blaize¶
Namespaces¶
| Name |
|---|
| blaize::isp |
| blaize::onnx |
| blaize::vx |
Classes¶
| Name | |
|---|---|
| class | blaize::ConfigurationValue |
| struct | blaize::float32x4 |
| struct | blaize::int16x4 |
| struct | blaize::int32x4 |
| struct | blaize::int8x4 |
| struct | blaize::uint16x4 |
| struct | blaize::uint32x4 |
| struct | blaize::uint8x4 |
Types¶
| Name | |
|---|---|
| enum | @909 |
| enum | OnnxStatus |
| enum class | TensorType |
| enum class | ShiftDirection |
| enum class | InterpolationMode |
| enum class | CoordinateTransformationMode |
| enum class | NearestMode |
| enum class | DepthToSpaceMode |
| enum class | RoiAlignMode |
| using int8_t | int8 |
| using uint8_t | uint8 |
| using int16_t | int16 |
| using uint16_t | uint16 |
| using int32_t | int32 |
| using uint32_t | uint32 |
| using int64_t | int64 |
| using uint64_t | uint64 |
| using float | float32 |
| using double | float64 |
| using vx_node | Node |
| using vx_graph | Graph |
| using vx_kernel | Kernel |
| using vx_parameter | Parameter |
| using vx_context | Context |
| using vx_tensor | Tensor |
| using vx_scalar | Scalar |
| using vx_image | Image |
| using vx_status | Status |
| using vx_reference | Reference |
| using vx_tensor_view | TensorView |
| using std::map< std::string, ConfigurationValue > | ConfigurationMapping |
Functions¶
| Name | |
|---|---|
| Context | CreateContext() Creates a context. |
| Status | SetConfiguration(Context c, std::string name, ConfigurationValue value) Change configuration item. |
| ConfigurationMapping | GetConfiguration(Context c) Get current configuration state. |
| Graph | CreateGraph(Context c) |
| Status | ProcessGraph(Graph g) |
| Status | ReleaseGraph(Graph * g) |
| Status | ReleaseContext(Context * c) |
| Status | GetStatus(Reference obj) |
| Status | AddParameterToGraph(Graph g, Parameter p) |
| Status | SetGraphParameterByIndex(Graph g, uint32 index, Reference obj) |
| Parameter | GetParameterByIndex(Node n, uint32 index) |
| Status | ReleaseParameter(Parameter * p) |
| Status | ReleaseNode(Node * n) |
| Status | ReleaseTensor(Tensor * t) |
| Status | SetReferenceName(Reference obj, const char * name) |
| Tensor | CreateTensorFromRawFile(Context c, const char * filename, int32 num_of_dims, uint32x4 sizes, TensorType data_format) |
| Tensor | CreateTensor(Context c, int32 num_of_dims, uint32x4 sizes, TensorType data_format) |
| Tensor | CreateVirtualTensor(Graph g, int32 num_of_dims, uint32x4 sizes, TensorType data_format) |
| Tensor | CreateWeightsTensorFromRawFile(Context c, const char * filename, uint32x4 sizes, TensorType data_format) |
| Status | SaveTensorToRawFile(Tensor t, const char * filename) |
| TensorView | CreateTensorView(Context c, uint32x4 start, uint32x4 end, int32 num_of_dims) |
| Tensor | CreateTensorFromView(Tensor t, TensorView view) |
| Node | AbsNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) AbsNode operator performs element-wise Absolute operation. |
| Node | AddNode(Graph graph, Tensor a, Tensor b, Tensor c, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) |
| Node | AveragePoolNode(Graph graph, Tensor a, Tensor b, uint8 ceil_mode =0, uint8x4 kernel_shape ={1, 1, 0, 0}, uint8x4 pads ={0, 0, 0, 0}, uint8x4 strides ={1, 1, 1, 1}, uint8 count_include_pad =0, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) AveragePool operator performs element-wise average pooling operation on subset of tensor. |
| Node | BatchNormalizationNode(Graph graph, Tensor a, Tensor b, Tensor weight, Tensor bias, Tensor weight_scale, float32 a_scale =0.0f, int32 a_zero_point =0, float32 bias_scale =0.0f, int32 bias_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, Tensor var =nullptr, Tensor mean =nullptr) BatchNorm operator performs element-wise max operation on subset of tensor. |
| Node | BitShiftNode(Graph g, Tensor a, Tensor b, Tensor c, ShiftDirection dir, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) Bitwise shift operator performs element-wise operation. |
| Node | EluNode(Graph graph, Tensor a, Tensor b, float32 alpha =1.0f, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) Elu operator performs element-wise function on input tensor. |
| Node | GemmNode(Graph graph, Tensor x, Tensor y, Tensor w, Tensor b =nullptr, float32 alpha =1.0f, float32 beta =1.0f, int32 transX =0, int32 transY =0, float32 x_scale =0.0f, int32 x_zero_point =0, float32 y_scale =0.0f, int32 y_zero_point =0, float32 w_scale =0.0f, int32 w_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, uint8 shift_flag =1, uint8x4 broadcast_axis ={0, 0, 0, 0}) |
| Node | GlobalAveragePoolNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) GlobalAveragePool operator performs global average pooling operation on subset of tensor. |
| Node | LeakyReluNode(Graph graph, Tensor a, Tensor b, float32 alpha =1.0f, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 alpha_scale =0.0f, int32 alpha_zero_point =0) |
| Node | LpNormalizationNode(Graph g, Tensor a, Tensor b, int32 p =2, int32 axis =-1, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, Tensor weight =nullptr, float32 weight_scale =0.0f, int32 weight_zero_point =0) |
| Node | LRNNode(Graph g, Tensor a, Tensor b, uint32 size, float32 alpha =0.0001, float32 beta =0.75, float32 bias =1.0, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) |
| Node | MatMulNode(Graph graph, Tensor a, Tensor b, Tensor c, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0) |
| Node | MaxPoolNode(Graph graph, Tensor a, Tensor b, uint8 ceil_mode =0, uint8x4 dilations ={1, 1, 1, 1}, uint8x4 kernel_shape ={1, 1, 0, 0}, uint8x4 pads ={0, 0, 0, 0}, uint8x4 strides ={1, 1, 1, 1}, uint8 storage_order =0, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) MaxPool operator performs element-wise max operation on subset of tensor. |
| Node | MulNode(Graph graph, Tensor a, Tensor b, Tensor c, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) |
| Node | QLinearConvNode(Graph graph, Tensor x, Tensor y, Tensor w, Tensor b =nullptr, uint8x4 dilations ={1, 1, 1, 1}, uint32 group =1, uint8x4 kernel_shape ={0, 0, 0, 0}, uint8x4 pads ={0, 0, 0, 0}, uint8x4 strides ={1, 1, 1}, float32 x_scale =0.0f, int32 x_zero_point =0, float32 y_scale =0.0f, int32 y_zero_point =0, float32 w_scale =0.0f, int32 w_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, uint8 shift_flag =1) |
| Node | QLinearConvNode2(Graph graph, Tensor x, Tensor y, Tensor w, Tensor b =nullptr, uint8x4 dilations ={1, 1, 1, 1}, uint32 group =1, uint8x4 kernel_shape ={0, 0, 0, 0}, uint8x4 pads ={0, 0, 0, 0}, uint8x4 strides ={1, 1, 1}, float32 x_scale =0.0f, int32 x_zero_point =0, Tensor y_scale =nullptr, int32 y_zero_point =0, float32 w_scale =0.0f, int32 w_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, uint8 shift_flag =1) |
| Node | ConvTransposeNode(Graph graph, Tensor x, Tensor y, Tensor w, Tensor b =nullptr, uint8x4 dilations ={1, 1, 1, 1}, uint32 group =1, uint8x4 kernel_shape ={0, 0, 0, 0}, uint8x4 output_padding ={0, 0, 0, 0}, uint8x4 pads ={0, 0, 0, 0}, uint8x4 strides ={1, 1, 1}, float32 x_scale =0.0f, int32 x_zero_point =0, float32 y_scale =0.0f, int32 y_zero_point =0, float32 w_scale =0.0f, int32 w_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, uint8 shift_flag =1) |
| Node | ReluNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) |
| Node | ResizeNode(Graph graph, Tensor a, Tensor b, float32x4 scales, Tensor roi, InterpolationMode mode =InterpolationMode::NEAREST, CoordinateTransformationMode transf_mode =CoordinateTransformationMode::HALF_PIXEL, NearestMode nearest_mode =NearestMode::ROUND_PREFER_FLOOR, int32 exclude_outside =0, float32 extrapolation_value =0.0f, float32 cubic_coeff_a =-0.75f, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) Onnx Resize-11 supported. |
| Node | ReshapeNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) |
| Node | SoftmaxNode(Graph graph, Tensor a, Tensor b, int32 axis =-1, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) |
| Node | SpaceToDepthNode(Graph graph, Tensor a, Tensor b, int32 blocksize, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) |
| Node | TransposeNode(Graph graph, Tensor input, Tensor out, uint8x4 perm ={0, 1, 2, 3}, float32 input_scale =0.0f, int32 input_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0) |
| Node | SeluNode(Graph graph, Tensor a, Tensor b, float32 alpha =1.67326f, float32 gamma =1.0507f, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) |
| Node | HardSigmoidNode(Graph graph, Tensor a, Tensor b, float32 alpha =0.2f, float32 beta =0.5f, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) |
| Node | AndNode(Graph graph, Tensor in0, Tensor in1, Tensor out, float32 in0_scale =0.0f, int32 in0_zero_point =0, float32 in1_scale =0.0f, int32 in1_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) AndNode operator performs element-wise Absolute operation. |
| Node | SubNode(Graph graph, Tensor in0, Tensor in1, Tensor out, float32 in0_scale =0.0f, int32 in0_zero_point =0, float32 in1_scale =0.0f, int32 in1_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) SubNode operator performs element-wise Absolute operation. |
| Node | Sum3Node(Graph graph, Tensor in0, Tensor in1, Tensor in2, Tensor out, float32 in0_scale =0.0f, int32 in0_zero_point =0, float32 in1_scale =0.0f, int32 in1_zero_point =0, float32 in2_scale =0.0f, int32 in2_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) SumNode(3-input) operator performs element-wise Absolute operation. |
