Gitiles
Code Review Sign In
review.mlplatform.org / ml / android-nn-driver / b23732bcd6cb51bc7a4da369b1654bfa02a70c00 / . / NnapiSupport.txt
blob: b4b2cc456e95daac2c8a367be4c5f1eeefba415e [file] [log] [blame]
surmeh0149b9e102018-05-17 14:11:25 +0100[diff] [blame]1------ ArmNN for Android NNAPI supported operations ------
telsoa015307bc12018-03-09 13:51:08 +0000[diff] [blame]2
Matthew Benthama3e23ca2019-05-13 12:49:59 +0100[diff] [blame]3This release of ArmNN for Android supports use as a driver for the Android Neural Networks API. It implements the
Aron Virginas-Tarb23732b2019-06-17 15:48:45 +0100[diff] [blame^]4android.hardware.neuralnetworks@1.0, android.hardware.neuralnetworks@1.1 and android.hardware.neuralnetworks@1.2
5HAL interfaces.
telsoa015307bc12018-03-09 13:51:08 +0000[diff] [blame]6
7For more information on the Android Neural Networks API, see https://developer.android.com/ndk/guides/neuralnetworks/index.html
8
9For integration and usage documentation, please see README.md.
10
11--- Support for Android Neural Networks HAL operations ---
12
Aron Virginas-Tarb23732b2019-06-17 15:48:45 +0100[diff] [blame^]13The following AndroidNN HAL 1.0 and 1.1 operations are currently supported:
telsoa015307bc12018-03-09 13:51:08 +0000[diff] [blame]14
15AndroidNN operator Tensor type supported
Matthew Benthamb92f8902019-06-05 09:24:31 +0100[diff] [blame]16ADD (FLOAT32,QUANT8_ASYMM)
telsoa015307bc12018-03-09 13:51:08 +0000[diff] [blame]17AVERAGE_POOL_2D (FLOAT32,QUANT8_ASYMM)
Sadik Armagan82fbeb32018-12-20 17:12:26 +0000[diff] [blame]18BATCH_TO_SPACE_ND (FLOAT32,QUANT8_ASYMM)
Matthew Benthamb92f8902019-06-05 09:24:31 +0100[diff] [blame]19CONCATENATION (FLOAT32,QUANT8_ASYMM)
20CONV_2D (FLOAT32,QUANT8_ASYMM)
21DEPTHWISE_CONV_2D (FLOAT32,QUANT8_ASYMM)
22DIV (FLOAT32,QUANT8_ASYMM)
23DEQUANTIZE (FLOAT32,QUANT8_ASYMM)
telsoa015307bc12018-03-09 13:51:08 +0000[diff] [blame]24FLOOR (FLOAT32)
Matthew Bentham378333d2018-10-01 16:07:27 +0100[diff] [blame]25FULLY_CONNECTED (FLOAT32,QUANT8_ASYMM)
telsoa015307bc12018-03-09 13:51:08 +0000[diff] [blame]26L2_NORMALIZATION (FLOAT32)
nikraj01c6494cc2019-05-22 16:30:44 +0100[diff] [blame]27L2_POOL_2D (FLOAT32,QUANT8_ASYMM)
telsoa015307bc12018-03-09 13:51:08 +0000[diff] [blame]28LOCAL_RESPONSE_NORMALIZATION (FLOAT32)
Matthew Benthamb92f8902019-06-05 09:24:31 +0100[diff] [blame]29LOGISTIC (FLOAT32,QUANT8_ASYMM)
Ferran Balaguerd04c0432018-11-15 14:48:05 +0000[diff] [blame]30LSTM (FLOAT32)
telsoa015307bc12018-03-09 13:51:08 +0000[diff] [blame]31MAX_POOL_2D (FLOAT32,QUANT8_ASYMM)
