Gitiles
Code Review Sign In
review.mlplatform.org / ml / ComputeLibrary / 0ec58215bf5322d370dbc7c0a7f3ced05af2174f / . / docs / 09_operators_list.dox
blob: 244f292f829d2fbc4810b7597a7f7c4738e975f5 [file] [log] [blame]
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]1///
2/// Copyright (c) 2021 Arm Limited.
3///
4/// SPDX-License-Identifier: MIT
5///
6/// Permission is hereby granted, free of charge, to any person obtaining a copy
7/// of this software and associated documentation files (the "Software"), to
8/// deal in the Software without restriction, including without limitation the
9/// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10/// sell copies of the Software, and to permit persons to whom the Software is
11/// furnished to do so, subject to the following conditions:
12///
13/// The above copyright notice and this permission notice shall be included in all
14/// copies or substantial portions of the Software.
15///
16/// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17/// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18/// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19/// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20/// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21/// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22/// SOFTWARE.
23///
24namespace arm_compute
25{
26/**
27@page operators_list Supported Operators
28
29@tableofcontents
30
31@section S9_1_operators_list Supported Operators
32
33Compute Library supports operators that are listed in below table.
34
35Compute Library supports a wide list of data-types, information can been directly found in the documentation of each kernel/function.
36The main data-types that the Machine Learning functions support are the following:
37 <ul>
38 <li>BFLOAT16: 16-bit non-standard brain floating point
39 <li>QASYMM8: 8-bit unsigned asymmetric quantized
40 <li>QASYMM8_SIGNED: 8-bit signed asymmetric quantized
41 <li>QSYMM8_PER_CHANNEL: 8-bit signed symmetric quantized (Used for the weights)
42 <li>QSYMM8: 8-bit unsigned symmetric quantized
43 <li>QSYMM16: 16-bit unsigned symmetric quantized
44 <li>F32: 32-bit single precision floating point
45 <li>F16: 16-bit half precision floating point
46 <li>S32: 32-bit signed integer
47 <li>U8: 8-bit unsigned char
48 <li>All: include all above data types
49 </ul>
50
51Compute Library supports the following data layouts (fast changing dimension from right to left):
52 <ul>
53 <li>NHWC: The native layout of Compute Library that delivers the best performance where channels are in the fastest changing dimension
54 <li>NCHW: Legacy layout where width is in the fastest changing dimension
55 <li>All: include all above data layouts
56 </ul>
57where N = batches, C = channels, H = height, W = width
58
59<table>
60<caption id="multi_row"></caption>
61<tr>
62 <th>Function
63 <th>Description
64 <th>Equivalent Android NNAPI Op
65 <th>Backends
66 <th>Data Layouts
67 <th>Data Types
68<tr>
69 <td rowspan="2">ActivationLayer
70 <td rowspan="2" style="width:200px;"> Function to simulate an activation layer with the specified activation function.
71 <td rowspan="2">
72 <ul>
73 <li>ANEURALNETWORKS_ELU
74 <li>ANEURALNETWORKS_HARD_SWISH
75 <li>ANEURALNETWORKS_LOGISTIC
76 <li>ANEURALNETWORKS_RELU
77 <li>ANEURALNETWORKS_RELU1
78 <li>ANEURALNETWORKS_RELU6
79 <li>ANEURALNETWORKS_TANH
80 </ul>
81 <td>NEActivationLayer
82 <td>
83 <ul>
84 <li>All
85 </ul>
86 <td>
87 <table>
88 <tr><th>src<th>dst
89 <tr><td>QASYMM8<td>QASYMM8
90 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
91 <tr><td>QSYMM16<td>QSYMM16
92 <tr><td>F16<td>F16
93 <tr><td>F32<td>F32
94 </table>
95<tr>
96 <td>CLActivationLayer
97 <td>
98 <ul>
99 <li>All
100 </ul>
101 <td>
102 <table>
103 <tr><th>src<th>dst
104 <tr><td>QASYMM8<td>QASYMM8
105 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
106 <tr><td>QSYMM16<td>QSYMM16
107 <tr><td>F16<td>F16
108 <tr><td>F32<td>F32
109 </table>
110<tr>
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]111 <td rowspan="2">ArgMinMaxLayer
112 <td rowspan="2" style="width:200px;"> Function to calculate the index of the minimum or maximum values in a tensor based on an axis.
113 <td rowspan="2">
114 <ul>
115 <li>ANEURALNETWORKS_ARGMAX
116 <li>ANEURALNETWORKS_ARGMIN
117 </ul>
118 <td>NEArgMinMaxLayer
119 <td>
120 <ul>
121 <li>All
122 </ul>
123 <td>
124 <table>
125 <tr><th>src<th>dst
126 <tr><td>QASYMM8<td>U32, S32
127 <tr><td>QASYMM8_SIGNED<td>U32, S32
128 <tr><td>S32<td>U32, S32
129 <tr><td>F16<td>U32, S32
130 <tr><td>F32<td>U32, S32
131 </table>
132<tr>
133 <td>CLArgMinMaxLayer
134 <td>
135 <ul>
136 <li>All
137 </ul>
138 <td>
139 <table>
140 <tr><th>src<th>dst
141 <tr><td>QASYMM8<td>U32, S32
142 <tr><td>QASYMM8_SIGNED<td>U32, S32
143 <tr><td>S32<td>U32, S32
144 <tr><td>F16<td>U32, S32
145 <tr><td>F32<td>U32, S32
146 </table>
147<tr>
148 <td rowspan="2">BatchNormalizationLayer
149 <td rowspan="2" style="width:200px;"> Function to perform batch normalization.
150 <td rowspan="2">
151 <ul>
152 <li>n/a
153 </ul>
154 <td>NEBatchNormalizationLayer
155 <td>
156 <ul>
157 <li>NHWC
158 <li>NCHW
159 </ul>
160 <td>
161 <table>
162 <tr><th>src<th>dst
163 <tr><td>F32<td>F32
164 <tr><td>F16<td>F16
165 </table>
166<tr>
167 <td>CLBatchNormalizationLayer
168 <td>
169 <ul>
170 <li>NHWC
171 <li>NCHW
172 </ul>
173 <td>
174 <table>
175 <tr><th>src<th>dst
176 <tr><td>F32<td>F32
177 <tr><td>F16<td>F16
178 </table>
179<tr>
180 <td rowspan="2">BatchToSpaceLayer
181 <td rowspan="2" style="width:200px;"> Batch to space transformation.
182 <td rowspan="2">
183 <ul>
184 <li>ANEURALNETWORKS_BATCH_TO_SPACE_ND
185 </ul>
186 <td>NEBatchToSpaceLayer
187 <td>
188 <ul>
189 <li>NHWC
190 <li>NCHW
191 </ul>
192 <td>
193 <table>
194 <tr><th>src0<th>src1<th>dst
195 <tr><td>All<td>s32<td>All
196 </table>
197<tr>
198 <td>CLBatchToSpaceLayer
199 <td>
200 <ul>
201 <li>NHWC
202 <li>NCHW
203 </ul>
204 <td>
205 <table>
206 <tr><th>src0<th>src1<th>dst
207 <tr><td>All<td>s32<td>All
208 </table>
209<tr>
210 <td rowspan="2">BitwiseAnd
211 <td rowspan="2" style="width:200px;"> Function to performe bitwise AND between 2 tensors.
212 <td rowspan="2">
213 <ul>
214 <li>ANEURALNETWORKS_LOGICAL_AND
215 </ul>
216 <td>NEBitwiseAnd
217 <td>
218 <ul>
219 <li>All
220 </ul>
221 <td>
222 <table>
223 <tr><th>src<th>dst
224 <tr><td>U8<td>U8
225 </table>
226<tr>
227 <td>CLBitwiseAnd
228 <td>
229 <ul>
230 <li>All
231 </ul>
232 <td>
233 <table>
234 <tr><th>src<th>dst
235 <tr><td>U8<td>U8
236 </table>
237<tr>
238 <td rowspan="2">BitwiseNot
239 <td rowspan="2" style="width:200px;"> Function to performe bitwise NOT.
240 <td rowspan="2">
241 <ul>
242 <li>ANEURALNETWORKS_LOGICAL_NOT
243 </ul>
244 <td>NEBitwiseNot
245 <td>
246 <ul>
247 <li>All
248 </ul>
249 <td>
250 <table>
251 <tr><th>src<th>dst
252 <tr><td>U8<td>U8
253 </table>
254<tr>
255 <td>CLBitwiseNot
256 <td>
257 <ul>
258 <li>All
259 </ul>
260 <td>
261 <table>
262 <tr><th>src<th>dst
263 <tr><td>U8<td>U8
264 </table>
265<tr>
266 <td rowspan="2">BitwiseOr
267 <td rowspan="2" style="width:200px;"> Function to performe bitwise OR between 2 tensors.
268 <td rowspan="2">
269 <ul>
270 <li>ANEURALNETWORKS_LOGICAL_OR
271 </ul>
272 <td>NEBitwiseOr
273 <td>
274 <ul>
275 <li>All
276 </ul>
277 <td>
278 <table>
279 <tr><th>src<th>dst
280 <tr><td>U8<td>U8
281 </table>
282<tr>
283 <td>CLBitwiseOr
284 <td>
285 <ul>
286 <li>All
287 </ul>
288 <td>
289 <table>
290 <tr><th>src<th>dst
291 <tr><td>U8<td>U8
292 </table>
293<tr>
294 <td rowspan="2">BitwiseXor
295 <td rowspan="2" style="width:200px;"> Function to performe bitwise XOR between 2 tensors.
296 <td rowspan="2">
297 <ul>
298 <li>n/a
299 </ul>
300 <td>NEBitwiseXor
301 <td>
302 <ul>
303 <li>All
304 </ul>
305 <td>
306 <table>
307 <tr><th>src<th>dst
308 <tr><td>U8<td>U8
309 </table>
310<tr>
311 <td>CLBitwiseXor
312 <td>
313 <ul>
314 <li>All
315 </ul>
316 <td>
317 <table>
318 <tr><th>src<th>dst
319 <tr><td>U8<td>U8
320 </table>
321<tr>
322 <td rowspan="2">BoundingBoxTransform
323 <td rowspan="2" style="width:200px;"> Transform proposal bounding boxes to target bounding box using bounding box deltas.
