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
|
/* Copyright 2017 The TensorFlow Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
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.
==============================================================================*/
// Kernel registrations for XLA JIT devices.
#include "tensorflow/compiler/tf2xla/xla_compilation_device.h"
#include "tensorflow/core/framework/op_kernel.h"
namespace tensorflow {
namespace {
// CPU JIT device registrations.
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("_Arg").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("_ArrayToList"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("_ListToArray"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("_Retval").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Abs").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Add").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("AddN").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("All"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("Any"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("AvgPool").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("AvgPoolGrad").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("BatchMatMul").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("BiasAdd").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("BiasAddV1").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("BiasAddGrad").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("BroadcastGradientArgs"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Cast")
.TypeConstraint("SrcT", kCpuAllTypes)
.TypeConstraint("DstT", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Ceil").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Concat").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("ConcatV2")
.TypeConstraint("T", kCpuAllTypes)
.TypeConstraint("Tidx", DT_INT32));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("ConcatOffset"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Conv2D").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(
DEVICE_CPU_XLA_JIT,
Name("Conv2DBackpropFilter").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(
DEVICE_CPU_XLA_JIT,
Name("Conv2DBackpropInput").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(
DEVICE_CPU_XLA_JIT,
Name("DepthwiseConv2dNative").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Diag").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("DiagPart").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Div").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("DynamicStitch").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Equal").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Exp").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("ExpandDims").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Fill").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Floor").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("FloorDiv").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("FloorMod").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Greater").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("GreaterEqual").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Inv").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Reciprocal").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("InvertPermutation").TypeConstraint("T", DT_INT32));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("L2Loss").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Less").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("LessEqual").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("LinSpace").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Log").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("LogicalAnd"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("LogicalNot"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("LogicalOr"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("LogSoftmax").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("LRN").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("LRNGrad").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Maximum").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("MatMul").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("MatrixDiag").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("MatrixDiagPart").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Max").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("MaxPool").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("MaxPoolGrad").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Mean").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Min").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Minimum").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Mod").TypeConstraint("T", kCpuIntTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Mul").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Neg").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("NotEqual").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Pack").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Pad").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Pow").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("PreventGradient").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Prod").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Range").TypeConstraint("Tidx", kCpuNumericTypes));
// TODO(b/31361304): disabled because of XLA bugs.
// REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("RandomStandardNormal"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("RandomUniform"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("RandomUniformInt"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("Rank"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("RealDiv").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Relu").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Relu6").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("ReluGrad").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Relu6Grad").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Reshape").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Rsqrt").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("RsqrtGrad").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Select").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("Shape"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("ShapeN"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Sigmoid").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("SigmoidGrad").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Sign").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("Size"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Slice").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Softmax").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(
DEVICE_CPU_XLA_JIT,
Name("SoftmaxCrossEntropyWithLogits").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Softplus").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("SoftplusGrad").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("SparseMatMul")
.TypeConstraint("Ta", kCpuFloatTypes)
.TypeConstraint("Tb", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Split").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("SplitV").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Square").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(
DEVICE_CPU_XLA_JIT,
Name("SquaredDifference").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Squeeze").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Sqrt").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("StopGradient").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("StridedSlice").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("StridedSliceGrad").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Sub").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Sum").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT, Name("SymbolicGradient"));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Tanh").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("TanhGrad").TypeConstraint("T", kCpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Tile").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Transpose").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("TruncateDiv").TypeConstraint("T", kCpuIntTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("TruncateMod").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Unpack").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_CPU_XLA_JIT,
Name("ZerosLike").TypeConstraint("T", kCpuNumericTypes));
REGISTER_XLA_JIT_ONLY_KERNEL(DEVICE_CPU_XLA_JIT,