| Node | Sum4Node(Graph graph, Tensor in0, Tensor in1, Tensor in2, Tensor in3, Tensor out, float32 in0_scale =0.0f, int32 in0_zero_point =0, float32 in1_scale =0.0f, int32 in1_zero_point =0, float32 in2_scale =0.0f, int32 in2_zero_point =0, float32 in3_scale =0.0f, int32 in3_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) SumNode(3-input) operator performs element-wise Absolute operation. |
| Node | TanhNode(Graph graph, Tensor input, Tensor out, float32 input_scale =0.0f, int32 input_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0) TanhNode operator performs element-wise Absolute operation. |
| Node | DepthToSpaceNode(Graph graph, Tensor a, Tensor b, int32 blocksize, DepthToSpaceMode mode =DepthToSpaceMode::DCR, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) DepthToSpace operator performs bitwise DepthToSpace conversion. |
| Node | SliceNode(Graph graph, Tensor input, Tensor output, int32x4 starts, int32x4 ends, int32x4 axes, int32x4 steps ={1, 1, 1, 1}, float32 input_scale =0.0f, int32 input_zero_point =0, float32 output_scale =0.0f, int32 output_zero_point =0) Slice operator performs bitwise slice operation. |
| Node | Split2Node(Graph graph, Tensor input, Tensor output0, Tensor output1, int32x4 split ={0, 0, 0, 0}, int32 axis =0, float32 input_scale =0.0f, int32 input_zero_point =0, float32 output0_scale =0.0f, int32 output0_zero_point =0, float32 output1_scale =0.0f, int32 output1_zero_point =0) Split operator performs bitwise split operation. |
| Node | Split3Node(Graph graph, Tensor input, Tensor output0, Tensor output1, Tensor output2, int32x4 split ={0, 0, 0, 0}, int32 axis =0, float32 input_scale =0.0f, int32 input_zero_point =0, float32 output0_scale =0.0f, int32 output0_zero_point =0, float32 output1_scale =0.0f, int32 output1_zero_point =0, float32 output2_scale =0.0f, int32 output2_zero_point =0) Split operator performs bitwise split operation. |
| Node | Split4Node(Graph graph, Tensor input, Tensor output0, Tensor output1, Tensor output2, Tensor output3, int32x4 split ={0, 0, 0, 0}, int32 axis =0, float32 input_scale =0.0f, int32 input_zero_point =0, float32 output0_scale =0.0f, int32 output0_zero_point =0, float32 output1_scale =0.0f, int32 output1_zero_point =0, float32 output2_scale =0.0f, int32 output2_zero_point =0, float32 output3_scale =0.0f, int32 output3_zero_point =0) Split operator performs bitwise split operation. |
| Node | ReduceMaxNode(Graph graph, Tensor a, Tensor b, int8x4 axes, int32 keepdim =1, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) ReduceMax onnx Operator. |
| Node | ReduceMinNode(Graph graph, Tensor a, Tensor b, int8x4 axes, int32 keepdim =1, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) ReduceMin onnx Operator. |
| Node | ReduceMeanNode(Graph graph, Tensor a, Tensor b, int8x4 axes, int32 keepdim =1, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) ReduceMean onnx Operator. |
| Node | ReduceSumNode(Graph graph, Tensor a, Tensor b, int8x4 axes, int32 keepdim =1, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) ReduceSum onnx Operator. |
| Node | ClipNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 min =1.17549435e-38, float32 max =3.38953139e+38) Clip onnx Operator. |
| Node | Max2Node(Graph graph, Tensor a, Tensor b, Tensor c, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) Max2 onnx Operator. |
| Node | Max3Node(Graph graph, Tensor a, Tensor b, Tensor c, Tensor d, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, float32 d_scale =0.0f, int32 d_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) Max3 onnx Operator. |
| Node | Max4Node(Graph graph, Tensor a, Tensor b, Tensor c, Tensor d, Tensor e, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, float32 d_scale =0.0f, int32 d_zero_point =0, float32 e_scale =0.0f, int32 e_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) Max4 onnx Operator. |
| Node | Mean2Node(Graph graph, Tensor a, Tensor b, Tensor c, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) Mean2 onnx Operator. |
| Node | Mean3Node(Graph graph, Tensor a, Tensor b, Tensor c, Tensor d, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, float32 d_scale =0.0f, int32 d_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) Mean3 onnx Operator. |
| Node | Mean4Node(Graph graph, Tensor a, Tensor b, Tensor c, Tensor d, Tensor e, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, float32 d_scale =0.0f, int32 d_zero_point =0, float32 e_scale =0.0f, int32 e_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) Mean4 onnx Operator. |
| Node | LessNode(Graph graph, Tensor a, Tensor b, Tensor c, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) |
| Node | SigmoidNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) |
| Node | XorNode(Graph graph, Tensor a, Tensor b, Tensor c, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) |
| Node | ShrinkNode(Graph graph, Tensor a, Tensor b, float bias =0.0f, float lambd =0.5, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) |
| Node | OrNode(Graph graph, Tensor a, Tensor b, Tensor c, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) |
| Node | GreaterNode(Graph graph, Tensor a, Tensor b, Tensor c, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) |
| Node | NotNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) |
| Node | NegNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) |
| Node | Min2Node(Graph graph, Tensor a, Tensor b, Tensor c, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) Min2 onnx Operator. |
| Node | Min3Node(Graph graph, Tensor a, Tensor b, Tensor c, Tensor d, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, float32 d_scale =0.0f, int32 d_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) Min3 onnx Operator. |
| Node | Min4Node(Graph graph, Tensor a, Tensor b, Tensor c, Tensor d, Tensor e, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, float32 d_scale =0.0f, int32 d_zero_point =0, float32 e_scale =0.0f, int32 e_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) Min4 onnx Operator. |
| Node | DivNode(Graph graph, Tensor a, Tensor b, Tensor c, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) Div onnx Operator. |
| Node | ThresholdedReluNode(Graph graph, Tensor a, Tensor b, float32 alpha =1.0f, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) ThresholdedReluNode operator performs element-wise function on input tensor. |
| Node | PReluNode(Graph graph, Tensor a, Tensor b, Tensor c, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) PRelu onnx Operator. |
| Node | SqueezeNode(Graph graph, Tensor input, Tensor out, int8x4 axes ={4, 4, 4, 4}, float32 input_scale =0.0f, int32 input_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0) |
| Node | UnsqueezeNode(Graph graph, Tensor input, Tensor out, int8x4 axes, float32 input_scale =0.0f, int32 input_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0) |
| Node | GatherElementsNode(Graph graph, Tensor input, Tensor out, Tensor index, int32 axis =0, float32 input_scale =0.0f, int32 input_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0, float32 index_scale =0.0f, int32 index_zero_point =0) |
| Node | ReduceProdNode(Graph graph, Tensor a, Tensor b, int8x4 axes, int32 keepdim =1, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) ReduceProd onnx Operator. |
| Node | ReduceSumSquareNode(Graph graph, Tensor a, Tensor b, int8x4 axes, int32 keepdim =1, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) ReduceSumSquare onnx Operator. |
| Node | SoftsignNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) SoftSign onnx Operator. |
| Node | EqualNode(Graph graph, Tensor in1, Tensor in2, Tensor out, float32 in1_scale =0.0f, int32 in1_zero_point =0, float32 in2_scale =0.0f, int32 in2_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) Equal onnx Operator. |
| Node | ModNode(Graph graph, Tensor in1, Tensor in2, Tensor out, int8 fmod =0, float32 in1_scale =0.0f, int32 in1_zero_point =0, float32 in2_scale =0.0f, int32 in2_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) Mod onnx Operator. |
| Node | IdentityNode(Graph graph, Tensor input, Tensor output, float32 input_scale =0.0f, int32 input_zero_point =0, float32 output_scale =0.0f, int32 output_zero_point =0) Identity onnx Operator. |
| Node | ScatterElementsNode(Graph graph, Tensor data, Tensor update, Tensor out, Tensor indices, int axis =1, float32 data_scale =0.0f, int32 data_zero_point =0, float32 update_scale =0.0f, int32 update_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0, float32 indices_scale =0.0f, int32 indices_zero_point =0) ScatterElements onnx Operator. |
| Node | SqrtNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) Sqrt operator performs element-wise sqrt operation. |
| Node | ReciprocalNode(Graph graph, Tensor input, Tensor out, float32 input_scale =0.0f, int32 input_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0) |
| Node | SinNode(Graph graph, Tensor input, Tensor out, float32 input_scale =0.0f, int32 input_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0) |
| Node | CosNode(Graph graph, Tensor input, Tensor out, float32 input_scale =0.0f, int32 input_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0) |
| Node | Sum2Node(Graph graph, Tensor in0, Tensor in1, Tensor out, float32 in0_scale =0.0f, int32 in0_zero_point =0, float32 in1_scale =0.0f, int32 in1_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) SumNode(2-input) operator performs element-wise Absolute operation. |
| Node | HardmaxNode(Graph graph, Tensor a, Tensor b, int32 axis =1, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) Hardmax onnx Operator. |