324 <td rowspan="2">
325 <ul>
326 <li>n/a
327 </ul>
328 <td>NEBoundingBoxTransform
329 <td>
330 <ul>
331 <li>NHWC
332 <li>NCHW
333 </ul>
334 <td>
335 <table>
336 <tr><th>src0<th>src1<th>dst
337 <tr><td>QASYMM16<td>QASYMM8<td>QASYMM16
338 <tr><td>F16<td>F16<td>F16
339 <tr><td>F32<td>F32<td>F32
340 </table>
341<tr>
342 <td>CLBoundingBoxTransform
343 <td>
344 <ul>
345 <li>NHWC
346 <li>NCHW
347 </ul>
348 <td>
349 <table>
350 <tr><th>src0<th>src1<th>dst
351 <tr><td>QASYMM16<td>QASYMM8<td>QASYMM16
352 <tr><td>F16<td>F16<td>F16
353 <tr><td>F32<td>F32<td>F32
354 </table>
355<tr>
356 <td rowspan="2">Cast
357 <td rowspan="2" style="width:200px;"> Function to cast a tensor.
358 <td rowspan="2">
359 <ul>
360 <li>ANEURALNETWORKS_CAST
361 </ul>
362 <td>NECast
363 <td>
364 <ul>
365 <li>All
366 </ul>
367 <td>
368 <table>
369 <tr><th>src<th>dst
370 <tr><td>QASYMM8_SIGNED<td>S16, S32, F32, F16
371 <tr><td>QASYMM8<td>U16, S16, S32, F32, F16
372 <tr><td>U8<td>U16, S16, S32, F32, F16
373 <tr><td>U16<td>U8, U32
374 <tr><td>S16<td>QASYMM8_SIGNED, U8, S32
375 <tr><td>F16<td>QASYMM8_SIGNED, QASYMM8, F32, S32, U8
376 <tr><td>S32<td>QASYMM8_SIGNED, QASYMM8, F16, F32, U8
377 <tr><td>F32<td>QASYMM8_SIGNED, QASYMM8, BFLOAT16, F16, S32, U8
378 </table>
379<tr>
380 <td>CLCast
381 <td>
382 <ul>
383 <li>All
384 </ul>
385 <td>
386 <table>
387 <tr><th>src<th>dst
388 <tr><td>U8<td>S8, U16, S16, U32, S32, F16, F32
389 <tr><td>U16<td>U8, S8, S16, U32, S32, F16, F32
390 <tr><td>S16<td>U8, S8, U16, U32, S32, F16, F32
391 <tr><td>U32<td>U8, S8, U16, S16, S32, F16, F32
392 <tr><td>S32<td>U8, S8, U16, S16, U32, F16, F32
393 <tr><td>F16<td>U8, S8, U16, S16, U32, F32
394 <tr><td>F32<td>U8, S8, U16, S16, U32, F16
395 </table>
396<tr>
397 <td rowspan="2">ChannelShuffleLayer
398 <td rowspan="2" style="width:200px;"> Function to shuffle the channels of the input tensor.
399 <td rowspan="2">
400 <ul>
401 <li>ANEURALNETWORKS_CHANNEL_SHUFFLE
402 </ul>
403 <td>NEChannelShuffleLayer
404 <td>
405 <ul>
406 <li>NCHW
407 </ul>
408 <td>
409 <table>
410 <tr><th>src<th>dst
411 <tr><td>All<td>All
412 </table>
413<tr>
414 <td>CLChannelShuffleLayer
415 <td>
416 <ul>
417 <li>NCHW
418 </ul>
419 <td>
420 <table>
421 <tr><th>src<th>dst
422 <tr><td>All<td>All
423 </table>
424<tr>
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]425 <td rowspan="2">ConcatenateLayer
426 <td rowspan="2" style="width:200px;"> Function to concatenate tensors along a given axis.
427 <td rowspan="2">
428 <ul>
429 <li>ANEURALNETWORKS_CONCATENATION
430 </ul>
431 <td>NEConcatenateLayer
432 <td>
433 <ul>
434 <li>All
435 </ul>
436 <td>
437 <table>
438 <tr><th>src<th>dst
439 <tr><td>QASYMM8<td>QASYMM8
440 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
441 <tr><td>F16<td>F16
442 <tr><td>F32<td>F32
443 </table>
444<tr>
445 <td>CLConcatenateLayer
446 <td>
447 <ul>
448 <li>All
449 </ul>
450 <td>
451 <table>
452 <tr><th>src<th>dst
453 <tr><td>QASYMM8<td>QASYMM8
454 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
455 <tr><td>F16<td>F16
456 <tr><td>F32<td>F32
457 </table>
458<tr>
459 <td rowspan="2">ConvertFullyConnectedWeights
460 <td rowspan="2" style="width:200px;"> Function to tranpose the wieghts for the fully connected layer.
461 <td rowspan="2">
462 <ul>
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]463 <li>n/a
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]464 </ul>
465 <td>NEConvertFullyConnectedWeights
466 <td>
467 <ul>
468 <li>NHWC
469 <li>NCHW
470 </ul>
471 <td>
472 <table>
473 <tr><th>src<th>dst
474 <tr><td>All<td>All
475 </table>
476<tr>
477 <td>CLConvertFullyConnectedWeights
478 <td>
479 <ul>
480 <li>NHWC
481 <li>NCHW
482 </ul>
483 <td>
484 <table>
485 <tr><th>src<th>dst
486 <tr><td>All<td>All
487 </table>
488<tr>
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]489 <td rowspan="2">ConvolutionLayer
490 <td rowspan="2" style="width:200px;"> Function to compute a convolution layer.
491 <td rowspan="2">
492 <ul>
493 <li>ANEURALNETWORKS_CONV_2D
494 </ul>
495 <td>NEConvolutionLayer
496 <td>
497 <ul>
498 <li>NHWC
499 <li>NCHW
500 </ul>
501 <td>
502 <table>
503 <tr><th>src0<th>src1<th>src2<th>dst
504 <tr><td>F16<td>F16<td>F16<td>F16
505 <tr><td>F32<td>F32<td>F32<td>F32
506 <tr><td>QASYMM8<td>QASYMM8<td>S32<td>QASYMM8
507 <tr><td>QASYMM8<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8
508 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>S32<td>QASYMM8_SIGNED
509 <tr><td>QASYMM8_SIGNED<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8_SIGNED
510 </table>
511<tr>
512 <td>CLConvolutionLayer
513 <td>
514 <ul>
515 <li>NHWC
516 <li>NCHW
517 </ul>
518 <td>
519 <table>
520 <tr><th>src0<th>src1<th>src2<th>dst
521 <tr><td>F16<td>F16<td>F16<td>F16
522 <tr><td>F32<td>F32<td>F32<td>F32
523 <tr><td>QASYMM8<td>QASYMM8<td>S32<td>QASYMM8
524 <tr><td>QASYMM8<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8
525 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>S32<td>QASYMM8_SIGNED
526 <tr><td>QASYMM8_SIGNED<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8_SIGNED
527 </table>
528<tr>
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]529 <td rowspan="2">Copy
530 <td rowspan="2" style="width:200px;"> Function to copy a tensor.
531 <td rowspan="2">
532 <ul>
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]533 <li>n/a
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]534 </ul>
535 <td>NECopy
536 <td>
537 <ul>
538 <li>All
539 </ul>
540 <td>
541 <table>
542 <tr><th>src<th>dst
543 <tr><td>All<td>All
544 </table>
545<tr>
546 <td>CLCopy
547 <td>
548 <ul>
549 <li>All
550 </ul>
551 <td>
552 <table>
553 <tr><th>src<th>dst
554 <tr><td>All<td>All
555 </table>
556<tr>
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]557 <td rowspan="2">CropResize
558 <td rowspan="2" style="width:200px;"> Function to perform cropping and resizing.
559 <td rowspan="2">
560 <ul>
561 <li>n/a
562 </ul>
563 <td>NECropResize
564 <td>
565 <ul>
566 <li>NHWC
567 </ul>
568 <td>
569 <table>
570 <tr><th>src0<th>src1<th>src2<th>dst
571 <tr><td>All<td>F32<td>F32<td>F32
572 </table>
573<tr>
574 <td>CLCropResize
575 <td>
576 <ul>
577 <li>NHWC
578 </ul>
579 <td>
580 <table>
581 <tr><th>src0<th>src1<th>src2<th>dst
582 <tr><td>All<td>F32<td>F32<td>F32
583 </table>
584<tr>
585 <td rowspan="2">DeconvolutionLayer
586 <td rowspan="2" style="width:200px;"> Function to compute a deconvolution or tranpose convolution.
587 <td rowspan="2">
588 <ul>
589 <li>ANEURALNETWORKS_TRANSPOSE_CONV_2D
590 </ul>
591 <td>NEDeconvolutionLayer
592 <td>
593 <ul>
594 <li>NHWC
595 <li>NCHW
596 </ul>
597 <td>
598 <table>
599 <tr><th>src0<th>src1<th>src2<th>dst
600 <tr><td>F16<td>F16<td>F16<td>F16
601 <tr><td>F32<td>F32<td>F32<td>F32
602 <tr><td>QASYMM8<td>QASYMM8<td>S32<td>QASYMM8
603 <tr><td>QASYMM8<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8
604 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>S32<td>QASYMM8_SIGNED
605 <tr><td>QASYMM8_SIGNED<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8_SIGNED
606 </table>
607<tr>
608 <td>CLDeconvolutionLayer
609 <td>
610 <ul>
611 <li>NHWC
612 <li>NCHW
613 </ul>
614 <td>
615 <table>
616 <tr><th>src0<th>src1<th>src2<th>dst
617 <tr><td>F16<td>F16<td>F16<td>F16
618 <tr><td>F32<td>F32<td>F32<td>F32
619 <tr><td>QASYMM8<td>QASYMM8<td>S32<td>QASYMM8
620 <tr><td>QASYMM8<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8
621 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>S32<td>QASYMM8_SIGNED
622 <tr><td>QASYMM8_SIGNED<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8_SIGNED
623 </table>
624<tr>
625 <td rowspan="2">DepthConvertLayer
626 <td rowspan="2" style="width:200px;"> Performs a down-scaling depth conversion.
