BI BI MI MI MI MI BI MI No level Level 8K ARGMAX Input tensor Axis in range from 0 to rank(shape1) - 1 Output tensor, with rank = rank(shape1) - 1 AVG_POOL2D Input tensor [kernel_y, kernel_x] [stride_y, stride_x] [pad_top, pad_bottom, pad_left, pad_right] Enumerated type, must be one of INT32, FP16, FP32, as defined in the Supported Data Types table for this operation Input tensor zero point. Must be zero for non-int8 types. Output tensor zero point. Must be zero for non-int8 types. Output tensor 4D CONV2D Input tensor Weight kernel size KH x KW Per output channel bias data. + Bias data will be broadcast if BC == 1. [pad_top, pad_bottom, pad_left, pad_right] [stride_y, stride_x] [dilation_y, dilation_x] Enumerated type, must be one of INT32, INT48, FP16, FP32, as defined in the Supported Data Types table for this operation Input tensor zero point. Must be zero for non-int8 types. Weight zero point. Must be zero for non-int8 types. This optional attribute affects the floating-point compliance error bound. The default of false allows for direct and transform based, fast convolution algorithms. Only set to true if direct dot-product calculation precision is required. Output tensor CONV3D Input tensor Weight kernel size KDxKHxKW Per output channel bias data. + Bias data will be broadcast if BC == 1. [pad_d0, pad_d1, pad_top, pad_bottom, pad_left, pad_right] [stride_d, stride_y, stride_x] [dilation_d, dilation_y, dilation_x] Enumerated type, must be one of INT32, INT48, FP16, FP32, as defined in the Supported Data Types table for this operation Input tensor zero point. Must be zero for non-int8 types. Weight zero point. Must be zero for non-int8 types. This optional attribute affects the floating-point compliance error bound. The default of false allows for direct and transform based, fast convolution algorithms. Only set to true if direct dot-product calculation precision is required. Output tensor DEPTHWISE_CONV2D Input tensor Weight kernel size KH x KW Per output channel bias data. + Bias data will be broadcast if BC == 1. [pad_top, pad_bottom, pad_left, pad_right] [stride_y, stride_x] [dilation_y, dilation_x] Enumerated type, must be one of INT32, INT48, FP16, FP32, as defined in the Supported Data Types table for this operation Input tensor zero point. Must be zero for non-int8 types. Weight zero point. Must be zero for non-int8 types. This optional attribute affects the floating-point compliance error bound. The default of false allows for direct and transform based, fast convolution algorithms. Only set to true if direct dot-product calculation precision is required. Output tensor FFT2D Real part of the complex input. H,W must be powers of two. Imaginary part of the complex input. H,W must be powers of two. false for forward FFT, true for inverse FFT Real part of the complex output. This optional attribute affects the floating-point compliance error bound. The default of false allows for direct and transform based, fast convolution algorithms. Only set to true if direct dot-product calculation precision is required. Imaginary part of the complex output. FULLY_CONNECTED Input tensor Weights Per output channel bias data. + Bias data will be broadcast if BC == 1. Input tensor zero point. Must be zero for non-int8 types. Weight zero point. Must be zero for non-int8 types. Output tensor MATMUL Input tensor A, N matrices of size HxC Input tensor B, N matrices of size CxW Input tensor A zero point. Must be zero for non-int8 types. Input tensor B zero point. Must be zero for non-int8 types. Output tensor, N matrices of size HxW MAX_POOL2D Input tensor 4D [kernel_y, kernel_x] [stride_y, stride_x] [pad_top, pad_bottom, pad_left, pad_right] Output tensor 4D RFFT2D Real input. H,W must be powers of two. Real part of the complex output This optional attribute affects the floating-point compliance error bound. The default of false allows for direct and transform based, fast convolution algorithms. Only set to true if direct dot-product calculation precision is required. Imaginary part of the complex output. TRANSPOSE_CONV2D Input tensor Weight kernel size KH x KW Per output channel bias data. + Bias data will be