DSP Backend Op Definition Supplement

ArgbToRgb

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

Argmax

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_INT_32, QNN_DATATYPE_UINT_32

Constraints

Configuration	in[0]	out[0]
All	Shape: Max supported input rank is 4. shape(input)[axis] should be less than INT32_MAX	Shape: Max supported output rank is 4.

Argmin

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_INT_32, QNN_DATATYPE_UINT_32

Constraints

Configuration	in[0]	out[0]
All	Shape: Max supported input rank is 4. shape(input)[axis] should be less than INT32_MAX	Shape: Max supported output rank is 4.

Batchnorm

Datatypes

Configuration	in[0]	in[1]	in[2]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

BatchToSpace

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

BoxWithNmsLimit

Datatypes

Configuration	in[0]	in[1]	in[2]	in[3]	out[0]	out[1]	out[2]	out[3]	out[4]	out[5]	out[6]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_INT_32	QNN_DATATYPE_INT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_INT_32	QNN_DATATYPE_INT_32	QNN_DATATYPE_INT_32	QNN_DATATYPE_INT_32	QNN_DATATYPE_INT_32

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

Cast

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_FLOAT_32, QNN_DATATYPE_INT_32, QNN_DATATYPE_UINT_32, QNN_DATATYPE_UINT_8, QNN_DATATYPE_BOOL_8, QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_32	QNN_DATATYPE_INT_32, QNN_DATATYPE_UINT_8, QNN_DATATYPE_BOOL_8, QNN_DATATYPE_FLOAT_32, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_UFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ChannelShuffle

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	out[0]	num_groups
All	Shape: Max supported input rank is 4.	Shape: Max supported output rank is 4.	should be less than INT32_MAX

Concat

Datatypes

Configuration	in[0..m]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32

Constraints

Configuration	in[0..m]
All	Shape: QNN_DATATYPE_INT_32’s max supported input rank is 4; Other data type’s max supported input rank is 5.

Conv2d

Datatypes

Configuration	in[0]	in[1]	in[2]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_32, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	reuse_sparse_indices
All	Shape: Max supported input rank is 4.	Only support default value 0

Conv3d

Datatypes

Configuration	in[0]	in[1]	in[2]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_32, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 5.

Convert

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_FLOAT_32, QNN_DATATYPE_BOOL_8, QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_UFIXED_POINT_16, QNN_DATATYPE_SFIXED_POINT_16	QNN_DATATYPE_FLOAT_32, QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_UFIXED_POINT_16, QNN_DATATYPE_SFIXED_POINT_16

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

Correlation1D

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

DepthToSpace

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

DepthWiseConv2d

Datatypes

Configuration	in[0]	in[1]	in[2]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

Dequantize

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_FLOAT_32

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4. Uint8 per-channel quant tensor is not supported

DetectionOutput

Datatypes

Configuration	in[0]	in[1]	in[2]	out[0]	out[1]	out[2]	out[3]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_FLOAT_32	QNN_DATATYPE_INT_32	QNN_DATATYPE_UINT_32

Constraints

Configuration	in[0]	out[0]
All	Shape: Max supported input rank is 4.	Datatype: out[0] must be same Datatype as in[1]

ElementWiseAbs

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ElementWiseAdd

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32

Constraints

Configuration	in[0]	in[1]	out[0]
All	Shape: Supports input of Rank between 1 and 5.	Shape: Supports input of Rank between 1 and 5. Mixed broadcast cases (e.g. (1,1,3,32), (1,5,3,1) are not supported Broadcast cases which rank is 5 are not supported	Shape: Supports output of Rank between 1 and 5. Datatype: Int32 cases with rank is 5 are not supported

ElementWiseBinary

Datatypes

Configuration	in[0]	in[1]	out[0]	operation
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32, QNN_DATATYPE_BOOL_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_FLOAT_32, QNN_DATATYPE_INT_32, QNN_DATATYPE_BOOL_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32, QNN_DATATYPE_BOOL_8	QNN_DATATYPE_UINT_32

Constraints

Configuration	in[0]	in[1]	out[0]	operation
All	Shape: Supports input of Rank between 1 and 5.	Shape: Supports input of Rank between 1 and 5.	Shape: Supports output of Rank between 1 and 5.	Unsupported operations: FMOD, MOD.