| Node | ArgMaxNode(Graph graph, Tensor a, Tensor b, int32 axis =0, int32 keepdims =1, int32 select_last_index =0, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) Argmax onnx Operator. |
| Node | ArgMinNode(Graph graph, Tensor a, Tensor b, int32 axis =0, int32 keepdims =1, int32 select_last_index =0, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) Argmin onnx Operator. |
| Node | CoshNode(Graph graph, Tensor input, Tensor out, float32 input_scale =0.0f, int32 input_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0) CoshNode operator performs element-wise Absolute operation. |
| Node | AcoshNode(Graph graph, Tensor input, Tensor out, float32 input_scale =0.0f, int32 input_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0) ACoshNode operator performs element-wise Absolute operation. |
| Node | AcosNode(Graph graph, Tensor input, Tensor out, float32 input_scale =0.0f, int32 input_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0) AcosNode operator performs element-wise Absolute operation. |
| Node | AtanhNode(Graph graph, Tensor input, Tensor out, float32 input_scale =0.0f, int32 input_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0) AtanhNode operator performs element-wise Absolute operation. |
| Node | AtanNode(Graph graph, Tensor input, Tensor out, float32 input_scale =0.0f, int32 input_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0) AtanNode operator performs element-wise Absolute operation. |
| Node | AsinNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) Asin onnx Operator. |
| Node | AsinhNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) Asinh onnx Operator. |
| Node | SinhNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) Sinh onnx Operator. |
| Node | ExpNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) Exp onnx Operator. |
| Node | TanNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) Tan onnx Operator. |
| Node | PowNode(Graph graph, Tensor a, Tensor b, Tensor c, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 c_scale =0.0f, int32 c_zero_point =0, uint8x4 broadcast_axis ={0, 0, 0, 0}) Pow onnx Operator. |
| Node | ExpandNode(Graph graph, Tensor input, Tensor out, float32 input_scale =0.0f, int32 input_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0) Expand onnx Operator. |
| Node | SoftplusNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) SoftplusNode operator performs element-wise Softplus operation. |
| Node | ReduceLogSumNode(Graph graph, Tensor a, Tensor b, int8x4 axes, int32 keepdim =1, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) ReduceLogSum onnx Operator. |
| Node | ReduceLogSumExpNode(Graph graph, Tensor a, Tensor b, int8x4 axes, int32 keepdim =1, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) ReduceLogSumExp onnx Operator. |
| Node | ReduceL1Node(Graph graph, Tensor a, Tensor b, int8x4 axes, int32 keepdim =1, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) ReduceL1 onnx Operator. |
| Node | ReduceL2Node(Graph graph, Tensor a, Tensor b, int8x4 axes, int32 keepdim =1, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) ReduceL2 onnx Operator. |
| Node | LogNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) LogNode onnx Operator. |
| Node | GlobalMaxPoolNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) GlobalMaxPool onnx Operator. |
| Node | LpPoolNode(Graph graph, Tensor a, Tensor b, uint8x4 kernel_shape ={1, 1}, int p =2, uint8x4 pads ={0, 0}, uint8x4 strides ={1, 1}, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) LpPool onnx Operator. |
| Node | GlobalLpPoolNode(Graph graph, Tensor a, Tensor b, int p =2, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) GlobalLpPool onnx Operator. |
| Node | GatherNode(Graph graph, Tensor input, Tensor out, Tensor index, int32 axis =0, float32 input_scale =0.0f, int32 input_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0, float32 index_scale =0.0f, int32 index_zero_point =0) |
| Node | LogSoftmaxNode(Graph graph, Tensor a, Tensor b, int32 axis =-1, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) LogSoftmax onnx Operator. |
| Node | MaxUnpoolNode(Graph graph, Tensor a, Tensor b, Tensor c, uint8x4 kernel_shape ={1, 1}, uint8x4 pads ={0, 0, 0, 0}, uint8x4 strides ={1, 1}, float32 a_scale =0, int32 a_zero_point =0, float32 c_scale =0, int32 c_zero_point =0) MaxUnpool onnx Operator. |
| Node | InstanceNormalizationNode(Graph graph, Tensor input, Tensor out, Tensor s, Tensor b, float epsilon =0.00001f, float32 input_scale =0.0f, int32 input_zero_point =0, float32 out_scale =0.0f, int32 out_zero_point =0, float32 s_scale =0.0f, int32 s_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) InstanceNormalization onnx Operator. |
| Node | CumSumNode(Graph graph, Tensor a, Tensor b, int32 axis =0, int8 exclusive =0, int8 reverse =0, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) CumSum onnx Operator. |
| Node | OneHotNode(Graph graph, Tensor a, Tensor b, int32 axis =-1, int32 depth =0, float32 off_value =0.0f, float32 on_value =0.0f, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) |
| Node | ErfNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) |
| Node | GatherNDNode(Graph graph, Tensor a, Tensor b, Tensor indices, int32 batch_dim =0, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0, float32 indices_scale =0.0f, int32 indices_zero_point =0) Gather onnx Operator. |
| Node | ScatterNDNode(Graph graph, Tensor input, Tensor updates, Tensor out, Tensor indices, float32 input_scale =0.0f, int32 input_zero_point =0, float32 updates_scale =0.0f, int updates_zero_point =0, float32 out_scale =0.0f, int out_zero_point =0, float32 indices_scale =0.0f, int indices_zero_point =0) |
| Node | SignNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) Sign onnx Operator. |
| Node | NonMaxSuppressionNode(Graph graph, Tensor in_boxes, Tensor scores, Tensor out_selected_indices, int32 center_point_box =0, int32 max_output_boxes_per_class =0, float iou_threshold =0, float score_threshold =0, float32 in_boxes_scale =0.0f, int32 in_boxes_zero_point =0, float32 scores_scale =0.0f, int32 scores_zero_point =0, float32 out_selected_indices_scale =0.0f, int32 out_selected_indices_zero_point =0, bool with_sort =true) NonMaxSuppression Operator. |
| Node | MaxRoiPoolNode(Graph graph, Tensor input, Tensor rois, Tensor output, int32 pooled_shape_height, int32 pooled_shape_width, float32 spatial_scale_factor =1.0f, float32 input_scale =0.0f, int32 input_zero_point =0, float32 rois_scale =0.0f, int32 rois_zero_point =0, float32 output_scale =0.0f, int32 output_zero_point =0) MaxRoiPoolNode. |
| Node | RoiAlignNode(Graph graph, Tensor input, Tensor rois, Tensor output, RoiAlignMode mode =RoiAlignMode::AVG, int32 output_height =1, int32 output_width =1, int32 sampling_ratio =0, float32 spatial_scale_factor =1.0, float32 input_scale =0, int32 input_zero_point =0, float32 output_scale =0, int32 output_zero_point =0, float32 rois_scale =0, int32 rois_zero_point =0) RoiAlignNode. |
| Node | CeilNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) Ceil onnx Operator. |
| Node | FloorNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) Floor onnx Operator. |
| Node | RoundNode(Graph graph, Tensor a, Tensor b, float32 a_scale =0, int32 a_zero_point =0, float32 b_scale =0, int32 b_zero_point =0) Round onnx Operator. |
| Node | IsNaNNode(Graph graph, Tensor input, Tensor output, float32 input_scale =0.0f, int32 input_zero_point =0, float32 output_scale =0.0f, int32 output_zero_point =0) IsNan onnx Operator. |
| Node | IsInfNode(Graph graph, Tensor input, Tensor output, int32 detect_negative, int32 detect_positive, float32 input_scale =0.0f, int32 input_zero_point =0, float32 output_scale =0.0f, int32 output_zero_point =0) IsInf onnx Operator. |
| Node | Concat2Node(Graph graph, Tensor input1, Tensor input2, Tensor output, int axis, float32 input1_scale =0.0f, int32 input1_zero_point =0, float32 input2_scale =0.0f, int32 input2_zero_point =0, float32 output_scale =0.0f, int32 output_zero_point =0) Concate2 onnx Operator. |
| Node | Concat3Node(Graph graph, Tensor input1, Tensor input2, Tensor input3, Tensor output, int axis, float32 input1_scale =0.0f, int32 input1_zero_point =0, float32 input2_scale =0.0f, int32 input2_zero_point =0, float32 input3_scale =0.0f, int32 input3_zero_point =0, float32 output_scale =0.0f, int32 output_zero_point =0) Concate3 onnx Operator. |
| Node | Concat4Node(Graph graph, Tensor input1, Tensor input2, Tensor input3, Tensor input4, Tensor output, int32 axis, float32 input1_scale =0.0f, int32 input1_zero_point =0, float32 input2_scale =0.0f, int32 input2_zero_point =0, float32 input3_scale =0.0f, int32 input3_zero_point =0, float32 input4_scale =0.0f, int32 input4_zero_point =0, float32 output_scale =0.0f, int32 output_zero_point =0) Concate4 onnx Operator. |
| uint32_t | GetInputParamIndex(Node node, int32 param_num) Get the index of the input parameter param_num for given node. |
| uint32_t | GetOutputParamIndex(Node node) Get the index of the output parameter for given node. |
| Node | CustomNodeFromKernel(Graph graph, Kernel kernel, std::vector< Tensor > outputs, std::vector< Tensor > inputs, std::vector< Scalar > scalars ={}, std::vector< int32 > input_zero_points ={}, std::vector< float32 > input_scales ={}, std::vector< int32 > output_zero_points ={}, std::vector< float32 > output_scales ={}, std::string kernel_file_name ={}) Create a custom node from an OpenVX kernel. |
| Node | ResizeNode(Graph graph, Tensor a, Tensor b, uint32x4 scales, InterpolationMode mode =InterpolationMode::NEAREST, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) Onnx Resize-10 supported. |
