Supported Network Layers

Supported Network Layers

Qualcomm® Neural Processing SDK supports the network layer types listed in the table below.

See Limitations for details on the limitations and constraints for the supported runtimes and individual layer types.

All of supported layers in GPU runtime are valid for both of GPU modes: GPU_FLOAT32_16_HYBRID and GPU_FLOAT16. GPU_FLOAT32_16_HYBRID - data storage is done in half float and computation is done in full float. GPU_FLOAT16 - both data storage and computation is done in half float.

A list of supported ONNX operations can be found at ONNX Operator Support.

Converters Equivalent

• COMMAND_LINE : indicates the Op is supported through command-line parameters provided during conversion and not as part of a source framework model. See the Source Framework’s converter help for more details.

• INFERRED: indicates Source Framework does not have a concrete definition for Op. However, converter pattern-matches a sequence of Ops to map to listed QNN Op.

• — : indicates there is no corresponding Source Framework Op, or the corresponding Op is not yet supported.

Runtime Support

• YES: Runtime has an implementation for Op.

• NO: Runtime does not have an implementation for Op.

No.	Operation	Converters Equivalent				Runtime Support
		Onnx	TensorFlow	TensorFlow Lite	PyTorch	CPU	GPU	AIP (HTA + DSP)	HTP (DSP v68+)	DSP (v66)
1	ArgbToRgb	COMMAND_LINE	COMMAND_LINE	—	—	YES	YES	YES	YES	YES
2	Argmax	ArgMax	argmax	—	argmax	YES	YES	YES	YES	YES
3	Argmin	ArgMin	argmin	—	argmin	YES	YES	YES	YES	YES
4	AxisAlignedBboxTransform	BBoxTransform(org.pytorch._caffe2) with im_info’s img_scale = 1	—	—	—	YES	NO	NO	NO	NO
5	Batchnorm	BatchNormalization	batch_normalization, fused_batch_norm(FusedBatchNorm, FusedBatchNormV3)	INFERRED	BatchNorm2d	YES	YES	YES	YES	YES
6	BatchPermutation	—	—	—	—	YES	NO	NO	NO	NO
7	BatchToSpace	—	batch_to_space	—	—	YES	YES	YES	YES	YES
8	BboxTransform	—	—	—	—	YES	NO	NO	NO	NO
9	BoxWithNmsLimit	BoxWithNMSLimit(org.pytorch._caffe2)	—	—	—	YES	NO	YES	NO	YES
10	Cast	Cast	cast	—	to	YES	YES	YES	YES	YES
11	ChannelShuffle	INFERRED	INFERRED	—	ChannelShuffle	YES	YES	YES	YES	YES
12	CollectRpnProposals	—	—	—	—	YES	NO	NO	NO	NO
13	Concat	Concat	concat(Concat, ConcatV2)	concatenation	cat	YES	YES	YES	YES	YES
14	ConstantOfShape	—	—	—	—	YES	NO	NO	NO	NO
15	Conv2d	Conv	conv2d	conv_2d	Conv2d	YES	YES	YES	YES	YES
16	Conv3d	Conv	conv3d	—	—	YES	NO	YES	YES	YES
17	Convert	—	—	—	—	YES	NO	YES	YES	YES