627 <td rowspan="2">
628 <ul>
629 <li>n/a
630 </ul>
631 <td>NEDepthConvertLayer
632 <td>
633 <ul>
634 <li>All
635 </ul>
636 <td>
637 <table>
638 <tr><th>src<th>dst
639 <tr><td>QASYMM8<td>F16, F32
640 <tr><td>U8<td>U16, S16, S32
641 <tr><td>U16<td>U8, U32
642 <tr><td>S16<td>U8, S32
643 <tr><td>BFLOAT16<td>F32
644 <tr><td>F16<td>QASYMM8, F32
645 <tr><td>F32<td>QASYMM8, F16, BFLOAT16
646 </table>
647<tr>
648 <td>CLDepthConvertLayer
649 <td>
650 <ul>
651 <li>All
652 </ul>
653 <td>
654 <table>
655 <tr><th>src<th>dst
656 <tr><td>U8<td>S8, U16, S16, U32, S32, F16, F32
657 <tr><td>U16<td>U8, S8, S16, U32, S32, F16, F32
658 <tr><td>S16<td>U8, S8, U16, U32, S32, F16, F32
659 <tr><td>U32<td>U8, S8, U16, S16, S32, F16, F32
660 <tr><td>S32<td>U8, S8, U16, S16, U32, F16, F32
661 <tr><td>F16<td>U8, S8, U16, S16, U32, F32
662 <tr><td>F32<td>U8, S8, U16, S16, U32, F16
663 </table>
664<tr>
665 <td rowspan="2">DepthToSpaceLayer
666 <td rowspan="2" style="width:200px;"> Depth to Space transformation.
667 <td rowspan="2">
668 <ul>
669 <li>ANEURALNETWORKS_DEPTH_TO_SPACE
670 </ul>
671 <td>NEDepthToSpaceLayer
672 <td>
673 <ul>
674 <li>NHWC
675 <li>NCHW
676 </ul>
677 <td>
678 <table>
679 <tr><th>src<th>dst
680 <tr><td>All<td>All
681 </table>
682<tr>
683 <td>CLDepthToSpaceLayer
684 <td>
685 <ul>
686 <li>NHWC
687 <li>NCHW
688 </ul>
689 <td>
690 <table>
691 <tr><th>src<th>dst
692 <tr><td>All<td>All
693 </table>
694<tr>
695 <td rowspan="2">DepthwiseConvolutionLayer
696 <td rowspan="2" style="width:200px;"> Function to perform depthwise separable convolution.
697 <td rowspan="2">
698 <ul>
699 <li>ANEURALNETWORKS_DEPTHWISE_CONV_2D
700 </ul>
701 <td>NEDepthwiseConvolutionLayer
702 <td>
703 <ul>
704 <li>NHWC
705 <li>NCHW
706 </ul>
707 <td>
708 <table>
709 <tr><th>src0<th>src1<th>src2<th>dst
710 <tr><td>F16<td>F16<td>F16<td>F16
711 <tr><td>F32<td>F32<td>F32<td>F32
712 <tr><td>QASYMM8<td>QASYMM8<td>S32<td>QASYMM8
713 <tr><td>QASYMM8<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8
714 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>S32<td>QASYMM8_SIGNED
715 <tr><td>QASYMM8_SIGNED<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8_SIGNED
716 </table>
717<tr>
718 <td>CLDepthwiseConvolutionLayer
719 <td>
720 <ul>
721 <li>NHWC
722 <li>NCHW
723 </ul>
724 <td>
725 <table>
726 <tr><th>src0<th>src1<th>src2<th>dst
727 <tr><td>F16<td>F16<td>F16<td>F16
728 <tr><td>F32<td>F32<td>F32<td>F32
729 <tr><td>QASYMM8<td>QASYMM8<td>S32<td>QASYMM8
730 <tr><td>QASYMM8<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8
731 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>S32<td>QASYMM8_SIGNED
732 <tr><td>QASYMM8_SIGNED<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8_SIGNED
733 </table>
734<tr>
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]735 <td rowspan="2">DequantizationLayer
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]736 <td rowspan="2" style="width:200px;"> Function to dequantize the values in a tensor.
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]737 <td rowspan="2">
738 <ul>
739 <li>ANEURALNETWORKS_DEQUANTIZE
740 </ul>
741 <td>NEDequantizationLayer
742 <td>
743 <ul>
744 <li>All
745 </ul>
746 <td>
747 <table>
748 <tr><th>src<th>dst
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]749 <tr><td>QASYMM8<td>F16, F32
750 <tr><td>QASYMM8_SIGNED<td>F16, F32
751 <tr><td>QSYMM8_PER_CHANNEL<td>F16, F32
752 <tr><td>QSYMM8<td>F16, F32
753 <tr><td>QSYMM16<td>F16, F32
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]754 </table>
755<tr>
756 <td>CLDequantizationLayer
757 <td>
758 <ul>
759 <li>All
760 </ul>
761 <td>
762 <table>
763 <tr><th>src<th>dst
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]764 <tr><td>QASYMM8<td>F16, F32
765 <tr><td>QASYMM8_SIGNED<td>F16, F32
766 <tr><td>QSYMM8_PER_CHANNEL<td>F16, F32
767 <tr><td>QSYMM8<td>F16, F32
768 <tr><td>QSYMM16<td>F16, F32
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]769 </table>
770<tr>
771 <td rowspan="2">DirectConvolutionLayer
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]772 <td rowspan="2" style="width:200px;"> Function to compute direct convolution.
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]773 <td rowspan="2">
774 <ul>
775 <li>ANEURALNETWORKS_CONV_2D
776 </ul>
777 <td>NEDirectConvolutionLayer
778 <td>
779 <ul>
780 <li>NHWC
781 <li>NCHW
782 </ul>
783 <td>
784 <table>
785 <tr><th>src0<th>src1<th>src2<th>dst
786 <tr><td>F16<td>F16<td>F16<td>F16
787 <tr><td>F32<td>F32<td>F32<td>F32
788 </table>
789<tr>
790 <td>CLDirectConvolutionLayer
791 <td>
792 <ul>
793 <li>NHWC
794 <li>NCHW
795 </ul>
796 <td>
797 <table>
798 <tr><th>src0<th>src1<th>src2<th>dst
799 <tr><td>F16<td>F16<td>F16<td>F16
800 <tr><td>F32<td>F32<td>F32<td>F32
801 <tr><td>QASYMM8<td>QASYMM8<td>S32<td>QASYMM8
802 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>S32<td>QASYMM8_SIGNED
803 </table>
804<tr>
805 <td rowspan="2">FFT1D
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]806 <td rowspan="2" style="width:200px;"> Fast Fourier Transform 1D.
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]807 <td rowspan="2">
808 <ul>
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]809 <li>n/a
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]810 </ul>
811 <td>NEFFT1D
812 <td>
813 <ul>
814 <li>All
815 </ul>
816 <td>
817 <table>
818 <tr><th>src<th>dst
819 <tr><td>F32<td>F32
820 </table>
821<tr>
822 <td>CLFFT1D
823 <td>
824 <ul>
825 <li>All
826 </ul>
827 <td>
828 <table>
829 <tr><th>src<th>dst
830 <tr><td>F32<td>F32
831 <tr><td>F16<td>F16
832 </table>
833<tr>
834 <td rowspan="2">FFT2D
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]835 <td rowspan="2" style="width:200px;"> Fast Fourier Transform 2D.
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]836 <td rowspan="2">
837 <ul>
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]838 <li>n/a
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]839 </ul>
840 <td>NEFFT2D
841 <td>
842 <ul>
843 <li>All
844 </ul>
845 <td>
846 <table>
847 <tr><th>src<th>dst
848 <tr><td>F32<td>F32
849 </table>
850<tr>
851 <td>CLFFT2D
852 <td>
853 <ul>
854 <li>All
855 </ul>
856 <td>
857 <table>
858 <tr><th>src<th>dst
859 <tr><td>F32<td>F32
860 <tr><td>F16<td>F16
861 </table>
862<tr>
863 <td rowspan="2">FFTConvolutionLayer
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]864 <td rowspan="2" style="width:200px;"> Fast Fourier Transform Convolution.
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]865 <td rowspan="2">
866 <ul>
867 <li>ANEURALNETWORKS_CONV_2D
868 </ul>
869 <td>NEFFTConvolutionLayer
870 <td>
871 <ul>
872 <li>All
873 </ul>
874 <td>
875 <table>
876 <tr><th>src<th>dst
877 <tr><td>F32<td>F32
878 </table>
879<tr>
880 <td>CLFFTConvolutionLayer
881 <td>
882 <ul>
883 <li>All
884 </ul>
885 <td>
886 <table>
887 <tr><th>src<th>dst
888 <tr><td>F32<td>F32
889 <tr><td>F16<td>F16
890 </table>
891<tr>
892 <td rowspan="2">Fill
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]893 <td rowspan="2" style="width:200px;"> Set the values of a tensor with a given value.
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]894 <td rowspan="2">
895 <ul>
896 <li>ANEURALNETWORKS_FILL
897 </ul>
898 <td>NEFill
899 <td>
900 <ul>
901 <li>All
902 </ul>
903 <td>
904 <table>
905 <tr><th>src<th>dst
906 <tr><td>All<td>All
907 </table>
908<tr>
909 <td>CLFill
910 <td>
911 <ul>
912 <li>All
913 </ul>
914 <td>
915 <table>
916 <tr><th>src<th>dst
917 <tr><td>All<td>All
918 </table>
919<tr>
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]920 <td rowspan="2">FillBorder
921 <td rowspan="2" style="width:200px;"> Function to .