| Node | ResizeNode(Graph graph, Tensor a, Tensor b, float32x4 scales, uint32x4 roi_start, uint32x4 roi_end, InterpolationMode mode =InterpolationMode::NEAREST, CoordinateTransformationMode transf_mode =CoordinateTransformationMode::HALF_PIXEL, NearestMode nearest_mode =NearestMode::ROUND_PREFER_FLOOR, int32 exclude_outside =0, float32 extrapolation_value =0.0f, float32 cubic_coeff_a =-0.75f, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) |
| Node | ResizeNode(Graph graph, Tensor a, Tensor b, uint32x4 scales, uint32x4 roi_start, uint32x4 roi_end, InterpolationMode mode =InterpolationMode::NEAREST, CoordinateTransformationMode transf_mode =CoordinateTransformationMode::HALF_PIXEL, NearestMode nearest_mode =NearestMode::ROUND_PREFER_FLOOR, int32 exclude_outside =0, float32 extrapolation_value =0.0f, float32 cubic_coeff_a =-0.75f, float32 a_scale =0.0f, int32 a_zero_point =0, float32 b_scale =0.0f, int32 b_zero_point =0) Onnx Resize-11 supported. |
| std::string | to_string(const ConfigurationValue & value) |
| std::ostream & | operator<<(std::ostream & s, const ConfigurationValue & value) |
Types Documentation¶
enum @909¶
| Enumerator | Value | Description |
|---|---|---|
| INVALID_PARAM_INDEX | 0xffffffff |
enum OnnxStatus¶
| Enumerator | Value | Description |
|---|---|---|
| ONNX_ERROR_INVALID_PARAMETERS | VX_ERROR_INVALID_PARAMETERS | |
| ONNX_FAILURE | VX_FAILURE | |
| ONNX_SUCCESS | VX_SUCCESS |
enum TensorType¶
| Enumerator | Value | Description |
|---|---|---|
| INVALID | 0x0000 | |
| INT8 | 0x0001 | |
| UINT8 | 0x0002 | |
| INT16 | 0x0003 | |
| UINT16 | 0x0004 | |
| INT32 | 0x0005 | |
| UINT32 | 0x0006 | |
| INT64 | 0x0007 | |
| UINT64 | 0x0008 | |
| FLOAT8 | 0x0009 | |
| FLOAT16 | 0x000A | |
| BFLOAT16 | 0x000B | |
| FLOAT32 | 0x000C | |
| FLOAT64 | 0x000D |
enum ShiftDirection¶
| Enumerator | Value | Description |
|---|---|---|
| LEFT | ||
| RIGHT |
enum InterpolationMode¶
| Enumerator | Value | Description |
|---|---|---|
| NEAREST | 0 | |
| BILINEAR | 1 | |
| CUBIC | 2 |
enum CoordinateTransformationMode¶
| Enumerator | Value | Description |
|---|---|---|
| HALF_PIXEL | 0 | |
| PYTORCH_HALF_PIXEL | 1 | |
| ALIGN_CORNERS | 2 | |
| ASYMMETRIC | 3 | |
| TF_HALF_PIXEL_FOR_NN | 4 | |
| TF_CROP_AND_RESIZE | 5 |
enum NearestMode¶
| Enumerator | Value | Description |
|---|---|---|
| ROUND_PREFER_FLOOR | 0 | |
| ROUND_PREFER_CEIL | 1 | |
| FLOOR | 2 | |
| CEIL | 3 |
enum DepthToSpaceMode¶
| Enumerator | Value | Description |
|---|---|---|
| DCR | 0 | |
| CRD | 1 |
enum RoiAlignMode¶
| Enumerator | Value | Description |
|---|---|---|
| MAX | ||
| AVG |
using int8¶
using int8 = int8_t;
using uint8¶
using uint8 = uint8_t;
using int16¶
using int16 = int16_t;
using uint16¶
using uint16 = uint16_t;
using int32¶
using int32 = int32_t;
using uint32¶
using uint32 = uint32_t;
using int64¶
using int64 = int64_t;
using uint64¶
using uint64 = uint64_t;
using float32¶
using float32 = float;
using float64¶
using float64 = double;
using Node¶
using Node = vx_node;
using Graph¶
using Graph = vx_graph;
using Kernel¶
using Kernel = vx_kernel;
using Parameter¶
using Parameter = vx_parameter;
using Context¶
using Context = vx_context;
using Tensor¶
using Tensor = vx_tensor;
using Scalar¶
using Scalar = vx_scalar;
using Image¶
using Image = vx_image;
using Status¶
using Status = vx_status;
using Reference¶
using Reference = vx_reference;
using TensorView¶
using TensorView = vx_tensor_view;
using ConfigurationMapping¶
using ConfigurationMapping = std::map<std::string, ConfigurationValue>;
Functions Documentation¶
function CreateContext¶
Context CreateContext()
Creates a context.
Return: The reference to the implementation context.
Note: This is required to do anything else.
Postcondition: ReleaseContext
This creates a top-level object context.
function SetConfiguration¶
Status SetConfiguration(
Context c,
std::string name,
ConfigurationValue value
)
Change configuration item.
Parameters:
- c The context to set the configuration item on.
- name Name of configuration item.
- value New value of configuration item.
Return: VX_SUCCESS if no error occurred. VX_ERROR_INVALID_PARAMETERS if the value is an invalid type or value.
Modify configuration item on context c. Check the vx error log for more information on why a failure occurred.
function GetConfiguration¶
ConfigurationMapping GetConfiguration(
Context c
)
Get current configuration state.
Parameters:
- c The context to get the configuration from.
Return: Mapping of configuration item to (type, value) where type is a ConfigurationValue.
function CreateGraph¶
Graph CreateGraph(
Context c
)
function ProcessGraph¶
Status ProcessGraph(
Graph g
)
function ReleaseGraph¶
Status ReleaseGraph(
Graph * g
)
function ReleaseContext¶
Status ReleaseContext(
Context * c
)
function GetStatus¶
Status GetStatus(
Reference obj
)
function AddParameterToGraph¶
Status AddParameterToGraph(
Graph g,
Parameter p
)
function SetGraphParameterByIndex¶
Status SetGraphParameterByIndex(
Graph g,
uint32 index,
Reference obj
)
function GetParameterByIndex¶
Parameter GetParameterByIndex(
Node n,
uint32 index
)
function ReleaseParameter¶
Status ReleaseParameter(
Parameter * p
)
function ReleaseNode¶
Status ReleaseNode(
Node * n
)
function ReleaseTensor¶
Status ReleaseTensor(
Tensor * t
)
function SetReferenceName¶
Status SetReferenceName(
Reference obj,
const char * name
)
function CreateTensorFromRawFile¶
Tensor CreateTensorFromRawFile(
Context c,
const char * filename,
int32 num_of_dims,
uint32x4 sizes,
TensorType data_format
)
function CreateTensor¶
Tensor CreateTensor(
Context c,
int32 num_of_dims,
uint32x4 sizes,
TensorType data_format
)
function CreateVirtualTensor¶
Tensor CreateVirtualTensor(
Graph g,
int32 num_of_dims,
uint32x4 sizes,
TensorType data_format
)
function CreateWeightsTensorFromRawFile¶
Tensor CreateWeightsTensorFromRawFile(
Context c,
const char * filename,
uint32x4 sizes,
TensorType data_format
)
function SaveTensorToRawFile¶
Status SaveTensorToRawFile(
Tensor t,
const char * filename
)
function CreateTensorView¶
TensorView CreateTensorView(
Context c,
uint32x4 start,
uint32x4 end,
int32 num_of_dims
)
function CreateTensorFromView¶
Tensor CreateTensorFromView(
Tensor t,
TensorView view
)
function AbsNode¶
Node AbsNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
AbsNode operator performs element-wise Absolute operation.
Parameters:
function AddNode¶
Node AddNode(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Parameters:
function AveragePoolNode¶
Node AveragePoolNode(
Graph graph,
Tensor a,
Tensor b,
uint8 ceil_mode =0,
uint8x4 kernel_shape ={1, 1, 0, 0},
uint8x4 pads ={0, 0, 0, 0},
uint8x4 strides ={1, 1, 1, 1},
uint8 count_include_pad =0,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
AveragePool operator performs element-wise average pooling operation on subset of tensor.
Parameters:
function BatchNormalizationNode¶
Node BatchNormalizationNode(
Graph graph,
Tensor a,
Tensor b,
Tensor weight,
Tensor bias,
Tensor weight_scale,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 bias_scale =0.0f,
int32 bias_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
Tensor var =nullptr,
Tensor mean =nullptr
)
BatchNorm operator performs element-wise max operation on subset of tensor.
Parameters:
- weight => input_var weight_scale => scale
function BitShiftNode¶
Node BitShiftNode(
Graph g,
Tensor a,
Tensor b,
Tensor c,
ShiftDirection dir,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Bitwise shift operator performs element-wise operation.
Parameters:
function EluNode¶
Node EluNode(
Graph graph,
Tensor a,
Tensor b,
float32 alpha =1.0f,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Elu operator performs element-wise function on input tensor.
Parameters:
function GemmNode¶
Node GemmNode(
Graph graph,
Tensor x,
Tensor y,
Tensor w,
Tensor b =nullptr,
float32 alpha =1.0f,
float32 beta =1.0f,
int32 transX =0,
int32 transY =0,
float32 x_scale =0.0f,
int32 x_zero_point =0,
float32 y_scale =0.0f,
int32 y_zero_point =0,
float32 w_scale =0.0f,
int32 w_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
uint8 shift_flag =1,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Parameters:
function GlobalAveragePoolNode¶
Node GlobalAveragePoolNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
GlobalAveragePool operator performs global average pooling operation on subset of tensor.
Parameters:
function LeakyReluNode¶
Node LeakyReluNode(
Graph graph,
Tensor a,
Tensor b,
float32 alpha =1.0f,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 alpha_scale =0.0f,
int32 alpha_zero_point =0
)
Parameters:
function LpNormalizationNode¶
Node LpNormalizationNode(
Graph g,
Tensor a,
Tensor b,
int32 p =2,
int32 axis =-1,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
Tensor weight =nullptr,
float32 weight_scale =0.0f,
int32 weight_zero_point =0
)
Parameters:
function LRNNode¶
Node LRNNode(
Graph g,
Tensor a,
Tensor b,
uint32 size,
float32 alpha =0.0001,
float32 beta =0.75,
float32 bias =1.0,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Parameters:
function MatMulNode¶
Node MatMulNode(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0
)
Parameters:
function MaxPoolNode¶
Node MaxPoolNode(
Graph graph,
Tensor a,
Tensor b,
uint8 ceil_mode =0,
uint8x4 dilations ={1, 1, 1, 1},
uint8x4 kernel_shape ={1, 1, 0, 0},
uint8x4 pads ={0, 0, 0, 0},
uint8x4 strides ={1, 1, 1, 1},
uint8 storage_order =0,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
MaxPool operator performs element-wise max operation on subset of tensor.