18	Correlation1D	—	—	—	—	YES	NO	YES	NO	YES
19	CropAndResize	—	crop_and_resize	—	—	YES	NO	NO	NO	NO
20	CumulativeSum	CumSum	cumsum	—	cumsum	YES	NO	NO	YES	NO
21	DepthToSpace	DepthToSpace	depth_to_space	depth_to_space	PixelShuffle	YES	YES	YES	YES	YES
22	DepthWiseConv2d	Conv with ‘num_output’ == ‘input channels’ == ‘group’	depthwise_conv2d	—	—	YES	YES	YES	YES	YES
23	Dequantize	DequantizeLinear	—	—	—	YES	YES	YES	YES	YES
24	DetectionOutput	—	INFERRED	TfliteDetectionPostProcess	—	YES	YES	YES	YES	YES
25	DistributeFpnProposals	—	—	—	—	YES	NO	NO	NO	NO
26	ElementWiseAbs	Abs	abs	abs	abs	YES	YES	YES	YES	YES
27	ElementWiseAdd	Add, Sum	add(Add, AddV2, Sum), bias_add	add	add	YES	YES	YES	YES	YES
28	ElementWiseAnd	And	logical_and	—	logical_and	YES	YES	NO	YES	NO
29	ElementWiseAsin	Asin	—	—	asin	YES	NO	NO	NO	NO
30	ElementWiseAtan	Atan	—	—	atan	YES	NO	NO	YES	NO
31	ElementWiseCeil	Ceil	ceil	—	ceil	YES	NO	YES	YES	YES
32	ElementWiseCos	Cos	—	—	cos	YES	YES	NO	YES	NO
33	ElementWiseDivide	Div, Reciprocal	divide, realdiv	div	div	YES	YES	YES	YES	YES
34	ElementWiseEqual	Equal	equal	—	eq	YES	YES	YES	YES	YES
35	ElementWiseExp	Exp	exp	exp	exp	YES	YES	YES	YES	YES
36	ElementWiseFloor	Floor	floor	floor	floor	YES	YES	YES	YES	YES
37	ElementWiseFloorDiv	—	floordiv	—	floor_divide	YES	NO	YES	YES	YES
38	ElementWiseFmod	—	—	—	—	NO	NO	NO	NO	NO
39	ElementWiseGreater	Greater	greater	—	gt	YES	YES	YES	YES	YES
40	ElementWiseGreaterEqual	GreaterOrEqual	greater_equal	—	ge	YES	YES	YES	YES	YES
41	ElementWiseLess	Less	less	—	lt	YES	YES	YES	YES	YES
42	ElementWiseLessEqual	LessOrEqual	less_equal	—	le	YES	YES	YES	YES	YES
43	ElementWiseLog	Log	log	—	log	YES	YES	YES	YES	YES
44	ElementWiseMaximum	Max	maximum	maximum	maximum	YES	YES	YES	YES	YES
45	ElementWiseMinimum	Min	minimum	minimum	minimum	YES	YES	YES	YES	YES
46	ElementWiseMod	—	—	—	—	YES	NO	NO	NO	NO
47	ElementWiseMultiply	Mul	mul	mul	mul	YES	YES	YES	YES	YES
48	ElementWiseNeg	Neg	negative	—	neg	YES	YES	YES	YES	YES
49	ElementWiseNot	Not	logical_not	—	logical_not	YES	YES	NO	YES	NO
50	ElementWiseNotEqual	—	not_equal	—	ne	YES	YES	YES	YES	YES
51	ElementWiseOr	Or	logical_or	—	logical_or	YES	YES	NO	NO	NO
52	ElementWisePower	Pow	pow, square	—	pow	YES	YES	YES	YES	YES
53	ElementWiseRound	Round	round	—	round	YES	YES	YES	YES	YES
54	ElementWiseRsqrt	—	rsqrt	—	rsqrt	YES	YES	YES	YES	YES
55	ElementWiseSelect	Where	where	—	—	YES	YES	YES	YES	YES
56	ElementWiseSign	Sign	—	—	sign	YES	NO	NO	YES	NO
57	ElementWiseSin	Sin	sin	—	sin	YES	YES	NO	YES	NO