922 <td rowspan="2">
923 <ul>
924 <li>n/a
925 </ul>
926 <td>NEFillBorder
927 <td>
928 <ul>
929 <li>All
930 </ul>
931 <td>
932 <table>
933 <tr><th>src<th>dst
934 <tr><td>All<td>All
935 </table>
936<tr>
937 <td>CLFillBorder
938 <td>
939 <ul>
940 <li>All
941 </ul>
942 <td>
943 <table>
944 <tr><th>src<th>dst
945 <tr><td>All<td>All
946 </table>
947<tr>
948 <td rowspan="2">FlattenLayer
949 <td rowspan="2" style="width:200px;"> Reshape a tensor to be 1D
950 <td rowspan="2">
951 <ul>
952 <li>ANEURALNETWORKS_RESHAPE
953 </ul>
954 <td>NEFlattenLayer
955 <td>
956 <ul>
957 <li>All
958 </ul>
959 <td>
960 <table>
961 <tr><th>src<th>dst
962 <tr><td>All<td>All
963 </table>
964<tr>
965 <td>CLFlattenLayer
966 <td>
967 <ul>
968 <li>All
969 </ul>
970 <td>
971 <table>
972 <tr><th>src<th>dst
973 <tr><td>All<td>All
974 </table>
975<tr>
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]976 <td rowspan="2">Floor
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]977 <td rowspan="2" style="width:200px;"> Round the value to the lowest number.
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]978 <td rowspan="2">
979 <ul>
980 <li>ANEURALNETWORKS_FLOOR
981 </ul>
982 <td>NEFloor
983 <td>
984 <ul>
985 <li>All
986 </ul>
987 <td>
988 <table>
989 <tr><th>src<th>dst
990 <tr><td>F32<td>F32
991 <tr><td>F16<td>F16
992 </table>
993<tr>
994 <td>CLFloor
995 <td>
996 <ul>
997 <li>All
998 </ul>
999 <td>
1000 <table>
1001 <tr><th>src<th>dst
1002 <tr><td>F32<td>F32
1003 <tr><td>F16<td>F16
1004 </table>
1005<tr>
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]1006 <td rowspan="2">FullyConnectedLayer
1007 <td rowspan="2" style="width:200px;"> Function to perform a fully connected / dense layer.
1008 <td rowspan="2">
1009 <ul>
1010 <li>ANEURALNETWORKS_FULLY_CONNECTED
1011 </ul>
1012 <td>NEFullyConnectedLayerReshapeWeightsManaged
1013 <td>
1014 <ul>
1015 <li>NHWC
1016 <li>NCHW
1017 </ul>
1018 <td>
1019 <table>
1020 <tr><th>src0<th>src1<th>src2<th>dst
1021 <tr><td>F16<td>F16<td>F16<td>F16
1022 <tr><td>F32<td>F32<td>F32<td>F32
1023 <tr><td>QASYMM8<td>QASYMM8<td>S32<td>QASYMM8
1024 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>S32<td>QASYMM8_SIGNED
1025 </table>
1026<tr>
1027 <td>CLFullyConnectedLayerReshapeWeightsManaged
1028 <td>
1029 <ul>
1030 <li>NHWC
1031 <li>NCHW
1032 </ul>
1033 <td>
1034 <table>
1035 <tr><th>src0<th>src1<th>src2<th>dst
1036 <tr><td>F16<td>F16<td>F16<td>F16
1037 <tr><td>F32<td>F32<td>F32<td>F32
1038 <tr><td>QASYMM8<td>QASYMM8<td>S32<td>QASYMM8
1039 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>S32<td>QASYMM8_SIGNED
1040 </table>
1041<tr>
1042 <td rowspan="2">FuseBatchNormalization
1043 <td rowspan="2" style="width:200px;"> Function to fuse the batch normalization node to a preceding convolution node.
1044 <td rowspan="2">
1045 <ul>
1046 <li>n/a
1047 </ul>
1048 <td>NEFuseBatchNormalization
1049 <td>
1050 <ul>
1051 <li>NHWC
1052 <li>NCHW
1053 </ul>
1054 <td>
1055 <table>
1056 <tr><th>src<th>dst
1057 <tr><td>F32<td>F32
1058 <tr><td>F16<td>F16
1059 </table>
1060<tr>
1061 <td>CLFuseBatchNormalization
1062 <td>
1063 <ul>
1064 <li>NHWC
1065 <li>NCHW
1066 </ul>
1067 <td>
1068 <table>
1069 <tr><th>src<th>dst
1070 <tr><td>F32<td>F32
1071 <tr><td>F16<td>F16
1072 </table>
1073<tr>
1074 <td rowspan="2">Gather
1075 <td rowspan="2" style="width:200px;"> Performs the Gather operation along the chosen axis.
1076 <td rowspan="2">
1077 <ul>
1078 <li>ANEURALNETWORKS_GATHER
1079 </ul>
1080 <td>NEGather
1081 <td>
1082 <ul>
1083 <li>All
1084 </ul>
1085 <td>
1086 <table>
1087 <tr><th>src<th>dst
1088 <tr><td>All<td>All
1089 </table>
1090<tr>
1091 <td>CLGather
1092 <td>
1093 <ul>
1094 <li>All
1095 </ul>
1096 <td>
1097 <table>
1098 <tr><th>src<th>dst
1099 <tr><td>All<td>All
1100 </table>
1101<tr>
1102 <td rowspan="2">GEMM
1103 <td rowspan="2" style="width:200px;"> General Matrix Multiplication.
1104 <td rowspan="2">
1105 <ul>
1106 <li>n/a
1107 </ul>
1108 <td>NEGEMM
1109 <td>
1110 <ul>
1111 <li>All
1112 </ul>
1113 <td>
1114 <table>
1115 <tr><th>src0<th>src1<th>src2<th>dst
1116 <tr><td>F32<td>F32<td>F32<td>F32
1117 <tr><td>F16<td>F16<td>F16<td>F16
1118 <tr><td>BFLOAT16<td>BFLOAT16<td>BFLOAT16<td>BFLOAT16
1119 </table>
1120<tr>
1121 <td>CLGEMMReshapeRHSMatrixKernelManaged
1122 <td>
1123 <ul>
1124 <li>All
1125 </ul>
1126 <td>
1127 <table>
1128 <tr><th>src0<th>src1<th>src2<th>dst
1129 <tr><td>F32<td>F32<td>F32<td>F32
1130 <tr><td>F16<td>F16<td>F16<td>F16
1131 </table>
1132<tr>
1133 <td rowspan="2">GEMMConvolutionLayer
1134 <td rowspan="2" style="width:200px;"> General Matrix Multiplication.
1135 <td rowspan="2">
1136 <ul>
1137 <li>ANEURALNETWORKS_CONV_2D
1138 </ul>
1139 <td>NEConvolutionLayerReshapeWeights
1140 <td>
1141 <ul>
1142 <li>NHWC
1143 <li>NCHW
1144 </ul>
1145 <td>
1146 <table>
1147 <tr><th>src0<th>src1<th>src2<th>dst
1148 <tr><td>F16<td>F16<td>F16<td>F16
1149 <tr><td>F32<td>F32<td>F32<td>F32
1150 <tr><td>BFLOAT16<td>BFLOAT16<td>BFLOAT16<td>BFLOAT16
1151 <tr><td>QASYMM8<td>QASYMM8<td>S32<td>QASYMM8
1152 <tr><td>QASYMM8<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8
1153 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>S32<td>QASYMM8_SIGNED
1154 <tr><td>QASYMM8_SIGNED<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8_SIGNED
1155 </table>
1156<tr>
1157 <td>CLConvolutionLayerReshapeWeights
1158 <td>
1159 <ul>
1160 <li>NHWC
1161 <li>NCHW
1162 </ul>
1163 <td>
1164 <table>
1165 <tr><th>src0<th>src1<th>src2<th>dst
1166 <tr><td>F16<td>F16<td>F16<td>F16
1167 <tr><td>F32<td>F32<td>F32<td>F32
1168 <tr><td>QASYMM8<td>QASYMM8<td>S32<td>QASYMM8
1169 <tr><td>QASYMM8<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8
1170 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>S32<td>QASYMM8_SIGNED
1171 <tr><td>QASYMM8_SIGNED<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8_SIGNED
1172 </table>
1173<tr>
1174 <td rowspan="2">GEMMLowpMatrixMultiplyCore
1175 <td rowspan="2" style="width:200px;"> General Matrix Multiplication.
1176 <td rowspan="2">
1177 <ul>
1178 <li>n/a
1179 </ul>
1180 <td>NEGEMMLowpMatrixMultiplyCore
1181 <td>
1182 <ul>
1183 <li>NHWC
1184 <li>NCHW
1185 </ul>
1186 <td>
1187 <table>
1188 <tr><th>src0<th>src1<th>src2<th>dst
1189 <tr><td>QASYMM8<td>QASYMM8<td>S32<td>QASYMM8
1190 <tr><td>QASYMM8<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8
1191 <tr><td>QASYMM8<td>QSYMM8<td>S32<td>QASYMM8
1192 <tr><td>QASYMM8<td>QASYMM8<td>S32<td>S32
1193 <tr><td>QASYMM8<td>QSYMM8_PER_CHANNEL<td>S32<td>S32
1194 <tr><td>QASYMM8<td>QSYMM8<td>S32<td>S32
1195 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>S32<td>QASYMM8_SIGNED
1196 <tr><td>QASYMM8_SIGNED<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8_SIGNED
1197 <tr><td>QASYMM8_SIGNED<td>QSYMM8<td>S32<td>QASYMM8_SIGNED
1198 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>S32<td>S32
1199 <tr><td>QASYMM8_SIGNED<td>QSYMM8_PER_CHANNEL<td>S32<td>S32
1200 <tr><td>QASYMM8_SIGNED<td>QSYMM8<td>S32<td>S32
1201 </table>
1202<tr>
1203 <td>CLGEMMLowpMatrixMultiplyCore
1204 <td>
1205 <ul>
1206 <li>NHWC
1207 <li>NCHW
1208 </ul>
1209 <td>
1210 <table>
1211 <tr><th>src0<th>src1<th>src2<th>dst
1212 <tr><td>QASYMM8<td>QASYMM8<td>S32<td>QASYMM8
1213 <tr><td>QASYMM8<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8
1214 <tr><td>QASYMM8<td>QSYMM8<td>S32<td>QASYMM8
1215 <tr><td>QASYMM8<td>QASYMM8<td>S32<td>S32
1216 <tr><td>QASYMM8<td>QSYMM8_PER_CHANNEL<td>S32<td>S32
1217 <tr><td>QASYMM8<td>QSYMM8<td>S32<td>S32
1218 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>S32<td>QASYMM8_SIGNED
1219 <tr><td>QASYMM8_SIGNED<td>QSYMM8_PER_CHANNEL<td>S32<td>QASYMM8_SIGNED
1220 <tr><td>QASYMM8_SIGNED<td>QSYMM8<td>S32<td>QASYMM8_SIGNED
1221 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>S32<td>S32
1222 <tr><td>QASYMM8_SIGNED<td>QSYMM8_PER_CHANNEL<td>S32<td>S32
1223 <tr><td>QASYMM8_SIGNED<td>QSYMM8<td>S32<td>S32
1224 </table>
1225<tr>
1226 <td rowspan="2">GenerateProposalsLayer
1227 <td rowspan="2" style="width:200px;"> Function to generate proposals for a RPN (Region Proposal Network).