Ferran Balaguerd04c0432018-11-15 14:48:05 +0000[diff] [blame]32MEAN (FLOAT32,QUANT8_ASYMM)
Matthew Benthamb92f8902019-06-05 09:24:31 +0100[diff] [blame]33MUL (FLOAT32,QUANT8_ASYMM)
Ferran Balaguerd04c0432018-11-15 14:48:05 +0000[diff] [blame]34PAD (FLOAT32,QUANT8_ASYMM)
Matthew Benthamb92f8902019-06-05 09:24:31 +0100[diff] [blame]35RELU (FLOAT32,QUANT8_ASYMM)
36RELU1 (FLOAT32,QUANT8_ASYMM)
37RELU6 (FLOAT32,QUANT8_ASYMM)
telsoa015307bc12018-03-09 13:51:08 +0000[diff] [blame]38RESHAPE (FLOAT32,QUANT8_ASYMM)
nikraj01c6494cc2019-05-22 16:30:44 +0100[diff] [blame]39RESIZE_BILINEAR (FLOAT32,QUANT8_ASYMM)
telsoa015307bc12018-03-09 13:51:08 +0000[diff] [blame]40SOFTMAX (FLOAT32,QUANT8_ASYMM)
Sadik Armagan82fbeb32018-12-20 17:12:26 +0000[diff] [blame]41SPACE_TO_BATCH_ND (FLOAT32,QUANT8_ASYMM)
saoste01b8471482018-10-10 09:44:51 +0100[diff] [blame]42SQUEEZE (FLOAT32,QUANT8_ASYMM)
Sadik Armagan82fbeb32018-12-20 17:12:26 +0000[diff] [blame]43STRIDED_SLICE (FLOAT32,QUANT8_ASYMM)
Matthew Benthamb92f8902019-06-05 09:24:31 +0100[diff] [blame]44SUB (FLOAT32,QUANT8_ASYMM)
45TANH (FLOAT32,QUANT8_ASYMM)
saoste01fe463152018-10-18 17:49:56 +0100[diff] [blame]46TRANSPOSE (FLOAT32,QUANT8_ASYMM)
telsoa015307bc12018-03-09 13:51:08 +0000[diff] [blame]47
Aron Virginas-Tarb23732b2019-06-17 15:48:45 +0100[diff] [blame^]48The following AndroidNN HAL 1.2 operations are currently supported:
49
50CONV_2D (FLOAT32,QUANT8_ASYMM)
51DEPTHWISE_CONV_2D (FLOAT32,QUANT8_ASYMM)
telsoa015307bc12018-03-09 13:51:08 +0000[diff] [blame]52
53--- Unsupported operators ---
54
Aron Virginas-Tarb23732b2019-06-17 15:48:45 +0100[diff] [blame^]55The following AndroidNN HAL 1.0 operations are currently not supported:
telsoa015307bc12018-03-09 13:51:08 +0000[diff] [blame]56
57DEPTH_TO_SPACE
telsoa015307bc12018-03-09 13:51:08 +0000[diff] [blame]58EMBEDDING_LOOKUP
59HASHTABLE_LOOKUP
60LSH_PROJECTION
telsoa015307bc12018-03-09 13:51:08 +0000[diff] [blame]61RNN
62SPACE_TO_DEPTH
63SVDF
64
Aron Virginas-Tarb23732b2019-06-17 15:48:45 +0100[diff] [blame^]65The following AndroidNN HAL 1.2 operations are currently not supported:
66
67CONCATENATION
68LSTM
69MINIMUM
70MAXIMUM
71PAD_V2
72QUANTIZE
73QUANTIZED_16BIT_LSTM
74PRELU
75RESIZE_NEAREST_NEIGHBOR
76TRANSPOSE_CONV_2D
77
Mike Kelly56df76c2019-06-14 15:51:39 +0100[diff] [blame]78Where operations are not supported by the ArmNN Android NN Driver, the driver indicates this to the framework
79appropriately and the framework implements those operations using a CPU implementation.