Parameters:
function MulNode¶
Node MulNode(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Parameters:
function QLinearConvNode¶
Node QLinearConvNode(
Graph graph,
Tensor x,
Tensor y,
Tensor w,
Tensor b =nullptr,
uint8x4 dilations ={1, 1, 1, 1},
uint32 group =1,
uint8x4 kernel_shape ={0, 0, 0, 0},
uint8x4 pads ={0, 0, 0, 0},
uint8x4 strides ={1, 1, 1},
float32 x_scale =0.0f,
int32 x_zero_point =0,
float32 y_scale =0.0f,
int32 y_zero_point =0,
float32 w_scale =0.0f,
int32 w_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
uint8 shift_flag =1
)
Parameters:
function QLinearConvNode2¶
Node QLinearConvNode2(
Graph graph,
Tensor x,
Tensor y,
Tensor w,
Tensor b =nullptr,
uint8x4 dilations ={1, 1, 1, 1},
uint32 group =1,
uint8x4 kernel_shape ={0, 0, 0, 0},
uint8x4 pads ={0, 0, 0, 0},
uint8x4 strides ={1, 1, 1},
float32 x_scale =0.0f,
int32 x_zero_point =0,
Tensor y_scale =nullptr,
int32 y_zero_point =0,
float32 w_scale =0.0f,
int32 w_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
uint8 shift_flag =1
)
function ConvTransposeNode¶
Node ConvTransposeNode(
Graph graph,
Tensor x,
Tensor y,
Tensor w,
Tensor b =nullptr,
uint8x4 dilations ={1, 1, 1, 1},
uint32 group =1,
uint8x4 kernel_shape ={0, 0, 0, 0},
uint8x4 output_padding ={0, 0, 0, 0},
uint8x4 pads ={0, 0, 0, 0},
uint8x4 strides ={1, 1, 1},
float32 x_scale =0.0f,
int32 x_zero_point =0,
float32 y_scale =0.0f,
int32 y_zero_point =0,
float32 w_scale =0.0f,
int32 w_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
uint8 shift_flag =1
)
Parameters:
function ReluNode¶
Node ReluNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Parameters:
function ResizeNode¶
Node ResizeNode(
Graph graph,
Tensor a,
Tensor b,
float32x4 scales,
Tensor roi,
InterpolationMode mode =InterpolationMode::NEAREST,
CoordinateTransformationMode transf_mode =CoordinateTransformationMode::HALF_PIXEL,
NearestMode nearest_mode =NearestMode::ROUND_PREFER_FLOOR,
int32 exclude_outside =0,
float32 extrapolation_value =0.0f,
float32 cubic_coeff_a =-0.75f,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Onnx Resize-11 supported.
Parameters:
function ReshapeNode¶
Node ReshapeNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Parameters:
function SoftmaxNode¶
Node SoftmaxNode(
Graph graph,
Tensor a,
Tensor b,
int32 axis =-1,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Parameters:
function SpaceToDepthNode¶
Node SpaceToDepthNode(
Graph graph,
Tensor a,
Tensor b,
int32 blocksize,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Parameters:
function TransposeNode¶
Node TransposeNode(
Graph graph,
Tensor input,
Tensor out,
uint8x4 perm ={0, 1, 2, 3},
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0
)
Parameters:
function SeluNode¶
Node SeluNode(
Graph graph,
Tensor a,
Tensor b,
float32 alpha =1.67326f,
float32 gamma =1.0507f,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Parameters:
function HardSigmoidNode¶
Node HardSigmoidNode(
Graph graph,
Tensor a,
Tensor b,
float32 alpha =0.2f,
float32 beta =0.5f,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Parameters:
function AndNode¶
Node AndNode(
Graph graph,
Tensor in0,
Tensor in1,
Tensor out,
float32 in0_scale =0.0f,
int32 in0_zero_point =0,
float32 in1_scale =0.0f,
int32 in1_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
AndNode operator performs element-wise Absolute operation.
Parameters:
function SubNode¶
Node SubNode(
Graph graph,
Tensor in0,
Tensor in1,
Tensor out,
float32 in0_scale =0.0f,
int32 in0_zero_point =0,
float32 in1_scale =0.0f,
int32 in1_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
SubNode operator performs element-wise Absolute operation.
Parameters:
function Sum3Node¶
Node Sum3Node(
Graph graph,
Tensor in0,
Tensor in1,
Tensor in2,
Tensor out,
float32 in0_scale =0.0f,
int32 in0_zero_point =0,
float32 in1_scale =0.0f,
int32 in1_zero_point =0,
float32 in2_scale =0.0f,
int32 in2_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
SumNode(3-input) operator performs element-wise Absolute operation.
Parameters:
function Sum4Node¶
Node Sum4Node(
Graph graph,
Tensor in0,
Tensor in1,
Tensor in2,
Tensor in3,
Tensor out,
float32 in0_scale =0.0f,
int32 in0_zero_point =0,
float32 in1_scale =0.0f,
int32 in1_zero_point =0,
float32 in2_scale =0.0f,
int32 in2_zero_point =0,
float32 in3_scale =0.0f,
int32 in3_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
SumNode(3-input) operator performs element-wise Absolute operation.
Parameters:
function TanhNode¶
Node TanhNode(
Graph graph,
Tensor input,
Tensor out,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0
)
TanhNode operator performs element-wise Absolute operation.
Parameters:
function DepthToSpaceNode¶
Node DepthToSpaceNode(
Graph graph,
Tensor a,
Tensor b,
int32 blocksize,
DepthToSpaceMode mode =DepthToSpaceMode::DCR,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
DepthToSpace operator performs bitwise DepthToSpace conversion.
Parameters:
function SliceNode¶
Node SliceNode(
Graph graph,
Tensor input,
Tensor output,
int32x4 starts,
int32x4 ends,
int32x4 axes,
int32x4 steps ={1, 1, 1, 1},
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 output_scale =0.0f,
int32 output_zero_point =0
)
Slice operator performs bitwise slice operation.
Parameters:
function Split2Node¶
Node Split2Node(
Graph graph,
Tensor input,
Tensor output0,
Tensor output1,
int32x4 split ={0, 0, 0, 0},
int32 axis =0,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 output0_scale =0.0f,
int32 output0_zero_point =0,
float32 output1_scale =0.0f,
int32 output1_zero_point =0
)
Split operator performs bitwise split operation.
Parameters:
function Split3Node¶
Node Split3Node(
Graph graph,
Tensor input,
Tensor output0,
Tensor output1,
Tensor output2,
int32x4 split ={0, 0, 0, 0},
int32 axis =0,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 output0_scale =0.0f,
int32 output0_zero_point =0,
float32 output1_scale =0.0f,
int32 output1_zero_point =0,
float32 output2_scale =0.0f,
int32 output2_zero_point =0
)
Split operator performs bitwise split operation.
Parameters:
function Split4Node¶
Node Split4Node(
Graph graph,
Tensor input,
Tensor output0,
Tensor output1,
Tensor output2,
Tensor output3,
int32x4 split ={0, 0, 0, 0},
int32 axis =0,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 output0_scale =0.0f,
int32 output0_zero_point =0,
float32 output1_scale =0.0f,
int32 output1_zero_point =0,
float32 output2_scale =0.0f,
int32 output2_zero_point =0,
float32 output3_scale =0.0f,
int32 output3_zero_point =0
)
Split operator performs bitwise split operation.
Parameters:
function ReduceMaxNode¶
Node ReduceMaxNode(
Graph graph,
Tensor a,
Tensor b,
int8x4 axes,
int32 keepdim =1,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
ReduceMax onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axes axes along which maximum values are claculated.
- keepdim The resulting tensor has the same rank as the input if keepdims equal 1. If keepdims equal 0, then the resulting tensor has the reduced dimension pruned.
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Computes max of input tensor's elements along provided axes.
function ReduceMinNode¶
Node ReduceMinNode(
Graph graph,
Tensor a,
Tensor b,
int8x4 axes,
int32 keepdim =1,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
ReduceMin onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axes axes along which minimum values are claculated.
- keepdim The resulting tensor has the same rank as the input if keepdim equals 1. If keepdim equals 0, then the resulting tensor has the reduced dimension pruned.
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Computes min of input tensor's elements along provided axes.
function ReduceMeanNode¶
Node ReduceMeanNode(
Graph graph,
Tensor a,
Tensor b,
int8x4 axes,
int32 keepdim =1,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
ReduceMean onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axes axes along which mean values are claculated.
- keepdim The resulting tensor has the same rank as the input if keepdim equals 1. If keepdim equals 0, then the resulting tensor has the reduced dimension pruned.
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Computes mean of input tensor's elements along provided axes.
function ReduceSumNode¶
Node ReduceSumNode(
Graph graph,
Tensor a,
Tensor b,
int8x4 axes,
int32 keepdim =1,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
ReduceSum onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axes axes along which maximum values are claculated.
- keepdim The resulting tensor has the same rank as the input if keepdim equals 1. If keepdim equals 0, then the resulting tensor has the reduced dimension pruned.
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Computes sum of input tensor's elements along provided axes.
function ClipNode¶
Node ClipNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 min =1.17549435e-38,
float32 max =3.38953139e+38
)
Clip onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- min minimum value to be clipped
- max maximum value to be clipped
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
performs element-wise Type Clip operation.
function Max2Node¶
Node Max2Node(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Max2 onnx Operator.