ElementWiseCeil

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ElementWiseDivide

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32

Constraints

Configuration	in[0]	in[1]	out[0]
All	Shape: Supports input of Rank between 1 and 4.	Shape: Supports input of Rank between 1 and 4.	Shape: Supports input of Rank between 1 and 4.

ElementWiseEqual

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_BOOL_8

Constraints

Configuration	in[0]	in[1]	out[0]
All	Shape: Supports input of Rank between 1 and 4.	Shape: Supports input of Rank between 1 and 4.	Shape: Supports output of Rank between 1 and 4.

ElementWiseExp

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ElementWiseFloor

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ElementWiseFloorDiv

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32

Constraints

Configuration	in[0]	in[1]	out[0]
All	Shape: Supports input of Rank between 1 and 4.	Shape: Supports input of Rank between 1 and 4.	Shape: Supports output of Rank between 1 and 4.

ElementWiseGreater

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_BOOL_8

Constraints

Configuration	in[0]	in[1]	out[0]
All	Shape: Supports input of Rank between 1 and 4.	Shape: Supports input of Rank between 1 and 4.	Shape: Supports output of Rank between 1 and 4.

ElementWiseGreaterEqual

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_BOOL_8

Constraints

Configuration	in[0]	in[1]	out[0]
All	Shape: Supports input of Rank between 1 and 4.	Shape: Supports input of Rank between 1 and 4.	Shape: Supports output of Rank between 1 and 4.

ElementWiseLess

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_BOOL_8

Constraints

Configuration	in[0]	in[1]	out[0]
All	Shape: Supports input of Rank between 1 and 4.	Shape: Supports input of Rank between 1 and 4.	Shape: Supports output of Rank between 1 and 4.

ElementWiseLessEqual

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_BOOL_8

Constraints

Configuration	in[0]	in[1]	out[0]
All	Shape: Supports input of Rank between 1 and 4.	Shape: Supports input of Rank between 1 and 4.	Shape: Supports output of Rank between 1 and 4.

ElementWiseLog

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ElementWiseMaximum

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	in[1]	out[0]
All	Shape: Supports input of Rank between 1 and 4.	Shape: Supports input of Rank between 1 and 4.	Shape: Supports input of Rank between 1 and 4.

ElementWiseMinimum

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	in[1]	out[0]
All	Shape: Supports input of Rank between 1 and 4.	Shape: Supports input of Rank between 1 and 4.	Shape: Supports output of Rank between 1 and 4.

ElementWiseMultiply

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32

Constraints

Configuration	in[0]	in[1]	out[0]
All	Shape: Supports input of Rank between 1 and 4.	Shape: Supports input of Rank between 1 and 4.	Shape: Supports output of Rank between 1 and 4.

ElementWiseNeg

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ElementWiseNeuron

Datatypes

Configuration	in[0]	out[0]	operation
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UINT_32

Constraints

Configuration	in[0]	operation
All	Shape: Max supported input rank is 5.	Unsupported operations: GELU, SOFTPLUS.

ElementWiseNotEqual

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_BOOL_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ElementWisePower

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_INT_32, QNN_DATATYPE_FLOAT_32, QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	out[0]
All	Shape: Max supported output rank is 4.	Shape: Max supported output rank is 4.

ElementWiseRound

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ElementWiseRsqrt

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ElementWiseSelect

Datatypes

Configuration	in[0]	in[1]	in[2]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_BOOL_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ElementWiseSquaredDifference

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	in[1]	out[0]
All	Shape: Supports input of Rank between 1 and 4.	Shape: Supports input of Rank between 1 and 4.	Shape: Supports output of Rank between 1 and 4.

ElementWiseSquareRoot

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ElementWiseSubtract

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32

Constraints

Configuration	in[0]	in[1]	out[0]
All	Shape: Supports input of Rank between 1 and 4.	Shape: Supports input of Rank between 1 and 4.	Shape: Supports output of Rank between 1 and 4.

ElementWiseUnary

Datatypes

Configuration	in[0]	out[0]	operation
All	QNN_DATATYPE_BOOL_8, QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_BOOL_8, QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UINT_32

Constraints

Configuration	in[0]	operation
All	Shape: Max supported input rank is 4.	Unsupported operations: ASIN, ATAN, COS, RECIPROCAL, SIGN, SIN.