58	ElementWiseSoftplus	Softplus	Softplus	—	Softplus	YES	NO	NO	NO	NO
59	ElementWiseSquaredDifference	—	—	—	—	YES	YES	YES	YES	YES
60	ElementWiseSquareRoot	Sqrt	sqrt	sqrt	sqrt	YES	YES	YES	YES	YES
61	ElementWiseSubtract	Sub	subtract	sub	sub	YES	YES	YES	YES	YES
62	ElementWiseUnary	—	—	—	—	YES	NO	NO	NO	NO
63	ElementWiseXor	Xor	logical_xor	—	logical_xor	YES	YES	NO	NO	NO
64	Elu	Elu	elu	—	—	YES	YES	YES	YES	YES
65	ExpandDims	—	—	—	—	YES	YES	NO	YES	NO
66	ExtractGlimpse	—	extract_glimpse	—	—	YES	NO	YES	YES	YES
67	ExtractPatches	—	extract_patches	—	—	YES	NO	NO	YES	NO
68	FullyConnected	MatMul(limited usecase), Gemm(limited usecase)	dense and tensordot(MatMul)	fully_connected	Linear	YES	YES	YES	YES	YES
69	Gather	Gather	gather(Gather, GatherV2)	—	—	YES	YES	YES	YES	YES
70	GatherElements	GatherElements	—	—	—	YES	NO	NO	YES	NO
71	GatherNd	GatherND	gather_nd	—	—	YES	NO	NO	YES	NO
72	Gelu	INFERRED / Gelu(for onnx version>=1.15)	INFERRED	—	GELU	YES	NO	NO	YES	NO
73	GenerateProposals	GenerateProposals(org.pytorch._caffe2) with im_info’s img_scale = 1	—	—	—	YES	NO	YES	NO	YES
74	GridSample	GridSample	—	—	—	YES	NO	NO	YES	NO
75	GroupNorm	—	—	—	GroupNorm	YES	NO	NO	NO	NO
76	Gru	—	—	—	—	NO	NO	NO	NO	NO
77	HardSwish	MATCHED	INFERRED	—	Hardswish	YES	YES	YES	YES	YES
78	HeatMapMaxKeyPoint	—	—	—	—	YES	YES	NO	YES	NO
79	ImageProjectionTransform	—	image.transform(ImageProjectiveTransform)	—	—	YES	NO	YES	YES	YES
80	InstanceNorm	InstanceNormalization	INFERRED	—	InstanceNorm2d	YES	YES	YES	YES	YES
81	L2Norm	LpNormalization	INFERRED	—	—	YES	YES	YES	YES	YES
82	L2Pool2d	LpPool	—	—	—	YES	YES	NO	NO	NO
83	LayerNorm	MATCHED	Layer_Normalization	—	LayerNorm	YES	YES	NO	YES	NO
84	LogSoftmax	LogSoftmax	log_softmax	—	LogSoftmax	YES	YES	NO	YES	NO
85	Lrn	LRN	local_response_normalization	—	—	YES	YES	YES	YES	YES
86	Lstm	LSTM	INFERRED	—	—	YES	YES	YES	YES	YES
87	MatMul	MatMul	matmul(BatchMatMul, BatchMatMulV2)	—	matmul	YES	YES	YES	YES	YES
88	Moments	—	INFERRED	—	—	YES	NO	YES	NO	YES
89	MultiClassNms	nms + gather	nms + gather	—	—	YES	NO	NO	YES	NO
90	NonMaxSuppression	NonMaxSuppression	—	—	—	YES	NO	NO	YES	NO
91	NonZero	NonZero	—	—	—	YES	NO	NO	YES	NO
92	Nv12ToRgb	COMMAND_LINE	COMMAND_LINE	—	—	YES	YES	YES	YES	YES
93	Nv21ToRgb	COMMAND_LINE	COMMAND_LINE	—	—	YES	YES	YES	YES	YES
94	OneHot	OneHot	one_hot	—	—	YES	NO	NO	YES	NO
95	Pack	—	stack(Stack, Pack)	—	stack	YES	YES	YES	YES	YES
96	Pad	Pad	pad(Pad, PadV2)	—	ConstantPad	YES	YES	YES	YES	YES
97	PoolAvg2d	AveragePool, GlobalAveragePool	average_pooling2d	average_pool_2d	AvgPool2d	YES	YES	YES	YES	YES