Parameters:
- a Input Tensor
- b Input Tensor
- c Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Input Tensor's scale
- b_zero_point Input Tensor's zero point value
- c_scale Output Tensor's scale
- c_zero_point Output Tensor's zero point value
performs element-wise Type Max operation with 2 inputs.
function Max3Node¶
Node Max3Node(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
Tensor d,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
float32 d_scale =0.0f,
int32 d_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Max3 onnx Operator.
Parameters:
- a Input Tensor
- b Input Tensor
- c Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Input Tensor's scale
- b_zero_point Input Tensor's zero point value
- c_scale Output Tensor's scale
- c_zero_point Output Tensor's zero point value
performs element-wise Type Max operation with 3 inputs.
function Max4Node¶
Node Max4Node(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
Tensor d,
Tensor e,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
float32 d_scale =0.0f,
int32 d_zero_point =0,
float32 e_scale =0.0f,
int32 e_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Max4 onnx Operator.
Parameters:
- a Input Tensor
- b Input Tensor
- c Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Input Tensor's scale
- b_zero_point Input Tensor's zero point value
- c_scale Output Tensor's scale
- c_zero_point Output Tensor's zero point value
performs element-wise Type Max operation with 4 inputs.
function Mean2Node¶
Node Mean2Node(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Mean2 onnx Operator.
Parameters:
- a Input Tensor
- b Input Tensor
- c Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Input Tensor's scale
- b_zero_point Input Tensor's zero point value
- c_scale Output Tensor's scale
- c_zero_point Output Tensor's zero point value
performs element-wise Type Mean operation with 2 inputs.
function Mean3Node¶
Node Mean3Node(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
Tensor d,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
float32 d_scale =0.0f,
int32 d_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Mean3 onnx Operator.
Parameters:
- a Input Tensor
- b Input Tensor
- c Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Input Tensor's scale
- b_zero_point Input Tensor's zero point value
- c_scale Output Tensor's scale
- c_zero_point Output Tensor's zero point value
performs element-wise Type Mean operation with 3 inputs.
function Mean4Node¶
Node Mean4Node(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
Tensor d,
Tensor e,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
float32 d_scale =0.0f,
int32 d_zero_point =0,
float32 e_scale =0.0f,
int32 e_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Mean4 onnx Operator.
Parameters:
- a Input Tensor
- b Input Tensor
- c Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Input Tensor's scale
- b_zero_point Input Tensor's zero point value
- c_scale Output Tensor's scale
- c_zero_point Output Tensor's zero point value
performs element-wise Type Mean operation with 4 inputs.
function LessNode¶
Node LessNode(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Parameters:
function SigmoidNode¶
Node SigmoidNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Parameters:
function XorNode¶
Node XorNode(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Parameters:
function ShrinkNode¶
Node ShrinkNode(
Graph graph,
Tensor a,
Tensor b,
float bias =0.0f,
float lambd =0.5,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Parameters:
function OrNode¶
Node OrNode(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Parameters:
function GreaterNode¶
Node GreaterNode(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Parameters:
function NotNode¶
Node NotNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Parameters:
function NegNode¶
Node NegNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Parameters:
function Min2Node¶
Node Min2Node(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Min2 onnx Operator.
Parameters:
- a Input Tensor
- b Input Tensor
- c Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Input Tensor's scale
- b_zero_point Input Tensor's zero point value
- c_scale Output Tensor's scale
- c_zero_point Output Tensor's zero point value
performs element-wise Type Min operation with 2 inputs.
function Min3Node¶
Node Min3Node(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
Tensor d,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
float32 d_scale =0.0f,
int32 d_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Min3 onnx Operator.
Parameters:
- a Input Tensor
- b Input Tensor
- c Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Input Tensor's scale
- b_zero_point Input Tensor's zero point value
- c_scale Output Tensor's scale
- c_zero_point Output Tensor's zero point value
performs element-wise Type Min operation with 3 inputs.
function Min4Node¶
Node Min4Node(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
Tensor d,
Tensor e,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
float32 d_scale =0.0f,
int32 d_zero_point =0,
float32 e_scale =0.0f,
int32 e_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Min4 onnx Operator.
Parameters:
- a Input Tensor
- b Input Tensor
- c Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Input Tensor's scale
- b_zero_point Input Tensor's zero point value
- c_scale Output Tensor's scale
- c_zero_point Output Tensor's zero point value
performs element-wise Type Min operation with 4 inputs.
function DivNode¶
Node DivNode(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Div onnx Operator.
Parameters:
- a Input Tensor
- b Input Tensor
- c Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Input Tensor's scale
- b_zero_point Input Tensor's zero point value
- c_scale Output Tensor's scale
- c_zero_point Output Tensor's zero point value
performs element-wise Type Div operation with 2 inputs.
function ThresholdedReluNode¶
Node ThresholdedReluNode(
Graph graph,
Tensor a,
Tensor b,
float32 alpha =1.0f,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
ThresholdedReluNode operator performs element-wise function on input tensor.
Parameters:
function PReluNode¶
Node PReluNode(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
PRelu onnx Operator.
Parameters:
- a Input Tensor
- b Input Tensor
- c Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Input Tensor's scale
- b_zero_point Input Tensor's zero point value
- c_scale Output Tensor's scale
- c_zero_point Output Tensor's zero point value
performs element-wise Type Div operation with 2 inputs.
function SqueezeNode¶
Node SqueezeNode(
Graph graph,
Tensor input,
Tensor out,
int8x4 axes ={4, 4, 4, 4},
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0
)
Squeeze Node
function UnsqueezeNode¶
Node UnsqueezeNode(
Graph graph,
Tensor input,
Tensor out,
int8x4 axes,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0
)
Unsqueeze Node
function GatherElementsNode¶
Node GatherElementsNode(
Graph graph,
Tensor input,
Tensor out,
Tensor index,
int32 axis =0,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0,
float32 index_scale =0.0f,
int32 index_zero_point =0
)
GatherElements Node
function ReduceProdNode¶
Node ReduceProdNode(
Graph graph,
Tensor a,
Tensor b,
int8x4 axes,
int32 keepdim =1,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
ReduceProd onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axes axes along which maximum values are claculated.
- keepdim The resulting tensor has the same rank as the input if keepdim equals 1. If keepdim equals 0, then the resulting tensor has the reduced dimension pruned.
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Computes produce of input tensor's elements along provided axes.
function ReduceSumSquareNode¶
Node ReduceSumSquareNode(
Graph graph,
Tensor a,
Tensor b,
int8x4 axes,
int32 keepdim =1,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
ReduceSumSquare onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axes axes along which maximum values are claculated.
- keepdim The resulting tensor has the same rank as the input if keepdim equal 1. If keepdim equals 0, then the resulting tensor has the reduced dimension pruned.
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Computes Sum of Squares of input tensor's elements along provided axes.
function SoftsignNode¶
Node SoftsignNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
SoftSign onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
SoftSign(x) = x/(1+|x|)
function EqualNode¶
Node EqualNode(
Graph graph,
Tensor in1,
Tensor in2,
Tensor out,
float32 in1_scale =0.0f,
int32 in1_zero_point =0,
float32 in2_scale =0.0f,
int32 in2_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Equal onnx Operator.
Parameters:
- in1 Input Tensor
- in2 Input Tensor
- out Output Tensor
- in1_scale Input Tensor's scale
- in1_zero_point Input Tensor's zero point value
- in2_scale Input Tensor's scale
- in2_zero_point Input Tensor's zero point value
- out_scale Output Tensor's scale
- out_zero_point Output Tensor's zero point value
performs element-wise equals operation with 2 inputs.
function ModNode¶
Node ModNode(
Graph graph,
Tensor in1,
Tensor in2,
Tensor out,
int8 fmod =0,
float32 in1_scale =0.0f,
int32 in1_zero_point =0,
float32 in2_scale =0.0f,
int32 in2_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Mod onnx Operator.
Parameters:
- in1 Input Tensor
- in2 Input Tensor
- out Output Tensor
- in1_scale Input Tensor's scale
- in1_zero_point Input Tensor's zero point value
- in2_scale Input Tensor's scale
- in2_zero_point Input Tensor's zero point value
- out_scale Output Tensor's scale
- out_zero_point Output Tensor's zero point value
performs element-wise modulus operation with 2 inputs.
function IdentityNode¶
Node IdentityNode(
Graph graph,
Tensor input,
Tensor output,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 output_scale =0.0f,
int32 output_zero_point =0
)
Identity onnx Operator.
Parameters:
- input Input Tensor
- output Output Tensor
performs element-wise data copying to output.
function ScatterElementsNode¶
Node ScatterElementsNode(
Graph graph,
Tensor data,
Tensor update,
Tensor out,
Tensor indices,
int axis =1,
float32 data_scale =0.0f,
int32 data_zero_point =0,
float32 update_scale =0.0f,
int32 update_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0,
float32 indices_scale =0.0f,
int32 indices_zero_point =0
)
ScatterElements onnx Operator.
Parameters:
- data Input Tensor
- indices Input Tensor
- update Input Tensor
- out Output Tensor
- in1_scale Input Tensor's scale
- in1_zero_point Input Tensor's zero point value
- indices_scale Input Tensor's scale
- indices_zero_point Input Tensor's zero point value
- update_scale Input Tensor's scale
- update_zero_point Input Tensor's zero point value
- out_scale Output Tensor's scale
- out_zero_point Output Tensor's zero point value
performs element-wise scatter operation.
function SqrtNode¶
Node SqrtNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Sqrt operator performs element-wise sqrt operation.
Parameters:
- a Input Tensor
- b Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Square root takes one input data (Tensor) and produces one output data (Tensor) where the square root is, y = x^0.5, is applied to the tensor elementwise.
function ReciprocalNode¶
Node ReciprocalNode(
Graph graph,
Tensor input,
Tensor out,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0
)
Reciprocal Node
function SinNode¶
Node SinNode(
Graph graph,
Tensor input,
Tensor out,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0
)
Sin Node
function CosNode¶
Node CosNode(
Graph graph,
Tensor input,
Tensor out,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0
)
Cos Node
function Sum2Node¶
Node Sum2Node(
Graph graph,
Tensor in0,
Tensor in1,
Tensor out,
float32 in0_scale =0.0f,
int32 in0_zero_point =0,
float32 in1_scale =0.0f,
int32 in1_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
SumNode(2-input) operator performs element-wise Absolute operation.