ElementWiseXor

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_BOOL_8	QNN_DATATYPE_BOOL_8	QNN_DATATYPE_BOOL_8

Constraints

Configuration	in[0]	in[1]
All	Shape: Max supported input rank is 4.	Shape: Max supported input rank is 4.

Elu

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ExtractGlimpse

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ExtractPatches

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

FullyConnected

Datatypes

Configuration	in[0]	in[1]	in[2]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_32, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	out[0]	keep_dims
All	Shape: Max supported input rank is 4.	Datatype: out[0] must be same Datatype as in[0]	Only support default value 0

Support

Configuration
All	Param keep_dims only supports default value

Gather

Datatypes

Configuration	in[0]	in[1]	in[2]	in[3]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_INT_32, QNN_DATATYPE_UINT_32	QNN_DATATYPE_INT_32	QNN_DATATYPE_INT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	in[1]	in[2]	in[3]
All	Shape: Max supported input rank is 4.	Value: Only supports index values in range [0, in[0].dim[axis] - 1]	optional	optional

GatherElements

Datatypes

Configuration	in[0]	in[1]	out[0]	axis
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_INT_32, QNN_DATATYPE_UINT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UINT_32

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

GatherNd

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_INT_32, QNN_DATATYPE_UINT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

GenerateProposals

Datatypes

Configuration	in[0]	in[1]	in[2]	in[3]	out[0]	out[1]	out[2]	bbox_xform_clip
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_INT_32	QNN_DATATYPE_BOOL_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

Support

Configuration
All	Param bbox_xform_clip only supports default value

HardSwish

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ImageProjectionTransform

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

InstanceNorm

Datatypes

Configuration	in[0]	in[1]	in[2]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	mode	region
All	Shape: Max supported input rank is 4.	Value: Only supports mode MU_SIGMA	Value: Only supports region ACROSS_SPATIAL

L2Norm

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

LayerNorm

Datatypes

Configuration	in[0]	in[1]	in[2]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8

Constraints

Configuration	in[0]	axes
All	Shape: Max supported input rank is 4.	Value: Only supports layer normalization on final dimension.

LogSoftmax

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	axis
All	Shape: Max supported input rank is 4.	Value: Only support value of rank-1

Lrn

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

Lstm

Datatypes

Configuration	in[0]	in[1]	in[2]	in[3]	in[4]	in[5]	in[6]	in[7]	in[8]	in[9]	in[10]	in[11]	in[12]	in[13]	in[14]	in[15]	in[16]	in[17]	in[18]	in[19]	in[20]	in[21]	in[22]	in[23]	in[24]	out[0]	out[1]	out[2]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_16	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_BOOL_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_16	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	in[1]	in[2]	in[3]	in[4]	in[5]	in[6]	in[7]	in[8]	in[9]	in[10]	in[11]	in[12]	in[13]	in[14]	in[15]	in[16]	in[17]	in[18]	in[19]	in[20]	in[21]	in[22]	in[23]	in[24]	out[0]	out[1]	out[2]	time_major
All	The input \(x_{t}\). Shape: Input rank must be 2.	input-to-forget weights \(W_{xf}\).	input-to-cell weights \(W_{xc}\).	input-to-output weights \(W_{xo}\).	recurrent-to-forget weights \(W_{hf}\).	recurrent-to-cell weights \(W_{hc}\).	recurrent-to-output weights \(W_{ho}\).	forget gate bias \(b_f\).	cell bias \(b_c\).	output gate bias \(b_o\).	output state (in) \(h_{t-1}\).	cell state (in) \(C_{t-1}\).	The input layer normalization weights. Used to rescale normalized inputs to activation at input gate.	The forget layer normalization weights. Used to rescale normalized inputs to activation at forget gate.	input-to-forget weights \(W_{xf}\)	The output layer normalization weights. Used to rescale normalized inputs to activation at output gate.	input-to-input weights \(W_{xi}\).	recurrent-to-input weights \(W_{hi}\).	cell-to-input weights \(W_{ci}\). It is a diagonal matrix by definition, and is expressed as a 1D vector.	cell-to-forget weights \(W_{cf}\). It is a diagonal matrix by definition, and is expressed as a 1D vector.	cell-to-output weights \(W_{co}\). It is a diagonal matrix by definition, and is expressed as a 1D vector.	input gate bias \(b_i\).	projection weights \(W_{proj}\).	projection bias \(b_{proj}\).	Reset for internal states. Only supports default.	output state (out) (\(h_{t}\)).	cell state (out) (\(C_{t}\)).	output (\(o_{t}\)). This is effectively the same as the current “output state (out)” value.	Not applicable for 2D input and is ignored.