1228 <td rowspan="2">
1229 <ul>
1230 <li>ANEURALNETWORKS_GENERATE_PROPOSALS
1231 </ul>
1232 <td>NEGenerateProposalsLayer
1233 <td>
1234 <ul>
1235 <li>All
1236 </ul>
1237 <td>
1238 <table>
1239 <tr><th>src0<th>src1<th>src2<th>dst
1240 <tr><td>F16<td>F16<td>F16<td>F16
1241 <tr><td>F32<td>F32<td>F32<td>F32
1242 <tr><td>QASYMM8<td>QSYMM8<td>QSYMM16<td>QASYMM8
1243 </table>
1244<tr>
1245 <td>CLGenerateProposalsLayer
1246 <td>
1247 <ul>
1248 <li>All
1249 </ul>
1250 <td>
1251 <table>
1252 <tr><th>src0<th>src1<th>src2<th>dst
1253 <tr><td>F16<td>F16<td>F16<td>F16
1254 <tr><td>F32<td>F32<td>F32<td>F32
1255 <tr><td>QASYMM8<td>QSYMM8<td>QSYMM16<td>QASYMM8
1256 </table>
1257<tr>
1258 <td rowspan="2">InstanceNormalizationLayer
1259 <td rowspan="2" style="width:200px;"> Function to perform a Instance normalization on a given axis.
1260 <td rowspan="2">
1261 <ul>
1262 <li>ANEURALNETWORKS_INSTANCE_NORMALIZATION
1263 </ul>
1264 <td>NEInstanceNormalizationLayer
1265 <td>
1266 <ul>
1267 <li>NHWC
1268 <li>NCHW
1269 </ul>
1270 <td>
1271 <table>
1272 <tr><th>src<th>dst
1273 <tr><td>F16<td>F16
1274 <tr><td>F32<td>F32
1275 </table>
1276<tr>
1277 <td>CLInstanceNormalizationLayer
1278 <td>
1279 <ul>
1280 <li>NHWC
1281 <li>NCHW
1282 </ul>
1283 <td>
1284 <table>
1285 <tr><th>src<th>dst
1286 <tr><td>F16<td>F16
1287 <tr><td>F32<td>F32
1288 </table>
1289<tr>
1290 <td rowspan="2">L2NormalizeLayer
1291 <td rowspan="2" style="width:200px;"> Function to perform a L2 normalization on a given axis.
1292 <td rowspan="2">
1293 <ul>
1294 <li>ANEURALNETWORKS_L2_NORMALIZATION
1295 </ul>
1296 <td>NEL2NormalizeLayer
1297 <td>
1298 <ul>
1299 <li>NHWC
1300 <li>NCHW
1301 </ul>
1302 <td>
1303 <table>
1304 <tr><th>src<th>dst
1305 <tr><td>F16<td>F16
1306 <tr><td>F32<td>F32
1307 </table>
1308<tr>
1309 <td>CLL2NormalizeLayer
1310 <td>
1311 <ul>
1312 <li>NHWC
1313 <li>NCHW
1314 </ul>
1315 <td>
1316 <table>
1317 <tr><th>src<th>dst
1318 <tr><td>F16<td>F16
1319 <tr><td>F32<td>F32
1320 </table>
1321<tr>
1322 <td rowspan="2">LSTMLayer
1323 <td rowspan="2" style="width:200px;"> Function to perform a single time step in a Long Short-Term Memory (LSTM) layer.
1324 <td rowspan="2">
1325 <ul>
1326 <li>ANEURALNETWORKS_LSTM
1327 </ul>
1328 <td>NELSTMLayer
1329 <td>
1330 <ul>
1331 <li>All
1332 </ul>
1333 <td>
1334 <table>
1335 <tr><th>src0 - src13<th>dst0 - dst3
1336 <tr><td>F16<td>F16
1337 <tr><td>F32<td>F32
1338 </table>
1339<tr>
1340 <td>CLLSTMLayer
1341 <td>
1342 <ul>
1343 <li>All
1344 </ul>
1345 <td>
1346 <table>
1347 <tr><th>src0 - src13<th>dst0 - dst3
1348 <tr><td>F16<td>F16
1349 <tr><td>F32<td>F32
1350 </table>
1351<tr>
1352 <td rowspan="2">LSTMLayerQuantized
1353 <td rowspan="2" style="width:200px;"> Function to perform quantized LSTM (Long Short-Term Memory)
1354 <td rowspan="2">
1355 <ul>
1356 <li>ANEURALNETWORKS_QUANTIZED_LSTM
1357 <li>ANEURALNETWORKS_QUANTIZED_16BIT_LSTM
1358 </ul>
1359 <td>NELSTMLayerQuantized
1360 <td>
1361 <ul>
1362 <li>All
1363 </ul>
1364 <td>
1365 <table>
1366 <tr><th>src0 - src8<th>src9 - src12<th>src13<th>src14<th>dst0<th>dst1
1367 <tr><td>QASYMM8<td>S32<td>QSYMM16<td>QASYMM8<td>QSYMM16<td>QASYMM8
1368 </table>
1369<tr>
1370 <td>CLLSTMLayerQuantized
1371 <td>
1372 <ul>
1373 <li>All
1374 </ul>
1375 <td>
1376 <table>
1377 <tr><th>src0 - src8<th>src9 - src12<th>src13<th>src14<th>dst0<th>dst1
1378 <tr><td>QASYMM8<td>S32<td>QSYMM16<td>QASYMM8<td>QSYMM16<td>QASYMM8
1379 </table>
1380<tr>
1381 <td rowspan="2">MaxUnpoolingLayer
1382 <td rowspan="2" style="width:200px;"> Function to perform MaxUnpooling.
1383 <td rowspan="2">
1384 <ul>
1385 <li>n/a
1386 </ul>
1387 <td>NEMaxUnpoolingLayer
1388 <td>
1389 <ul>
1390 <li>NHWC
1391 <li>NCHW
1392 </ul>
1393 <td>
1394 <table>
1395 <tr><th>src<th>dst
1396 <tr><td>QASYMM8<td>QASYMM8
1397 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
1398 <tr><td>F16<td>F16
1399 <tr><td>F32<td>F32
1400 </table>
1401<tr>
1402 <td>CLMaxUnpoolingLayer
1403 <td>
1404 <ul>
1405 <li>NHWC
1406 <li>NCHW
1407 </ul>
1408 <td>
1409 <table>
1410 <tr><th>src<th>dst
1411 <tr><td>QASYMM8<td>QASYMM8
1412 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
1413 <tr><td>F16<td>F16
1414 <tr><td>F32<td>F32
1415 </table>
1416<tr>
1417 <td rowspan="2">MeanStdDevNormalizationLayer
1418 <td rowspan="2" style="width:200px;"> Function to execute mean and standard deviation normalization.
1419 <td rowspan="2">
1420 <ul>
1421 <li>n/a
1422 </ul>
1423 <td>NEMeanStdDevNormalizationLayer
1424 <td>
1425 <ul>
1426 <li>NHWC
1427 <li>NCHW
1428 </ul>
1429 <td>
1430 <table>
1431 <tr><th>src<th>dst
1432 <tr><td>F32<td>F32
1433 <tr><td>F16<td>F16
1434 </table>
1435<tr>
1436 <td>CLMeanStdDevNormalizationLayer
1437 <td>
1438 <ul>
1439 <li>NHWC
1440 <li>NCHW
1441 </ul>
1442 <td>
1443 <table>
1444 <tr><th>src<th>dst
1445 <tr><td>F32<td>F32
1446 <tr><td>F16<td>F16
1447 </table>
1448<tr>
1449 <td rowspan="2">NormalizationLayer
1450 <td rowspan="2" style="width:200px;"> Function to compute normalization layer.
1451 <td rowspan="2">
1452 <ul>
1453 <li>ANEURALNETWORKS_LOCAL_RESPONSE_NORMALIZATION
1454 </ul>
1455 <td>NENormalizationLayer
1456 <td>
1457 <ul>
1458 <li>NHWC
1459 <li>NCHW
1460 </ul>
1461 <td>
1462 <table>
1463 <tr><th>src<th>dst
1464 <tr><td>F32<td>F32
1465 <tr><td>F16<td>F16
1466 </table>
1467<tr>
1468 <td>CLNormalizationLayer
1469 <td>
1470 <ul>
1471 <li>NHWC
1472 <li>NCHW
1473 </ul>
1474 <td>
1475 <table>
1476 <tr><th>src<th>dst
1477 <tr><td>F32<td>F32
1478 <tr><td>F16<td>F16
1479 </table>
1480<tr>
1481 <td rowspan="2">PadLayer
1482 <td rowspan="2" style="width:200px;"> Function to pad a tensor.
1483 <td rowspan="2">
1484 <ul>
1485 <li>ANEURALNETWORKS_PAD
1486 <li>ANEURALNETWORKS_PAD_V2
1487 </ul>
1488 <td>NEPadLayer
1489 <td>
1490 <ul>
1491 <li>NHWC
1492 <li>NCHW
1493 </ul>
1494 <td>
1495 <table>
1496 <tr><th>src<th>dst
1497 <tr><td>All<td>All
1498 </table>
1499<tr>
1500 <td>CLPadLayer
1501 <td>
1502 <ul>
1503 <li>NHWC
1504 <li>NCHW
1505 </ul>
1506 <td>
1507 <table>
1508 <tr><th>src<th>dst
1509 <tr><td>All<td>All
1510 </table>
1511<tr>
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]1512 <td rowspan="2">Permute
1513 <td rowspan="2" style="width:200px;"> Function to transpose an ND tensor.