98	PoolAvg3d	AveragePool, GlobalAveragePool	—	—	—	YES	NO	YES	YES	YES
99	PoolMax2d	MaxPool, GlobalMaxPool	max_pooling2d	max_pool_2d	MaxPool2d	YES	YES	YES	YES	YES
100	PoolMax3d	MaxPool, GlobalMaxPool	—	—	—	YES	NO	YES	NO	YES
101	Prelu	PRelu, LeakyRelu	PReLU	—	PReLU	YES	YES	YES	YES	YES
102	Quantize	QuantizeLinear	—	—	—	YES	NO	YES	YES	YES
103	ReduceMax	ReduceMax	reduce_max	—	max	YES	YES	YES	YES	YES
104	ReduceMean	ReduceMean	reduce_mean	—	mean	YES	YES	YES	YES	YES
105	ReduceMin	ReduceMin	reduce_min	—	min	YES	YES	YES	YES	YES
106	ReduceProd	ReduceProd	reduce_prod	—	prod	YES	YES	NO	NO	NO
107	ReduceSum	ReduceSum	reduce_sum	—	sum	YES	YES	YES	YES	YES
108	Relu	Relu	relu	relu	ReLU	YES	YES	YES	YES	YES
109	Relu1	—	—	—	—	NO	YES	NO	YES	NO
110	Relu6	—	relu6	—	ReLU6	YES	YES	YES	YES	YES
111	ReluMinMax	Clip	INFERRED	relu6	Hardtanh	YES	YES	YES	YES	YES
112	Reshape	Reshape, Flatten, Squeeze, UnSqueeze	reshape, squeeze, expand_dims	reshape	reshape	YES	YES	YES	YES	YES
113	Resize	Resize	—	—	Resize	YES	NO	NO	YES	NO
114	ResizeBilinear	Resize	resize_bilinear	resize_bilinear	UpsamplingBilinear2d	YES	YES	YES	YES	YES
115	ResizeNearestNeighbor	Resize, ResizeNearest(org.pytorch._caffe2)	resize_nearest_neighbor	—	—	YES	YES	YES	YES	YES
116	RoiAlign	RoiAlign, RoIAlign(org.pytorch._caffe2)	—	—	—	YES	YES	YES	YES	YES
117	RoiPooling	MaxRoiPool	—	—	—	YES	NO	YES	NO	YES
118	ScatterElements	ScatterElements, Scatter (deprecated)	—	—	—	YES	NO	NO	YES	NO
119	ScatterNd	ScatterND	—	—	—	YES	NO	NO	YES	NO
120	Shape	—	—	—	—	YES	NO	NO	NO	NO
121	Sigmoid	Sigmoid	sigmoid	—	sigmoid	YES	YES	YES	YES	YES
122	Softmax	Softmax	softmax	softmax	Softmax	YES	YES	YES	YES	YES
123	SpaceToBatch	—	space_to_batch(SpaceToBatchND)	—	—	YES	YES	YES	YES	YES
124	SpaceToDepth	SpaceToDepth	space_to_depth	—	—	YES	YES	YES	YES	YES
125	Split	Split	split(Split, SplitV)	—	split	YES	YES	YES	YES	YES
126	Squeeze	—	—	—	—	YES	YES	YES	YES	YES
127	StridedSlice	Slice	strided_slice	slice	—	YES	YES	YES	YES	YES
128	Tanh	Tanh	tanh	tanh	tanh	YES	YES	YES	YES	YES
129	Tile	Tile	tile	—	—	YES	YES	YES	YES	YES
130	TopK	TopK	top_k	—	topk	YES	YES	NO	YES	NO
131	Transpose	Transpose	transpose	—	transpose	YES	YES	YES	YES	YES
132	TransposeConv2d	ConvTranspose	conv2d_transpose	transpose_conv	ConvTranspose2d	YES	YES	YES	YES	YES
133	TransposeConv3d	ConvTranspose	—	—	—	YES	NO	NO	YES	NO
134	UnPack	—	unstack	—	unbind	YES	YES	YES	YES	YES

Note : AIP Runtime supports all layers supported by the DSP runtime, as layers not supported by HTA run on HVX.