MatMul

Datatypes

Configuration	in[0]	in[1]	in[2]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	in[1]	in[2]	out[0]
All	Shape: Max supported input rank is 4. Dynamic Shape: Dynamic dims not supported.	Shape: Rank must be equal to in[0] Shape: Batch dimensions (outermost N-2) must be equal to in[0] Dynamic Shape: Dynamic dims not supported.	Dynamic Shape: Dynamic dims not supported.	Shape: Rank must be equal to in[0] Shape: Batch dimensions (outermost N-2) must be equal to in[0] Dynamic Shape: Dynamic dims not supported.

Moments

Datatypes

Configuration	in[0]	out[0]	out[1]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

MultiClassNms

Datatypes

Configuration	in[0]	in[1]	out[0]	out[1]	out[2]	out[3]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_FLOAT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_INT_32	QNN_DATATYPE_UINT_32

Nv12ToRgb

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

Nv21ToRgb

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

Pack

Datatypes

Configuration	in[0..m]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0..m]
All	Shape: Max supported input rank is 3.

Pad

Datatypes

Configuration	in[0]	out[0]	pad_constant_value
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_FLOAT_32, QNN_DATATYPE_UINT_8, QNN_DATATYPE_INT_32

Constraints

Configuration	in[0]	pad_amount	pad_constant_value
All	Shape: Max supported input rank is 4.	MIRROR_SYMMETRIC and MIRROR_REFLECT type padding support padding on height (1) and width (2) axis only	pad_constant_value with QNN_DATATYPE_UINT_8 and QNN_DATATYPE_INT_32 are accepted with quantization param same as in[0] and the value provided is interpreted as the same data type as in[0].

PoolAvg2d

Datatypes

Configuration	in[0]	out[0]	rounding_mode
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UINT_32

Constraints

Configuration	in[0]	out[0]	count_pad_for_edges
All	Shape: Max supported input rank is 4.	Shape: Max supported output rank is 4.	Does not support PADDING_TYPE_VALID with explicit padding when count_pad_for_edges is zero

Support

Configuration
All	Param rounding_mode only supports default value

PoolAvg3d

Datatypes

Configuration	in[0]	out[0]	rounding_mode
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UINT_32

Constraints

Configuration	in[0]	out[0]	count_pad_for_edges
All	Shape: Max supported input rank is 5.	Shape: Max supported output rank is 5.	Does not support PADDING_TYPE_VALID with explicit padding when count_pad_for_edges is zero

Support

Configuration
All	Param rounding_mode only supports default value

PoolMax2d

Datatypes

Configuration	in[0]	out[0]	rounding_mode
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UINT_32

Constraints

Configuration	in[0]	out[0]
All	Shape: Max supported output rank is 4.	Shape: Max supported output rank is 4.

Support

Configuration
All	Param rounding_mode only supports default value

PoolMax3d

Datatypes

Configuration	in[0]	out[0]	rounding_mode
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UINT_32

Constraints

Configuration	in[0]	out[0]
All	Shape: Max supported output rank is 5.	Shape: Max supported output rank is 5.

Support

Configuration
All	Param rounding_mode only supports default value

Prelu

Datatypes

Configuration	in[0]	in[1]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	in[1]
All	Shape: Max supported input rank is 4.	Shape: Only support alpha with batch, height and width equal to 1.

Quantize

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_FLOAT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ReduceMax

Datatypes

Configuration	in[0]	out[0]	axes	keep_dims
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UINT_32	QNN_DATATYPE_BOOL_8

Constraints

Configuration	in[0]	out[0]
All	Shape: Max supported input rank is 4.	Shape: Max supported output rank is 4.

ReduceMean

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	out[0]
All	Shape: Max supported input rank is 4.	Shape: Max supported output rank is 4.

ReduceMin

Datatypes

Configuration	in[0]	out[0]	axes	keep_dims
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UINT_32	QNN_DATATYPE_BOOL_8

Constraints

Configuration	in[0]	out[0]
All	Shape: Max supported input rank is 4.	Shape: Max supported output rank is 4.