1514 <td rowspan="2">
1515 <ul>
1516 <li>ANEURALNETWORKS_TRANSPOSE
1517 </ul>
1518 <td>NEPermute
1519 <td>
1520 <ul>
1521 <li>NHWC
1522 <li>NCHW
1523 </ul>
1524 <td>
1525 <table>
1526 <tr><th>src<th>dst
1527 <tr><td>All<td>All
1528 </table>
1529<tr>
1530 <td>CLPermute
1531 <td>
1532 <ul>
1533 <li>NHWC
1534 <li>NCHW
1535 </ul>
1536 <td>
1537 <table>
1538 <tr><th>src<th>dst
1539 <tr><td>All<td>All
1540 </table>
1541<tr>
1542 <td rowspan="2">PixelWiseMultiplication
1543 <td rowspan="2" style="width:200px;"> Function to performe a multiplication.
1544 <td rowspan="2">
1545 <ul>
1546 <li>ANEURALNETWORKS_MUL
1547 </ul>
1548 <td>NEPixelWiseMultiplication
1549 <td>
1550 <ul>
1551 <li>All
1552 </ul>
1553 <td>
1554 <table>
1555 <tr><th>src0<th>src1<th>dst
1556 <tr><td>QASYMM8<td>QASYMM8<td>QASYMM8
1557 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
1558 <tr><td>QSYMM16<td>QSYMM16<td>QASYMM16
1559 <tr><td>QSYMM16<td>QSYMM16<td>S32
1560 <tr><td>U8<td>U8<td>U8
1561 <tr><td>U8<td>U8<td>S16
1562 <tr><td>U8<td>S16<td>S16
1563 <tr><td>S16<td>U8<td>S16
1564 <tr><td>S16<td>S16<td>S16
1565 <tr><td>F16<td>F16<td>F16
1566 <tr><td>F32<td>S32<td>F32
1567 </table>
1568<tr>
1569 <td>CLPixelWiseMultiplication
1570 <td>
1571 <ul>
1572 <li>All
1573 </ul>
1574 <td>
1575 <table>
1576 <tr><th>src0<th>src1<th>dst
1577 <tr><td>QASYMM8<td>QASYMM8<td>QASYMM8
1578 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
1579 <tr><td>QSYMM16<td>QSYMM16<td>QASYMM16
1580 <tr><td>QSYMM16<td>QSYMM16<td>S32
1581 <tr><td>U8<td>U8<td>U8
1582 <tr><td>U8<td>U8<td>S16
1583 <tr><td>U8<td>S16<td>S16
1584 <tr><td>S16<td>U8<td>S16
1585 <tr><td>S16<td>S16<td>S16
1586 <tr><td>F16<td>F16<td>F16
1587 <tr><td>F32<td>S32<td>F32
1588 </table>
1589<tr>
1590 <td rowspan="2">PoolingLayer
1591 <td rowspan="2" style="width:200px;"> Function to performe pooling with the specified pooling operation.
1592 <td rowspan="2">
1593 <ul>
1594 <li>ANEURALNETWORKS_AVERAGE_POOL_2D
1595 <li>ANEURALNETWORKS_L2_POOL_2D
1596 <li>ANEURALNETWORKS_MAX_POOL_2D
1597 </ul>
1598 <td>NEPoolingLayer
1599 <td>
1600 <ul>
1601 <li>NHWC
1602 <li>NCHW
1603 </ul>
1604 <td>
1605 <table>
1606 <tr><th>src<th>dst
1607 <tr><td>QASYMM8<td>QASYMM8
1608 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
1609 <tr><td>F16<td>F16
1610 <tr><td>F32<td>F32
1611 </table>
1612<tr>
1613 <td>CLPoolingLayer
1614 <td>
1615 <ul>
1616 <li>NHWC
1617 <li>NCHW
1618 </ul>
1619 <td>
1620 <table>
1621 <tr><th>src<th>dst
1622 <tr><td>QASYMM8<td>QASYMM8
1623 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
1624 <tr><td>F16<td>F16
1625 <tr><td>F32<td>F32
1626 </table>
1627<tr>
1628 <td rowspan="2">PReluLayer
1629 <td rowspan="2" style="width:200px;"> Function to compute the activation layer with the PRELU activation function.
1630 <td rowspan="2">
1631 <ul>
1632 <li>ANEURALNETWORKS_PRELU
1633 </ul>
1634 <td>NEPReluLayer
1635 <td>
1636 <ul>
1637 <li>All
1638 </ul>
1639 <td>
1640 <table>
1641 <tr><th>src<th>dst
1642 <tr><td>QASYMM8<td>QASYMM8
1643 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
1644 <tr><td>F16<td>F16
1645 <tr><td>F32<td>F32
1646 </table>
1647<tr>
1648 <td>CLPReluLayer
1649 <td>
1650 <ul>
1651 <li>All
1652 </ul>
1653 <td>
1654 <table>
1655 <tr><th>src<th>dst
1656 <tr><td>QASYMM8<td>QASYMM8
1657 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
1658 <tr><td>F16<td>F16
1659 <tr><td>F32<td>F32
1660 </table>
1661<tr>
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]1662 <td rowspan="2">PriorBoxLayer
1663 <td rowspan="2" style="width:200px;"> Function to .
1664 <td rowspan="2">
1665 <ul>
1666 <li>n/a
1667 </ul>
1668 <td>NEPriorBoxLayer
1669 <td>
1670 <ul>
1671 <li>NHWC
1672 <li>NCHW
1673 </ul>
1674 <td>
1675 <table>
1676 <tr><th>src0<th>src1<th>dst
1677 <tr><td>F32<td>F32<td>F32
1678 </table>
1679<tr>
1680 <td>CLPriorBoxLayer
1681 <td>
1682 <ul>
1683 <li>NHWC
1684 <li>NCHW
1685 </ul>
1686 <td>
1687 <table>
1688 <tr><th>src0<th>src1<th>dst
1689 <tr><td>F32<td>F32<td>F32
1690 </table>
1691<tr>
1692 <td rowspan="2">QLSTMLayer
1693 <td rowspan="2" style="width:200px;"> Function to perform quantized LSTM (Long Short-Term Memory).
1694 <td rowspan="2">
1695 <ul>
1696 <li>ANEURALNETWORKS_QUANTIZED_LSTM
1697 <li>ANEURALNETWORKS_QUANTIZED_16BIT_LSTM
1698 </ul>
1699 <td>NEQLSTMLayer
1700 <td>
1701 <ul>
1702 <li>All
1703 </ul>
1704 <td>
1705 <table>
1706 <tr><th>src0<th>src1 - src6<th>src7 -src9<th>src10<th>src11<th>dst0<th>dst1 - dst2
1707 <tr><td>QASYMM8_SIGNED<td>QASYMM8<td>S32<td>QSYMM16<td>QASYMM8_SIGNED<td>QSYMM16<td>QASYMM8_SIGNED
1708 </table>
1709<tr>
1710 <td>CLQLSTMLayer
1711 <td>
1712 <ul>
1713 <li>All
1714 </ul>
1715 <td>
1716 <table>
1717 <tr><th>src0<th>src1 - src6<th>src7 -src9<th>src10<th>src11<th>dst0<th>dst1 - dst2
1718 <tr><td>QASYMM8_SIGNED<td>QASYMM8<td>S32<td>QSYMM16<td>QASYMM8_SIGNED<td>QSYMM16<td>QASYMM8_SIGNED
1719 </table>
1720<tr>
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]1721 <td rowspan="2">QuantizationLayer
1722 <td rowspan="2" style="width:200px;"> Function to perform quantization layer
1723 <td rowspan="2">
1724 <ul>
1725 <li>ANEURALNETWORKS_QUANTIZE
1726 </ul>
1727 <td>NEQuantizationLayer
1728 <td>
1729 <ul>
1730 <li>All
1731 </ul>
1732 <td>
1733 <table>
1734 <tr><th>src<th>dst
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]1735 <tr><td>QASYMM8<td>QASYMM8, QASYMM8_SIGNED, QASYMM16
1736 <tr><td>QASYMM8_SIGNED<td>QASYMM8, QASYMM8_SIGNED, QASYMM16
1737 <tr><td>F16<td>QASYMM8, QASYMM8_SIGNED, QASYMM16
1738 <tr><td>F32<td>QASYMM8, QASYMM8_SIGNED, QASYMM16
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]1739 </table>
1740<tr>
1741 <td>CLQuantizationLayer
1742 <td>
1743 <ul>
1744 <li>All
1745 </ul>
1746 <td>
1747 <table>
1748 <tr><th>src<th>dst
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]1749 <tr><td>QASYMM8<td>QASYMM8, QASYMM8_SIGNED, QASYMM16
1750 <tr><td>QASYMM8_SIGNED<td>QASYMM8, QASYMM8_SIGNED, QASYMM16
1751 <tr><td>F16<td>QASYMM8, QASYMM8_SIGNED, QASYMM16
1752 <tr><td>F32<td>QASYMM8, QASYMM8_SIGNED, QASYMM16
1753 </table>
1754<tr>
1755 <td rowspan="2">Range
1756 <td rowspan="2" style="width:200px;"> Function to generates a sequence of numbers starting from START and extends by increments of 'STEP' up to but not including 'END'.
1757 <td rowspan="2">
1758 <ul>
1759 <li>n/a
1760 </ul>
1761 <td>NERange
1762 <td>
1763 <ul>
1764 <li>All
1765 </ul>
1766 <td>
1767 <table>
1768 <tr><th>dst
1769 <tr><td>U8
1770 <tr><td>S8
1771 <tr><td>U16
1772 <tr><td>S16
1773 <tr><td>U32
1774 <tr><td>S32
1775 <tr><td>F16
1776 <tr><td>F32
1777 </table>
1778<tr>
1779 <td>CLRange
1780 <td>
1781 <ul>
1782 <li>All
1783 </ul>
1784 <td>
1785 <table>
1786 <tr><th>dst
1787 <tr><td>U8
1788 <tr><td>S8
1789 <tr><td>QASYMM8
1790 <tr><td>U16
1791 <tr><td>S16
1792 <tr><td>U32
1793 <tr><td>S32
1794 <tr><td>F16
1795 <tr><td>F32
1796 </table>
1797<tr>
1798 <td rowspan="2">ReduceMean
1799 <td rowspan="2" style="width:200px;"> Function to performe reduce mean operation.
