Op Definition Revision History

Op Set Version

SDK Version

Description

2.7.0

2.39.0

  • Added HadamardTransform Op. Not initially supported.

2.6.0

2.38.0

  • Added RandomUniformLike Op to support ONNX RandomUniformLike. Initially supported on CPU and HTP backend.

  • Added STFT Op to support ONNX STFT. Initially supported on CPU and HTP backend.

2.5.0

2.37.0

  • Added IsNan Op to support ONNX IsNan. Initially supported on CPU and HTP backend.

  • Added Logit Op to support ONNX Logit. Initially supported on CPU and HTP backend.

2.4.0

2.36.0

  • Added new reuse_sparse_indicies param to Conv2d to support tensor sparsity.

2.3.0

2.35.0

  • Added IsInf Op.

2.2.0

2.34.0

  • Added buffer_padding parameter to Buffer Op.

  • Add dynamic shape support for FullyConnected.

2.1.0

2.30.0

  • Added dynamic shape support for ExpandDims, ReduceMean, Squeeze, and StridedSlice.

2.0.0

2.29.0

  • Modify NonMaxSuppression to produce indices per batch. Note this is backwards incompatible change.

1.38.0

2.28.0

  • Added dynamic shape support for Cast, Concat, Conv2d, Convert, ElementWiseAdd, ElementWiseBinary, ElementWiseDivide, ElementWiseMultiply, ElementWisePower, ElementWiseSquareRoot, ElementWiseSubtract, ElementWiseUnary, Gather, ReduceMean, RmsNorm, and StridedSlice.

1.37.0

2.27.0

  • Added dynamic shape support for ElementWiseNeuron, Sigmoid, Softmax, and Transpose.

1.36.0

2.26.0

  • Added antialias param for ResizeBilinear.

1.35.0

2.25.0

  • Added op definition for RmsNorm.

  • Added largest parameter to Topk.

1.34.0

2.24.0

  • Added op definition for Im2Col and Col2Im.

  • Added dynamic tensor support for MatMul and Reshape ops.

1.33.0

2.23.0

  • Added op definition for CombinedNMS.

  • Updated Buffer Op definition for multi-frame support.

1.32.0

2.22.0

  • Added op definition for ReduceSumSquare.

  • Added 0-D support for ElementWiseBinary and Reshape ops.

1.31.0

2.21.0

  • Added op definition for Buffer.

  • Added optional reset input to GRU op.

  • Updated mask tensor input description for the MaskedSoftmax op.

  • Added QNN_DATATYPE_SFIXED_POINT_4 and QNN_DATATYPE_UFIXED_POINT_4 support for Quantize and Dequantize ops.

1.30.0

2.20.0

  • Added new ops to support tensor sparsity: CreateSparse, GetSparseIndices, GetSparseValues, SparseToDense

  • Allow sparse tensors usage in Conv3d, BatchNorm, ElementWiseAdd, and Relu.

  • Added new reuse_sparse_indicies param to Conv3d to support tensor sparsity.

  • Added support for single batch input and output in DistributeFPNProposals op.

1.29.0

2.19.0

  • Fix issue where constraint for split_index parameter for Split op allowed creation of empty output tensors.

1.28.0

2.18.0

  • Added op definition for MaskedSoftmax.

  • Added support for optional param time_major and support for multi time-step input and output in LSTM op.

  • Added support for optional param time_major in GRU op.

  • Added support for negative index values in the Gather op.

1.27.0

2.16.0

  • Fixed QNN_OP_GRID_SAMPLE_PADDING_MODE_REFLECTION example in GridSample Op.

  • Fixed index value constraint range for in[1] in Gather Op.

1.26.0

2.15.0

  • Added If op definition.

  • Clarified behavior with regards to how the parameters normalize and centered affect glimpse window in ExtractGlimpse.

  • Added support for cubic as an interpolation mode in Resize Op.

1.25.0

2.14.0

  • Added optional parameter mode to SpaceToDepth.

  • Updated BatchPermutation Op to use shape of in[1] to determine batch dimension of out[0]. Relaxed constraint on index values of in[1].

  • Updated constraint of out[0] index values to be based on FPN levels for DistributeFpnProposals.

1.24.0

2.13.0

  • Added Conv1d, DepthWiseConv1d, and TransposeConv1d op definitions.

  • Added ElementWiseXor op definition and added XOR option to ElementWiseBinary.

1.23.0

2.12.0

  • Added ElementWiseBinary op definition.

  • Added GroupNorm op definition.

  • Added optional parameter weights and optional output batch splits to AxisAlignedBboxTransform.

  • Added optional input batch splits and optional outputs batch splits, keeps, and keeps size to BoxWithNmsLimit.

  • DistributeFpnProposals rois_idx_restore (out[0]) will output -1 for invalid ROIs.

  • Added optional parameter bbox_xform_clip to GenerateProposals.

  • Added optional parameters aligned and allow_invalid_roi to RoiAlign.

1.22.0

2.11.0

  • Added ElementWiseNeuron op definition.

1.21.0

2.10.0

  • Added ElementWiseAsin, ElementWiseFmod and ElementWiseMod op definitions.

1.20.0

2.9.0

  • Constrained all ND inputs to have a rank greater than 0.

  • Added constraint for dilation > 0 in Convolution Ops.

  • Added ElementWiseUnary op definition.

1.19.1

2.8.2

  • Added NonMaxSuppression op definition.

1.19.0

2.8.0

  • Added ElementWiseSign op definition.

  • Added QNN_DATATYPE_UINT_32 support for Gather in[1].

1.18.0

2.7.0

  • Added ExtractPatches op definition.

  • Added QNN_DATATYPE_INT_32 support for GatherNd in[1].

1.17.0

2.6.0

  • Added NonZero, CollectRpnProposals, and DistributeFpnProposals op definitions.

  • Added QNN_DATATYPE_INT_32 support for CropAndResize in[2] and ScatterNd in[1].

  • Added support for broadcasting in ElementWiseLess Op.

1.16.0

2.5.0

  • Added BatchPermutation op definition.

  • Added QNN_DATATYPE_INT_32 support for ArgMax out[0], ArgMin out[0], and GatherElements in[1].

1.15.0

2.4.0

  • Added ScatterElements op definition.

  • Remove incorrect shape constraints for Tile out[0] and multiples param.

  • Added support for broadcasting in ElementwiseSelect Op.

1.14.0

2.2.0

  • Added ConstantOfShape op definition.

  • Added GridSample op definition.

  • Added PoolMax3d op definition.

  • Added Shape op definition.

  • Add 32-bit data types for Quantize and Dequantize.

  • Extend OneHot to support N-dimensional inputs.

  • Add optional bias input tensor for MatMul.

  • Remove Prelu in[1] data type constraint.

1.13.0

2.1.0

  • Added GRU op definition.

  • Added constraint that Conv2d and TransposeConv2d strides must be positive.

1.12.0

2.0.0

  • Initial independent version of op definition set to match prior API version.