ReduceSum

Datatypes

Configuration	in[0]	out[0]	axes	keep_dims
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32	QNN_DATATYPE_UINT_32	QNN_DATATYPE_BOOL_8

Constraints

Configuration	in[0]	out[0]
All	Shape: Min supported input rank is 1, Max supported output rank is 4.	Datatype: out[0] must be same Datatype as in[0]

Relu

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 5.

Relu6

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

ReluMinMax

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

Reshape

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_FLOAT_32, QNN_DATATYPE_UINT_8, QNN_DATATYPE_INT_32, QNN_DATATYPE_UFIXED_POINT_16, QNN_DATATYPE_SFIXED_POINT_16	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_INT_32, QNN_DATATYPE_UINT_8, QNN_DATATYPE_UFIXED_POINT_16, QNN_DATATYPE_SFIXED_POINT_16

Constraints

Configuration	in[0]	in[1]	out[0]
All	Shape: Supports Rank in range [1, 5]. Input with zero size is not supported Dynamic Shape: Dynamic dims not supported.	This input is not supported.	Shape: Supports Rank in range [1, 5]. Dynamic Shape: Dynamic dims not supported.

Resize

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	out[0]
All	Shape: Max supported input rank is 4.	Shape: Max supported output rank is 4.

ResizeBilinear

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	out[0]	antialias
All	Shape: Max supported input rank is 4.	Shape: Max supported output rank is 4.	Only support default value 0

Support

Configuration
All	Param antialias only supports default value

ResizeNearestNeighbor

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	out[0]
All	Shape: Max supported input rank is 4.	Shape: Max supported output rank is 4.

RoiAlign

Datatypes

Configuration	in[0]	in[1]	in[2]	out[0]	img_size_ratio	num_samples_y	num_samples_x	aligned	allow_invalid_roi
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_UFIXED_POINT_16, QNN_DATATYPE_FLOAT_32	QNN_DATATYPE_UINT_32, QNN_DATATYPE_INT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_FLOAT_32	QNN_DATATYPE_INT_32	QNN_DATATYPE_INT_32	QNN_DATATYPE_BOOL_8	QNN_DATATYPE_BOOL_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

Support

Configuration
All	Param aligned only supports default value Param allow_invalid_roi only supports default value

RoiPooling

Datatypes

Configuration	in[0]	in[1]	in[2]	out[0]	img_size_ratio
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UINT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_FLOAT_32

Constraints

Configuration	in[0]	out[0]
All	Shape: Max supported input rank is 4.	Datatype: out[0] must be same Datatype as in[0]

Sigmoid

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

Softmax

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_FLOAT_32, QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

Support

Configuration
All	Param axis only supports default value

SpaceToBatch

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

SpaceToDepth

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	mode
All	Shape: Max supported input rank is 4.	Only support default mode QNN_OP_SPACE_TO_DEPTH_MODE_DCR

Support

Configuration
All	Param mode only supports default value

Split

Datatypes

Configuration	in[0]	out[0]…out[m]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	out[0]…out[m]
All	Shape: Max supported input rank is 4.	Shape: Max supported output rank is 4.

Squeeze

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	out[0]
All	Shape: Max supported input rank is 5.	Shape: Max supported output rank is 4. Shape: If rank of input is 5, the output rank should be 4.

StridedSlice

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 5.

Tanh

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

Tile

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]
All	Shape: Max supported input rank is 4.

TopK

Datatypes

Configuration	in[0]	out[0]	out[1]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_FLOAT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_FLOAT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_INT_32

Constraints

Configuration	in[0]	largest
All	Shape: Max supported input rank is 4.	Only support default value 1

Transpose

Datatypes

Configuration	in[0]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	out[0]
All	Shape: Max supported input rank is 5.	Shape: Same Rank as in[0].

TransposeConv2d

Datatypes

Configuration	in[0]	in[1]	in[2]	out[0]
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_32	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8

Constraints

Configuration	in[0]	out[0]
All	Shape: Max supported input rank is 4.	Only supports encodingDefinition as QNN_DEFINITION_DEFINED

UnPack

Datatypes

Configuration	in[0]	out[0..m]	axis
All	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UFIXED_POINT_8, QNN_DATATYPE_SFIXED_POINT_8	QNN_DATATYPE_UINT_32

Constraints

Configuration	in[0]	out[0..m]	axis
All	Shape: Min supported input rank is 1, Max supported input rank is 4	Datatype: out[0] must be same Datatype as in[0]	Specifies axis to unpack on.