1800 <td rowspan="2">
1801 <ul>
1802 <li>ANEURALNETWORKS_MEAN
1803 </ul>
1804 <td>NEReduceMean
1805 <td>
1806 <ul>
1807 <li>All
1808 </ul>
1809 <td>
1810 <table>
1811 <tr><th>src<th>dst
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]1812 <tr><td>QASYMM8<td>QASYMM8
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]1813 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]1814 <tr><td>F16<td>F16
1815 <tr><td>F32<td>F32
1816 </table>
1817<tr>
1818 <td>CLReduceMean
1819 <td>
1820 <ul>
1821 <li>All
1822 </ul>
1823 <td>
1824 <table>
1825 <tr><th>src<th>dst
1826 <tr><td>QASYMM8<td>QASYMM8
1827 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
1828 <tr><td>F16<td>F16
1829 <tr><td>F32<td>F32
1830 </table>
1831<tr>
1832 <td rowspan="2">ReductionOperation
1833 <td rowspan="2" style="width:200px;"> Function to performe reduce with the following operations - ARG_IDX_MAX: Index of the max value - ARG_IDX_MIN: Index of the min value - MEAN_SUM: Mean of sum - PROD: Product - SUM_SQUARE: Sum of squares - SUM: Sum - MIN: Min - MAX: Max
1834 <td rowspan="2">
1835 <ul>
1836 <li>ANEURALNETWORKS_REDUCE_ALL
1837 <li>ANEURALNETWORKS_REDUCE_ANY
1838 <li>ANEURALNETWORKS_REDUCE_MAX
1839 <li>ANEURALNETWORKS_REDUCE_MIN
1840 <li>ANEURALNETWORKS_REDUCE_PROD
1841 <li>ANEURALNETWORKS_REDUCE_SUM
1842 </ul>
1843 <td>NEReductionOperation
1844 <td>
1845 <ul>
1846 <li>All
1847 </ul>
1848 <td>
1849 <table>
1850 <tr><th>src<th>dst
1851 <tr><td>QASYMM8<td>QASYMM8
1852 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
1853 <tr><td>F16<td>F16
1854 <tr><td>F32<td>F32
1855 <tr><td>S32<td>S32
1856 </table>
1857<tr>
1858 <td>CLReductionOperation
1859 <td>
1860 <ul>
1861 <li>All
1862 </ul>
1863 <td>
1864 <table>
1865 <tr><th>src<th>dst
1866 <tr><td>QASYMM8<td>QASYMM8
1867 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
1868 <tr><td>F16<td>F16
1869 <tr><td>F32<td>F32
1870 <tr><td>S32<td>S32
1871 </table>
1872<tr>
1873 <td rowspan="2">ReorgLayer
1874 <td rowspan="2" style="width:200px;"> Performs a reorganization layer of input tensor to the output tensor.
1875 <td rowspan="2">
1876 <ul>
1877 <li>n/a
1878 </ul>
1879 <td>NEReorgLayer
1880 <td>
1881 <ul>
1882 <li>NHWC
1883 <li>NCHW
1884 </ul>
1885 <td>
1886 <table>
1887 <tr><th>src<th>dst
1888 <tr><td>All<td>All
1889 </table>
1890<tr>
1891 <td>CLReorgLayer
1892 <td>
1893 <ul>
1894 <li>NHWC
1895 <li>NCHW
1896 </ul>
1897 <td>
1898 <table>
1899 <tr><th>src<th>dst
1900 <tr><td>All<td>All
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]1901 </table>
1902<tr>
1903 <td rowspan="2">ReshapeLayer
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]1904 <td rowspan="2" style="width:200px;"> Function to reshape a tensor.
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]1905 <td rowspan="2">
1906 <ul>
1907 <li>ANEURALNETWORKS_RESHAPE
1908 <li>ANEURALNETWORKS_SQUEEZE
1909 </ul>
1910 <td>NEReshapeLayer
1911 <td>
1912 <ul>
1913 <li>All
1914 </ul>
1915 <td>
1916 <table>
1917 <tr><th>src<th>dst
1918 <tr><td>All<td>All
1919 </table>
1920<tr>
1921 <td>CLReshapeLayer
1922 <td>
1923 <ul>
1924 <li>All
1925 </ul>
1926 <td>
1927 <table>
1928 <tr><th>src<th>dst
1929 <tr><td>All<td>All
1930 </table>
1931<tr>
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]1932 <td rowspan="2">Reverse
1933 <td rowspan="2" style="width:200px;"> Function to reverse tensor according to axis.
1934 <td rowspan="2">
1935 <ul>
1936 <li>n/a
1937 </ul>
1938 <td>NEReverse
1939 <td>
1940 <ul>
1941 <li>All
1942 </ul>
1943 <td>
1944 <table>
1945 <tr><th>src0<th>src1<th>dst
1946 <tr><td>All<td>U32<td>All
1947 </table>
1948<tr>
1949 <td>CLReverse
1950 <td>
1951 <ul>
1952 <li>All
1953 </ul>
1954 <td>
1955 <table>
1956 <tr><th>src0<th>src1<th>dst
1957 <tr><td>All<td>U32<td>All
1958 </table>
1959<tr>
1960 <td rowspan="2">RNNLayer
1961 <td rowspan="2" style="width:200px;"> Function to perform recurrent neural network layer.
1962 <td rowspan="2">
1963 <ul>
1964 <li>ANEURALNETWORKS_RNN
1965 </ul>
1966 <td>NERNNLayer
1967 <td>
1968 <ul>
1969 <li>NHWC
1970 <li>NCHW
1971 </ul>
1972 <td>
1973 <table>
1974 <tr><th>src0<th>src1<th>src2<th>src3<th>dst0<th>dst1
1975 <tr><td>F16<td>F16<td>F16<td>F16<td>F16<td>F16
1976 <tr><td>F32<td>F32<td>F32<td>F32<td>F32<td>F32
1977 </table>
1978<tr>
1979 <td>CLRNNLayer
1980 <td>
1981 <ul>
1982 <li>NHWC
1983 <li>NCHW
1984 </ul>
1985 <td>
1986 <table>
1987 <tr><th>src0<th>src1<th>src2<th>src3<th>dst0<th>dst1
1988 <tr><td>F16<td>F16<td>F16<td>F16<td>F16<td>F16
1989 <tr><td>F32<td>F32<td>F32<td>F32<td>F32<td>F32
1990 </table>
1991<tr>
1992 <td rowspan="2">ROIAlignLayer
1993 <td rowspan="2" style="width:200px;"> Function to perform ROI alignment.
1994 <td rowspan="2">
1995 <ul>
1996 <li>ANEURALNETWORKS_ROI_ALIGN
1997 </ul>
1998 <td>NEROIAlignLayer
1999 <td>
2000 <ul>
2001 <li>All
2002 </ul>
2003 <td>
2004 <table>
2005 <tr><th>src0<th>src1<th>dst
2006 <tr><td>F16<td>F16<td>F16
2007 <tr><td>F32<td>F32<td>F32
2008 <tr><td>QASYMM8<td>QASYMM16<td>QASYMM8
2009 <tr><td>QASYMM8_SIGNED<td>QASYMM16<td>QASYMM8_SIGNED
2010 </table>
2011<tr>
2012 <td>CLROIAlignLayer
2013 <td>
2014 <ul>
2015 <li>All
2016 </ul>
2017 <td>
2018 <table>
2019 <tr><th>src0<th>src1<th>dst
2020 <tr><td>F16<td>F16<td>F16
2021 <tr><td>F32<td>F32<td>F32
2022 <tr><td>QASYMM8<td>QASYMM16<td>QASYMM8
2023 <tr><td>QASYMM8_SIGNED<td>QASYMM16<td>QASYMM8_SIGNED
2024 </table>
2025<tr>
2026 <td rowspan="2">ROIPoolingLayer
2027 <td rowspan="2" style="width:200px;"> Function to perform ROI pooling.
2028 <td rowspan="2">
2029 <ul>
2030 <li>ANEURALNETWORKS_ROI_POOLING
2031 </ul>
2032 <td>NEROIPoolingLayer
2033 <td>
2034 <ul>
2035 <li>All
2036 </ul>
2037 <td>
2038 <table>
2039 <tr><th>src0<th>src1<th>dst
2040 <tr><td>F32<td>U16<td>F32
2041 <tr><td>QASYMM8<td>U16<td>QASYMM8
2042 </table>
2043<tr>
2044 <td>CLROIPoolingLayer
2045 <td>
2046 <ul>
2047 <li>All
2048 </ul>
2049 <td>
2050 <table>
2051 <tr><th>src0<th>src1<th>dst
2052 <tr><td>F16<td>U16<td>F16
2053 <tr><td>F32<td>U16<td>F32
2054 <tr><td>QASYMM8<td>U16<td>QASYMM8
2055 </table>
2056<tr>
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]2057 <td rowspan="2">Scale
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]2058 <td rowspan="2" style="width:200px;"> Function to perform resize a tensor using to interpolate: - Bilinear - Nearest neighbor
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]2059 <td rowspan="2">
2060 <ul>
2061 <li>ANEURALNETWORKS_RESIZE_BILINEAR
2062 <li>ANEURALNETWORKS_RESIZE_NEAREST_NEIGHBOR
2063 </ul>
2064 <td>NEScale
2065 <td>
2066 <ul>
2067 <li>NHWC
2068 <li>NCHW
2069 </ul>
2070 <td>
2071 <table>
2072 <tr><th>src<th>dst
2073 <tr><td>QASYMM8<td>QASYMM8
2074 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
2075 <tr><td>F16<td>F16
2076 <tr><td>F32<td>F32
2077 <tr><td>U8<td>U8
2078 <tr><td>S16<td>S16
2079 </table>
2080<tr>
2081 <td>CLScale
2082 <td>
2083 <ul>
2084 <li>NHWC
2085 <li>NCHW
2086 </ul>
2087 <td>
2088 <table>
2089 <tr><th>src<th>dst
2090 <tr><td>QASYMM8<td>QASYMM8
2091 <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED
2092 <tr><td>F16<td>F16
2093 <tr><td>F32<td>F32
2094 <tr><td>U8<td>U8
2095 <tr><td>S16<td>S16
2096 </table>
2097<tr>
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]2098 <td rowspan="2">Select
2099 <td rowspan="2" style="width:200px;"> Function to select values from 2 tensors depending on an input tensor of booleans.