broadcast if BC == 1. [out_pad_top, out_pad_bottom, out_pad_left, out_pad_right] [stride_y, stride_x] Enumerated type, must be one of INT32, INT48, FP16, FP32, as defined in the Supported Data Types table for this operation Input tensor zero point. Must be zero for non-int8 types. Weight zero point. Must be zero for non-int8 types. This optional attribute affects the floating-point compliance error bound. The default of false allows for direct and transform based, fast convolution algorithms. Only set to true if direct dot-product calculation precision is required. Output tensor CLAMP Input tensor Minimum clip value Maximum clip value Output tensor of same type and shape as input ERF Input tensor Output tensor of same type and shape as input SIGMOID Input tensor Output tensor of same type and shape as input TANH Input tensor Output tensor of same type and shape as input ADD Input tensor Input tensor with the same rank as input1 Output tensor ARITHMETIC_RIGHT_SHIFT Input tensor Input tensor with the same rank as input1 If true then the shift is rounded Output tensor BITWISE_AND Input tensor Input tensor with the same rank as input1 Output tensor BITWISE_OR Input tensor Input tensor with the same rank as input1 Output tensor BITWISE_XOR Input tensor Input tensor with the same rank as input1 Output tensor INTDIV Input tensor Input tensor with the same rank as input1 Output tensor LOGICAL_AND Input tensor Input tensor with the same rank as input1 Output tensor LOGICAL_LEFT_SHIFT Input tensor Input tensor with the same rank as input1 Output tensor LOGICAL_RIGHT_SHIFT Input tensor Input tensor with the same rank as input1 Output tensor LOGICAL_OR Input tensor Input tensor with the same rank as input1 Output tensor LOGICAL_XOR Input tensor Input tensor with the same rank as input1 Output tensor MAXIMUM Input tensor Input tensor with the same rank as input1 Output tensor MINIMUM Input tensor Input tensor with the same rank as input1 Output tensor MUL Input tensor Input tensor with the same rank as input1 Result right shift (i32_t data type only) Output tensor POW Input tensor Input tensor with the same rank as input1 Output tensor SUB Input tensor Input tensor with the same rank as input1 Output tensor TABLE Input tensor Lookup table tensor Output tensor ABS Input tensor Output tensor of same type, size as the input tensor BITWISE_NOT Input tensor Output tensor of same type, size as the input tensor CEIL Input tensor Output tensor of same type, size as the input tensor CLZ Input tensor Output tensor of same type, size as the input tensor COS Input tensor Output tensor of same type and shape as input EXP Input tensor Output tensor of same type, size as the input tensor FLOOR Input tensor Output tensor of same type, size as the input tensor LOG Input tensor Output tensor of same type, size as the input tensor LOGICAL_NOT Input tensor Output tensor of same type, size as the input tensor NEGATE Input tensor Input 1 zero point. Must be zero for non-int8 types. Output zero point. Must be zero for non-int8 types. Output tensor of same type, size as the input tensor RECIPROCAL Input tensor Output tensor of same type, size as the input tensor RSQRT Input tensor Output tensor of same type, size as the input tensor SELECT Input selector tensor Input value tensor if input1 is True Input value tensor if input1 is False Output tensor of same type as input2 and input3 SIN Input tensor Output tensor of same type and shape as input EQUAL Input tensor Input tensor with the same rank as input1 Output tensor GREATER Input tensor Input tensor with the same rank as input1 Output tensor GREATER_EQUAL Input tensor Input tensor with the same rank as input1 Output tensor REDUCE_ALL Input tensor Axis to reduce, in range from 0 to rank(shape1)-1 Output tensor. Same rank as the input tensor. REDUCE_ANY Input tensor Axis to reduce, in range from 0 to rank(shape1)-1 Output tensor. Same rank as the input tensor. REDUCE_MAX Input tensor Axis to reduce, in range from 0 to rank(shape1)-1 Output tensor. Same rank as the input tensor. REDUCE_MIN Input tensor Axis to reduce, in range from 0 to rank(shape1)-1 Output tensor. Same rank as the input