Parameters:
function HardmaxNode¶
Node HardmaxNode(
Graph graph,
Tensor a,
Tensor b,
int32 axis =1,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Hardmax onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axis indicating conversion type to 2d representation
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
performs row wise max operation in 2d representation of given input tensor
function ArgMaxNode¶
Node ArgMaxNode(
Graph graph,
Tensor a,
Tensor b,
int32 axis =0,
int32 keepdims =1,
int32 select_last_index =0,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Argmax onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axis indicating on which axis Argmax operation to be performed
- select_last_index indicates which index to be taken
- keepdims indicates to clip axis in the output
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
performs axis wise Argmax operation
function ArgMinNode¶
Node ArgMinNode(
Graph graph,
Tensor a,
Tensor b,
int32 axis =0,
int32 keepdims =1,
int32 select_last_index =0,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Argmin onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axis indicating on which axis Argmin operation to be performed
- select_last_index indicates which index to be taken
- keepdims indicates to clip axis in the output
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
performs axis wise Argmin operation
function CoshNode¶
Node CoshNode(
Graph graph,
Tensor input,
Tensor out,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0
)
CoshNode operator performs element-wise Absolute operation.
Parameters:
function AcoshNode¶
Node AcoshNode(
Graph graph,
Tensor input,
Tensor out,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0
)
ACoshNode operator performs element-wise Absolute operation.
Parameters:
function AcosNode¶
Node AcosNode(
Graph graph,
Tensor input,
Tensor out,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0
)
AcosNode operator performs element-wise Absolute operation.
Parameters:
function AtanhNode¶
Node AtanhNode(
Graph graph,
Tensor input,
Tensor out,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0
)
AtanhNode operator performs element-wise Absolute operation.
Parameters:
function AtanNode¶
Node AtanNode(
Graph graph,
Tensor input,
Tensor out,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0
)
AtanNode operator performs element-wise Absolute operation.
Parameters:
function AsinNode¶
Node AsinNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Asin onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Calculates the arcsine (inverse of sine) of the given input tensor, element-wise.
function AsinhNode¶
Node AsinhNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Asinh onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Calculates the hyperbolic arcsine of the given input tensor element-wise.
function SinhNode¶
Node SinhNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Sinh onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axis indicating conversion type to 2d representation
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Calculates the hyperbolic sine of the given input tensor element-wise
function ExpNode¶
Node ExpNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Exp onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axis indicating conversion type to 2d representation
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Calculates the exponential of the given input tensor, element-wise.
function TanNode¶
Node TanNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Tan onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axis indicating conversion type to 2d representation
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Calculates the tangent of the given input tensor element-wise.
function PowNode¶
Node PowNode(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 c_scale =0.0f,
int32 c_zero_point =0,
uint8x4 broadcast_axis ={0, 0, 0, 0}
)
Pow onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axis indicating conversion type to 2d representation
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Calculates element wise exponent of input1^input2 tensors
function ExpandNode¶
Node ExpandNode(
Graph graph,
Tensor input,
Tensor out,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0
)
Expand onnx Operator.
Parameters:
- in Input Tensor
- out Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
performs element wise expand operation
function SoftplusNode¶
Node SoftplusNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
SoftplusNode operator performs element-wise Softplus operation.
Parameters:
- a Input Tensor
- b Output Tensor
function ReduceLogSumNode¶
Node ReduceLogSumNode(
Graph graph,
Tensor a,
Tensor b,
int8x4 axes,
int32 keepdim =1,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
ReduceLogSum onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axes axes along which maximum values are claculated.
- keepdim The resulting tensor has the same rank as the input if keepdims equal 1. If keepdims equal 0, then the resulting tensor has the reduced dimension pruned.
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Computes log of sum of input tensor's elements along provided axes.
function ReduceLogSumExpNode¶
Node ReduceLogSumExpNode(
Graph graph,
Tensor a,
Tensor b,
int8x4 axes,
int32 keepdim =1,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
ReduceLogSumExp onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axes axes along which maximum values are claculated.
- keepdim The resulting tensor has the same rank as the input if keepdims equal 1. If keepdims equal 0, then the resulting tensor has the reduced dimension pruned.
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Computes log of sum of exponent of input tensor's elements along provided axes.
function ReduceL1Node¶
Node ReduceL1Node(
Graph graph,
Tensor a,
Tensor b,
int8x4 axes,
int32 keepdim =1,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
ReduceL1 onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axes axes along which maximum values are claculated.
- keepdim The resulting tensor has the same rank as the input if keepdims equal 1. If keepdims equal 0, then the resulting tensor has the reduced dimension pruned.
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Computes L1 norm of input tensor's elements along provided axes.
function ReduceL2Node¶
Node ReduceL2Node(
Graph graph,
Tensor a,
Tensor b,
int8x4 axes,
int32 keepdim =1,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
ReduceL2 onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axes axes along which maximum values are claculated.
- keepdim The resulting tensor has the same rank as the input if keepdims equal 1. If keepdims equal 0, then the resulting tensor has the reduced dimension pruned.
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Computes L2 norm of input tensor's elements along provided axes.
function LogNode¶
Node LogNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
LogNode onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Calculates the natural log of the given input tensor, element-wise.
function GlobalMaxPoolNode¶
Node GlobalMaxPoolNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
GlobalMaxPool onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
fids the max element in the spatial dimension of the given input tensor
function LpPoolNode¶
Node LpPoolNode(
Graph graph,
Tensor a,
Tensor b,
uint8x4 kernel_shape ={1, 1},
int p =2,
uint8x4 pads ={0, 0},
uint8x4 strides ={1, 1},
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
LpPool onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- kernel_shape The size of the kernel along each axis.
- p p value of the Lp norm used to pool over the input data.
- pads Padding for the beginning and ending along each spatial axis
- strides Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
LpPool consumes an input tensor X and applies Lp pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. Lp pooling consisting of computing the Lp norm on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing.
function GlobalLpPoolNode¶
Node GlobalLpPoolNode(
Graph graph,
Tensor a,
Tensor b,
int p =2,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
GlobalLpPool onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- p p value of the Lp norm used to pool over the input data.
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
GlobalLpPool consumes an input tensor X and applies lp pool pooling across the values in the same channel. This is equivalent to LpPool with kernel size equal to the spatial dimension of input tensor.
function GatherNode¶
Node GatherNode(
Graph graph,
Tensor input,
Tensor out,
Tensor index,
int32 axis =0,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0,
float32 index_scale =0.0f,
int32 index_zero_point =0
)
Gather details param
function LogSoftmaxNode¶
Node LogSoftmaxNode(
Graph graph,
Tensor a,
Tensor b,
int32 axis =-1,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
LogSoftmax onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axis scalar
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Calculates log to the softmax to given input tensor, element-wise.
function MaxUnpoolNode¶
Node MaxUnpoolNode(
Graph graph,
Tensor a,
Tensor b,
Tensor c,
uint8x4 kernel_shape ={1, 1},
uint8x4 pads ={0, 0, 0, 0},
uint8x4 strides ={1, 1},
float32 a_scale =0,
int32 a_zero_point =0,
float32 c_scale =0,
int32 c_zero_point =0
)
MaxUnpool onnx Operator.
Parameters:
- a Input data tensor that has to be unpooled.
- b Input data tensor containing the indices corresponding to elements in the first input tensor.
- c Output data tensor that contains the result of the unpooling.
- kernel_shape The size of the kernel along each axis.
- pads Padding for the beginning and ending along each spatial axis
- strides Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
- c_scale Output Tensor's scale
- c_zero_point Output Tensor's zero point value
MaxUnpool essentially computes the partial inverse of the MaxPool op.
function InstanceNormalizationNode¶
Node InstanceNormalizationNode(
Graph graph,
Tensor input,
Tensor out,
Tensor s,
Tensor b,
float epsilon =0.00001f,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 out_scale =0.0f,
int32 out_zero_point =0,
float32 s_scale =0.0f,
int32 s_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
InstanceNormalization onnx Operator.
Parameters:
- in Input Tensor
- out Output Tensor
- s 1-D Tensor
- b 1-D Tensor
- epsilon avoids divide by zero condition
- in_scale Output Tensor's scale
- in_zero_point Input Tensor's zero point value
- out_scale Output Tensor's scale
- out_zero_point Output Tensor's zero point value
- s_scale s Tensor's scale
- s_zero_point s Tensor's zero point value
- b_scale b Tensor's scale
- b_zero_point b Tensor's zero point value
performs axis wise Argmin operation
function CumSumNode¶
Node CumSumNode(
Graph graph,
Tensor a,
Tensor b,
int32 axis =0,
int8 exclusive =0,
int8 reverse =0,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
CumSum onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- axis Axis along which CumSum operation is performed
- exclusive If set to 1 will return exclusive sum in which the top element is not included.
- reverse If set to 1 will perform the sums in reverse direction.