2100 <td rowspan="2">
2101 <ul>
2102 <li>ANEURALNETWORKS_SELECT
2103 </ul>
2104 <td>NESelect
2105 <td>
2106 <ul>
2107 <li>All
2108 </ul>
2109 <td>
2110 <table>
2111 <tr><th>src0<th>src1<th>src2<th>dst
2112 <tr><td>U8<td>All<td>All<td>All
2113 </table>
2114<tr>
2115 <td>CLSelect
2116 <td>
2117 <ul>
2118 <li>All
2119 </ul>
2120 <td>
2121 <table>
2122 <tr><th>src0<th>src1<th>src2<th>dst
2123 <tr><td>U8<td>All<td>All<td>All
2124 </table>
2125<tr>
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]2126 <td rowspan="2">Slice
2127 <td rowspan="2" style="width:200px;"> Function to perform tensor slicing.
2128 <td rowspan="2">
2129 <ul>
2130 <li>ANEURALNETWORKS_SLICE
2131 </ul>
2132 <td>NESlice
2133 <td>
2134 <ul>
2135 <li>All
2136 </ul>
2137 <td>
2138 <table>
2139 <tr><th>src<th>dst
2140 <tr><td>All<td>All
2141 </table>
2142<tr>
2143 <td>CLSlice
2144 <td>
2145 <ul>
2146 <li>All
2147 </ul>
2148 <td>
2149 <table>
2150 <tr><th>src<th>dst
2151 <tr><td>All<td>All
2152 </table>
2153<tr>
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]2154 <td rowspan="2">SpaceToBatchLayer
2155 <td rowspan="2" style="width:200px;"> Function to divide a tensor spatially.
2156 <td rowspan="2">
2157 <ul>
2158 <li>ANEURALNETWORKS_SPACE_TO_BATCH_ND
2159 </ul>
2160 <td>NESpaceToBatchLayer
2161 <td>
2162 <ul>
2163 <li>NHWC
2164 <li>NCHW
2165 </ul>
2166 <td>
2167 <table>
2168 <tr><th>src0<th>src1<th>src2<th>dst
2169 <tr><td>All<td>S32<td>S32<td>All
2170 </table>
2171<tr>
2172 <td>CLSpaceToBatchLayer
2173 <td>
2174 <ul>
2175 <li>NHWC
2176 <li>NCHW
2177 </ul>
2178 <td>
2179 <table>
2180 <tr><th>src0<th>src1<th>src2<th>dst
2181 <tr><td>All<td>S32<td>S32<td>All
2182 </table>
2183<tr>
2184 <td rowspan="2">SpaceToDepthLayer
2185 <td rowspan="2" style="width:200px;"> Function to rearrange blocks of spatial data into depth.
2186 <td rowspan="2">
2187 <ul>
2188 <li>ANEURALNETWORKS_SPACE_TO_DEPTH
2189 </ul>
2190 <td>NESpaceToDepthLayer
2191 <td>
2192 <ul>
2193 <li>NHWC
2194 <li>NCHW
2195 </ul>
2196 <td>
2197 <table>
2198 <tr><th>src<th>dst
2199 <tr><td>All<td>All
2200 </table>
2201<tr>
2202 <td>CLSpaceToDepthLayer
2203 <td>
2204 <ul>
2205 <li>NHWC
2206 <li>NCHW
2207 </ul>
2208 <td>
2209 <table>
2210 <tr><th>src<th>dst
2211 <tr><td>All<td>All
2212 </table>
2213<tr>
2214 <td rowspan="2">Split
2215 <td rowspan="2" style="width:200px;"> Function to split a tensor along a given axis.
2216 <td rowspan="2">
2217 <ul>
2218 <li>ANEURALNETWORKS_SPLIT
2219 </ul>
2220 <td>NESplit
2221 <td>
2222 <ul>
2223 <li>All
2224 </ul>
2225 <td>
2226 <table>
2227 <tr><th>src<th>dst
2228 <tr><td>All<td>All
2229 </table>
2230<tr>
2231 <td>CLSplit
2232 <td>
2233 <ul>
2234 <li>All
2235 </ul>
2236 <td>
2237 <table>
2238 <tr><th>src<th>dst
2239 <tr><td>All<td>All
2240 </table>
2241<tr>
2242 <td rowspan="2">StackLayer
2243 <td rowspan="2" style="width:200px;"> Function to stack tensors along an axis.
2244 <td rowspan="2">
2245 <ul>
2246 <li>n/a
2247 </ul>
2248 <td>NEStackLayer
2249 <td>
2250 <ul>
2251 <li>All
2252 </ul>
2253 <td>
2254 <table>
2255 <tr><th>src<th>dst
2256 <tr><td>All<td>All
2257 </table>
2258<tr>
2259 <td>CLStackLayer
2260 <td>
2261 <ul>
2262 <li>All
2263 </ul>
2264 <td>
2265 <table>
2266 <tr><th>src<th>dst
2267 <tr><td>All<td>All
2268 </table>
2269<tr>
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]2270 <td rowspan="2">StridedSlice
2271 <td rowspan="2" style="width:200px;"> Function to extract a strided slice of a tensor.
2272 <td rowspan="2">
2273 <ul>
2274 <li>ANEURALNETWORKS_STRIDED_SLICE
2275 </ul>
2276 <td>NEStridedSlice
2277 <td>
2278 <ul>
2279 <li>All
2280 </ul>
2281 <td>
2282 <table>
2283 <tr><th>src<th>dst
2284 <tr><td>All<td>All
2285 </table>
2286<tr>
2287 <td>CLStridedSlice
2288 <td>
2289 <ul>
2290 <li>All
2291 </ul>
2292 <td>
2293 <table>
2294 <tr><th>src<th>dst
2295 <tr><td>All<td>All
2296 </table>
2297<tr>
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]2298 <td rowspan="2">Tile
2299 <td rowspan="2" style="width:200px;"> Function to construct a tensor by tiling a given tensor.
2300 <td rowspan="2">
2301 <ul>
2302 <li>ANEURALNETWORKS_TILE
2303 </ul>
2304 <td>NETile
2305 <td>
2306 <ul>
2307 <li>All
2308 </ul>
2309 <td>
2310 <table>
2311 <tr><th>src<th>dst
2312 <tr><td>All<td>All
2313 </table>
2314<tr>
2315 <td>CLTile
2316 <td>
2317 <ul>
2318 <li>All
2319 </ul>
2320 <td>
2321 <table>
2322 <tr><th>src<th>dst
2323 <tr><td>All<td>All
2324 </table>
2325<tr>
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]2326 <td rowspan="2">Transpose
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]2327 <td rowspan="2" style="width:200px;"> Function to transpose a 2D tensor.
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]2328 <td rowspan="2">
2329 <ul>
2330 <li>ANEURALNETWORKS_TRANSPOSE
2331 </ul>
2332 <td>NETranspose
2333 <td>
2334 <ul>
2335 <li>All
2336 </ul>
2337 <td>
2338 <table>
2339 <tr><th>src<th>dst
2340 <tr><td>All<td>All
2341 </table>
2342<tr>
2343 <td>CLTranspose
2344 <td>
2345 <ul>
2346 <li>All
2347 </ul>
2348 <td>
2349 <table>
2350 <tr><th>src<th>dst
2351 <tr><td>All<td>All
2352 </table>
Teresa Charlin62687422021-04-28 10:58:49 +0100[diff] [blame]2353<tr>
2354 <td rowspan="2">Unstack
2355 <td rowspan="2" style="width:200px;"> Function to unpack a rank-R tensor into rank-(R-1) tensors.
2356 <td rowspan="2">
2357 <ul>
2358 <li>n/a
2359 </ul>
2360 <td>NEUnstack
2361 <td>
2362 <ul>
2363 <li>All
2364 </ul>
2365 <td>
2366 <table>
2367 <tr><th>src<th>dst
2368 <tr><td>All<td>All
2369 </table>
2370<tr>
2371 <td>CLUnstack
2372 <td>
2373 <ul>
2374 <li>All
2375 </ul>
2376 <td>
2377 <table>
2378 <tr><th>src<th>dst
2379 <tr><td>All<td>All
2380 </table>
2381<tr>
2382 <td rowspan="2">WinogradConvolutionLayer
2383 <td rowspan="2" style="width:200px;"> Function to do Winograd Convolution.
2384 <td rowspan="2">
2385 <ul>
2386 <li>ANEURALNETWORKS_CONV_2D
2387 </ul>
2388 <td>NEWinogradConvolutionLayer
2389 <td>
2390 <ul>
2391 <li>NHWC
2392 <li>NCHW
2393 </ul>
2394 <td>
2395 <table>
2396 <tr><th>src0<th>src1<th>src2<th>dst
2397 <tr><td>F16<td>F16<td>F16<td>F16
2398 <tr><td>F32<td>F32<td>F32<td>F32
2399 </table>
2400<tr>
2401 <td>CLWinogradConvolutionLayer
2402 <td>
2403 <ul>
2404 <li>NHWC
2405 <li>NCHW
2406 </ul>
2407 <td>
2408 <table>
2409 <tr><th>src0<th>src1<th>src2<th>dst
2410 <tr><td>F16<td>F16<td>F16<td>F16
2411 <tr><td>F32<td>F32<td>F32<td>F32
2412 </table>
Sheri Zhanga47dcc22021-04-22 14:41:12 +0100[diff] [blame]2413</table>
2414
2415*/
2416} // namespace