32, DType.int64]:
raise ValueError(
f"Invalid indices dtype: '{indices.dtype}'."
" Indices must be of type int32 or int64."
)
assert_same_device(input=input, updates=updates, indices=indices)
axis_constant = constant(axis, DType.int64, DeviceRef.CPU())
return Graph.current._add_op_generated(
rmo.MoScatterMinOp,
input.type,
input,
updates,
indices,
axis_constant,
kgen.ParamDeclArrayAttr([]),
)[0].tensor
def scatter_mul(
input: TensorValueLike,
updates: TensorValueLike,
indices: TensorValueLike,
axis: int = -1,
) -> TensorValue:
"""Creates a new symbolic tensor by scattering the product of updates into input.
Produces an output tensor by scattering elements from updates into input
according to indices, multiplying values at duplicate indices. For a 2-D
input with ``axis=0``, each element updates as ``output[indices[i][j]][j]
*= updates[i][j]``; with ``axis=1`` the index selects the column, so
``output[i][indices[i][j]] *= updates[i][j]``.
The following example multiplies into each axis of a one-filled input,
where cells hit twice along an axis accumulate both factors:
.. code-block:: python
import numpy as np
from max.driver import CPU
from max.dtype import DType
from max.engine import InferenceSession
from max.graph import DeviceRef, Graph, TensorType, ops
device = DeviceRef.CPU()
input_type = TensorType(DType.float32, [3, 3], device=device)
with Graph("scatter_mul", input_types=[input_type]) as graph:
x = graph.inputs[0].tensor
updates = ops.constant(
np.array([[5.0, 2.0, 3.0], [4.0, 7.0, 6.0]],
dtype=np.float32),
DType.float32,
device=device,
)
indices = ops.constant(
np.array([[0, 1, 2], [0, 1, 0]], dtype=np.int64),
DType.int64,
device=device,
)
graph.output(
ops.scatter_mul(x, updates, indices, axis=0),
ops.scatter_mul(x, updates, indices, axis=1),
)
model = InferenceSession(devices=[CPU()]).load(graph)
along_rows, along_cols = model.execute(np.ones((3, 3), np.float32))
# along_rows: [[20.0, 1.0, 6.0], [1.0, 14.0, 1.0], [1.0, 1.0, 3.0]]
# along_cols: [[5.0, 2.0, 3.0], [24.0, 7.0, 1.0], [1.0, 1.0, 1.0]]
.. invisible-code-block: python
np.testing.assert_allclose(
along_rows.to_numpy(),
[[20.0, 1.0, 6.0], [1.0, 14.0, 1.0], [1.0, 1.0, 3.0]],
)
np.testing.assert_allclose(
along_cols.to_numpy(),
[[5.0, 2.0, 3.0], [24.0, 7.0, 1.0], [1.0, 1.0, 1.0]],