tensor. REDUCE_PRODUCT Input tensor Axis to reduce, in range from 0 to rank(shape1)-1 Output tensor. Same rank as the input tensor. REDUCE_SUM Input tensor Axis to reduce, in range from 0 to rank(shape1)-1 Output tensor. Same rank as the input tensor. CONCAT List of input tensors. All inputs must have the same rank and data type Axis along which concatenation is to occur, in range from 0 to rank(shape)-1 Output tensor PAD Input tensor Number of pad elements at the start and end of each dimension. The values in padding are interpreted as start, end of each dimension. As an example for rank 2, the values would be interpreted as [start_dim0, end_dim0, start_dim1, end_dim1]. Constant value to be used as padding Output tensor of same type as the input tensor RESHAPE Input tensor shape_t giving the new shape. Output tensor of same type, size as the input tensor REVERSE Input tensor Axis to reverse, in range from 0 to rank(shape)-1 Output tensor. Same shape as input tensor SLICE Input tensor List of integer coordinates, of length equal to the rank of input1. Start coordinate for slicing. List of integer size values, of length equal to the rank of input1. Size of the input to be used. Output tensor of same type as the input tensor TILE Input tensor Number of times to replicate input1 in each dimension Output tensor of same type, rank as the input tensor TRANSPOSE Input tensor List of integers of length equal to the rank of input1. Values must be valid dimensions within shape1, and may not be repeated. Output tensor of same type, rank as the input tensor GATHER 3D value tensor 2D index tensor 3D output tensor SCATTER 3D values in tensor 2D index tensor 3D input tensor 3D output tensor RESIZE Input tensor [scale_y_n, scale_y_d, scale_x_n, scale_x_d] [offset_y, offset_x] [border_y, border_x] BILINEAR or NEAREST Output tensor CAST Input tensor Output tensor RESCALE Input tensor Output tensor with the same shape as input Input tensor zero point. int8/uint8 can have zero point within their valid range. uint16 zero point must be either 0 or 32768. All other types must have zero point equal to 0. Output tensor zero point.int8/uint8 can have zero point within their valid range. uint16 zero point must be either 0 or 32768. All other types must have zero point equal to 0. Scaling multiplier array Scaling shift array if (scale32) mul_t=i32_t else mul_t=i16_t Select double round mode if (per_channel) NC=shape[rank(shape)-1] else NC=1 If True, treat the input values as unsigned. If True, treat the output values as unsigned. CONST Constant values Output tensor IDENTITY Input tensor Output tensor of the same type, size as the input tensor CUSTOM List of input tensors String which tells the backend which custom operator is being called String idenifier which can help avoid name collisions on the operator field. Different implementations of a given operator would be in different domains. Implementations can choose which domains they want to support. String value containing implementation specific attributes which apply to the operation List of output tensors COND_IF Input condition as a size 1 tensor List of input tensors TOSA graph to execute if condition is true TOSA graph to execute if condition is false List of output tensors WHILE_LOOP List of input tensors TOSA graph to evaluate the condition TOSA graph to execute the loop body List of output tensors VARIABLE Globally unique identifier for the declared variable tensor. The variable tensor shape Type of the tensor variable elements. Initial value of the variable tensor. This argument is optional with default value NULL. VARIABLE_WRITE Globally unique identifier of the variable tensor that is writing to Input tensor VARIABLE_READ Globally unique identifier of the variable tensor that is reading from Output tensor ADD_SHAPE Input 1 Input 2 Output shape CONCAT_SHAPE List of input shape values Output shape CONST_SHAPE Constant shape Output shape DIM Input tensor Axis in range from 0 to rank(shape) - 1 Output shape type of size 1 giving the size of the shape for the given axis DIV_SHAPE Input 1 Input 2 Output shape MUL_SHAPE Input 1 Input 2 Output shape SUB_SHAPE Input 1 Input 2 Output shape