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Performs cumulative sum of the input elements along the given axis.
function OneHotNode¶
Node OneHotNode(
Graph graph,
Tensor a,
Tensor b,
int32 axis =-1,
int32 depth =0,
float32 off_value =0.0f,
float32 on_value =0.0f,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
function ErfNode¶
Node ErfNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
function GatherNDNode¶
Node GatherNDNode(
Graph graph,
Tensor a,
Tensor b,
Tensor indices,
int32 batch_dim =0,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0,
float32 indices_scale =0.0f,
int32 indices_zero_point =0
)
Gather onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- c Indices Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale
- b_zero_point Output Tensor's zero point value
Performs gather operation on input elements along the given indices
function ScatterNDNode¶
Node ScatterNDNode(
Graph graph,
Tensor input,
Tensor updates,
Tensor out,
Tensor indices,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 updates_scale =0.0f,
int updates_zero_point =0,
float32 out_scale =0.0f,
int out_zero_point =0,
float32 indices_scale =0.0f,
int indices_zero_point =0
)
function SignNode¶
Node SignNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
Sign onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Input Tensor's scale, always zero
- b_zero_point Input Tensor's zero point value, always zero
Calculate the sign of the given input tensor element-wise. If input
0, output 1. if input < 0, output -1. if input == 0, output 0.
function NonMaxSuppressionNode¶
Node NonMaxSuppressionNode(
Graph graph,
Tensor in_boxes,
Tensor scores,
Tensor out_selected_indices,
int32 center_point_box =0,
int32 max_output_boxes_per_class =0,
float iou_threshold =0,
float score_threshold =0,
float32 in_boxes_scale =0.0f,
int32 in_boxes_zero_point =0,
float32 scores_scale =0.0f,
int32 scores_zero_point =0,
float32 out_selected_indices_scale =0.0f,
int32 out_selected_indices_zero_point =0,
bool with_sort =true
)
NonMaxSuppression Operator.
Parameters:
function MaxRoiPoolNode¶
Node MaxRoiPoolNode(
Graph graph,
Tensor input,
Tensor rois,
Tensor output,
int32 pooled_shape_height,
int32 pooled_shape_width,
float32 spatial_scale_factor =1.0f,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 rois_scale =0.0f,
int32 rois_zero_point =0,
float32 output_scale =0.0f,
int32 output_zero_point =0
)
MaxRoiPoolNode.
Parameters:
function RoiAlignNode¶
Node RoiAlignNode(
Graph graph,
Tensor input,
Tensor rois,
Tensor output,
RoiAlignMode mode =RoiAlignMode::AVG,
int32 output_height =1,
int32 output_width =1,
int32 sampling_ratio =0,
float32 spatial_scale_factor =1.0,
float32 input_scale =0,
int32 input_zero_point =0,
float32 output_scale =0,
int32 output_zero_point =0,
float32 rois_scale =0,
int32 rois_zero_point =0
)
RoiAlignNode.
Parameters:
function CeilNode¶
Node CeilNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
Ceil onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Input Tensor's scale, always zero
- b_zero_point Input Tensor's zero point value, always zero
Calculate the ceil of the given input tensor element-wise.
function FloorNode¶
Node FloorNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
Floor onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Input Tensor's scale, always zero
- b_zero_point Input Tensor's zero point value, always zero
Calculate the floor of the given input tensor element-wise.
function RoundNode¶
Node RoundNode(
Graph graph,
Tensor a,
Tensor b,
float32 a_scale =0,
int32 a_zero_point =0,
float32 b_scale =0,
int32 b_zero_point =0
)
Round onnx Operator.
Parameters:
- a Input Tensor
- b Output Tensor
- a_scale Input Tensor's scale
- a_zero_point Input Tensor's zero point value
- b_scale Output Tensor's scale, always zero
- b_zero_point Output Tensor's zero point value, always zero
Calculate the round of the given input tensor element-wise.
function IsNaNNode¶
Node IsNaNNode(
Graph graph,
Tensor input,
Tensor output,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 output_scale =0.0f,
int32 output_zero_point =0
)
IsNan onnx Operator.
Parameters:
- input Input Tensor
- output Output Tensor
- input_scale Input Tensor's scale
- input_zero_point Input Tensor's zero point value
- output_scale Output Tensor's scale
- output_zero_point Output Tensor's zero point value
Returns which elements of the input are NaN.
function IsInfNode¶
Node IsInfNode(
Graph graph,
Tensor input,
Tensor output,
int32 detect_negative,
int32 detect_positive,
float32 input_scale =0.0f,
int32 input_zero_point =0,
float32 output_scale =0.0f,
int32 output_zero_point =0
)
IsInf onnx Operator.
Parameters:
- input Input Tensor
- output Output Tensor
Returns which elements of the input are Inf.
function Concat2Node¶
Node Concat2Node(
Graph graph,
Tensor input1,
Tensor input2,
Tensor output,
int axis,
float32 input1_scale =0.0f,
int32 input1_zero_point =0,
float32 input2_scale =0.0f,
int32 input2_zero_point =0,
float32 output_scale =0.0f,
int32 output_zero_point =0
)
Concate2 onnx Operator.
Parameters:
- input1 Input Tensor
- input2 Input Tensor
- output Output Tensor
- axis Axis along concate operation is performed
Concatenate two tensors into a single tensor.
function Concat3Node¶
Node Concat3Node(
Graph graph,
Tensor input1,
Tensor input2,
Tensor input3,
Tensor output,
int axis,
float32 input1_scale =0.0f,
int32 input1_zero_point =0,
float32 input2_scale =0.0f,
int32 input2_zero_point =0,
float32 input3_scale =0.0f,
int32 input3_zero_point =0,
float32 output_scale =0.0f,
int32 output_zero_point =0
)
Concate3 onnx Operator.
Parameters:
- input1 Input Tensor
- input2 Input Tensor
- input3 Input Tensor
- output Output Tensor
- axis Axis along concate operation is performed
Concatenate two tensors into a single tensor.
function Concat4Node¶
Node Concat4Node(
Graph graph,
Tensor input1,
Tensor input2,
Tensor input3,
Tensor input4,
Tensor output,
int32 axis,
float32 input1_scale =0.0f,
int32 input1_zero_point =0,
float32 input2_scale =0.0f,
int32 input2_zero_point =0,
float32 input3_scale =0.0f,
int32 input3_zero_point =0,
float32 input4_scale =0.0f,
int32 input4_zero_point =0,
float32 output_scale =0.0f,
int32 output_zero_point =0
)
Concate4 onnx Operator.
Parameters:
- input1 Input Tensor
- input2 Input Tensor
- input3 Input Tensor
- input4 Input Tensor
- output Output Tensor
- axis Axis along concate operation is performed
Concatenate two tensors into a single tensor.
function GetInputParamIndex¶
uint32_t GetInputParamIndex(
Node node,
int32 param_num
)
Get the index of the input parameter param_num for given node.
Parameters:
- node The node to query
- param_num One based index of the input parameter index, e.g. for the first input parameter pass 1.
Return: Zero based parameter index of the requested input parameter, or INVALID_PARAM_INDEX on error
Returns the zero based parameter index of the one based param_num input parameter This takes account of the number of output parameters and the ordering of input/output parameters in the given node.
The returned index can be passed directly to GetParameterByIndex.
function GetOutputParamIndex¶
uint32_t GetOutputParamIndex(
Node node
)
Get the index of the output parameter for given node.
Parameters:
- node The node to query
Return: Zero based parameter index of the output parameter, or INVALID_PARAM_INDEX on error
Returns the zero based parameter index of the output parameter for the given node. This takes account of the number of input parameters and the ordering of input/output parameters in the given node.
The returned index can be passed directly to GetParameterByIndex.
function CustomNodeFromKernel¶
Node CustomNodeFromKernel(
Graph graph,
Kernel kernel,
std::vector< Tensor > outputs,
std::vector< Tensor > inputs,
std::vector< Scalar > scalars ={},
std::vector< int32 > input_zero_points ={},
std::vector< float32 > input_scales ={},
std::vector< int32 > output_zero_points ={},
std::vector< float32 > output_scales ={},
std::string kernel_file_name ={}
)
Create a custom node from an OpenVX kernel.
Parameters:
- graph to add the node to.
- kernel vx_kernel to create a node around.
- outputs List of output tensors.
- inputs List of input tensors.
- scalars List of input scalars.
- input_zero_points List of input zeropoints, one per input tensor, missing items are defaulted to 0.
- input_scales List of input scales, one per input tensor, missing items are defaulted to 0.0.
- output_zero_points List of zeropoints, one per output tensor, missing items are defaulted to 0.
- output_scales List of output scales, one per output tensor, missing items are defaulted to 0.0.
- kernel_file_name Path to kernel source, not required.
This function can be used to add an existing vx_kernel to an onnx graph.
function ResizeNode¶
Node ResizeNode(
Graph graph,
Tensor a,
Tensor b,
uint32x4 scales,
InterpolationMode mode =InterpolationMode::NEAREST,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Onnx Resize-10 supported.
Parameters:
function ResizeNode¶
Node ResizeNode(
Graph graph,
Tensor a,
Tensor b,
float32x4 scales,
uint32x4 roi_start,
uint32x4 roi_end,
InterpolationMode mode =InterpolationMode::NEAREST,
CoordinateTransformationMode transf_mode =CoordinateTransformationMode::HALF_PIXEL,
NearestMode nearest_mode =NearestMode::ROUND_PREFER_FLOOR,
int32 exclude_outside =0,
float32 extrapolation_value =0.0f,
float32 cubic_coeff_a =-0.75f,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
function ResizeNode¶
Node ResizeNode(
Graph graph,
Tensor a,
Tensor b,
uint32x4 scales,
uint32x4 roi_start,
uint32x4 roi_end,
InterpolationMode mode =InterpolationMode::NEAREST,
CoordinateTransformationMode transf_mode =CoordinateTransformationMode::HALF_PIXEL,
NearestMode nearest_mode =NearestMode::ROUND_PREFER_FLOOR,
int32 exclude_outside =0,
float32 extrapolation_value =0.0f,
float32 cubic_coeff_a =-0.75f,
float32 a_scale =0.0f,
int32 a_zero_point =0,
float32 b_scale =0.0f,
int32 b_zero_point =0
)
Onnx Resize-11 supported.
Parameters:
function to_string¶
std::string to_string(
const ConfigurationValue & value
)
function operator<<¶
std::ostream & operator<<(
std::ostream & s,
const ConfigurationValue & value
)
Updated on 2023-11-29 at 18:22:22 +0000