Name("Const").TypeConstraint("dtype",
kCpuAllTypes));
REGISTER_XLA_JIT_ONLY_KERNEL(
DEVICE_CPU_XLA_JIT, Name("Identity").TypeConstraint("T", kCpuAllTypes));
REGISTER_XLA_JIT_ONLY_KERNEL(DEVICE_CPU_XLA_JIT, Name("NoOp"));
// GPU JIT device registrations
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("_Arg").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("_ArrayToList"));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("_ListToArray"));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("_Retval").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Abs").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Add").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("AddN").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("All"));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("Any"));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("AvgPool").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("AvgPoolGrad").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("BatchMatMul").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("BiasAdd").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("BiasAddV1").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("BiasAddGrad").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("BroadcastGradientArgs"));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Cast")
.TypeConstraint("SrcT", kGpuAllTypes)
.TypeConstraint("DstT", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Ceil").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Concat").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("ConcatV2").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("ConcatOffset"));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Conv2D").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(
DEVICE_GPU_XLA_JIT,
Name("Conv2DBackpropFilter").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(
DEVICE_GPU_XLA_JIT,
Name("Conv2DBackpropInput").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(
DEVICE_GPU_XLA_JIT,
Name("DepthwiseConv2dNative").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Diag").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("DiagPart").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Div").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("DynamicStitch").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Equal").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Exp").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("ExpandDims").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Fill").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Floor").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("FloorDiv").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("FloorMod").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Greater").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("GreaterEqual").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Inv").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Reciprocal").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("InvertPermutation").TypeConstraint("T", DT_INT32));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("L2Loss").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Less").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("LessEqual").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("LinSpace").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Log").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("LogicalAnd"));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("LogicalNot"));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("LogicalOr"));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("LogSoftmax").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("LRN").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("LRNGrad").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Maximum").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("MatMul").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("MatrixDiag").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("MatrixDiagPart").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Max").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("MaxPool").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("MaxPoolGrad").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Mean").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Min").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Minimum").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Mod").TypeConstraint("T", kGpuIntTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Mul").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Neg").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("NotEqual").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Pack").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Pad").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Pow").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("PreventGradient").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Prod").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Range").TypeConstraint("Tidx", kGpuNumericTypes));
// TODO(b/31361304): disabled because of XLA bugs.
// REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("RandomStandardNormal"));
// REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("RandomUniform"));
// REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("RandomUniformInt"));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("Rank"));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("RealDiv").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Relu").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Relu6").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("ReluGrad").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Relu6Grad").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Reshape").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Rsqrt").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("RsqrtGrad").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Select").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("Shape"));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("ShapeN"));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Sigmoid").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("SigmoidGrad").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Sign").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("Size"));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Slice").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Softmax").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(
DEVICE_GPU_XLA_JIT,
Name("SoftmaxCrossEntropyWithLogits").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Softplus").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("SoftplusGrad").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("SparseMatMul")
.TypeConstraint("Ta", kGpuFloatTypes)
.TypeConstraint("Tb", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Split").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("SplitV").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Square").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(
DEVICE_GPU_XLA_JIT,
Name("SquaredDifference").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Squeeze").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Sqrt").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("StopGradient").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("StridedSlice").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("StridedSliceGrad").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Sub").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Sum").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT, Name("SymbolicGradient"));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Tanh").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("TanhGrad").TypeConstraint("T", kGpuFloatTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Tile").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Transpose").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("TruncateDiv").TypeConstraint("T", kGpuIntTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("TruncateMod").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Unpack").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_KERNEL(DEVICE_GPU_XLA_JIT,
Name("ZerosLike").TypeConstraint("T", kGpuNumericTypes));
REGISTER_XLA_JIT_ONLY_KERNEL(DEVICE_GPU_XLA_JIT,
Name("Const").TypeConstraint("dtype",
kGpuAllTypes));
REGISTER_XLA_JIT_ONLY_KERNEL(
DEVICE_GPU_XLA_JIT, Name("Identity").TypeConstraint("T", kGpuAllTypes));
REGISTER_XLA_JIT_ONLY_KERNEL(DEVICE_GPU_XLA_JIT, Name("NoOp"));
} // anonymous namespace
} // namespace tensorflow
|
