diff options
46 files changed, 2291 insertions, 63 deletions
diff --git a/arm_compute/runtime/CL/functions/CLActivationLayer.h b/arm_compute/runtime/CL/functions/CLActivationLayer.h index aa62183909..e6bb192532 100644 --- a/arm_compute/runtime/CL/functions/CLActivationLayer.h +++ b/arm_compute/runtime/CL/functions/CLActivationLayer.h @@ -59,6 +59,19 @@ public: CLActivationLayer &operator=(CLActivationLayer &&); /** Set the input and output tensor. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |QSYMM16 |QSYMM16 | + * |F16 |F16 | + * |F32 |F32 | + * + * * @note If the output tensor is a nullptr or is equal to the input, the activation function will be performed in-place * * @param[in, out] input Source tensor. In case of @p output tensor = nullptr, this tensor will store the result diff --git a/arm_compute/runtime/CL/functions/CLConcatenateLayer.h b/arm_compute/runtime/CL/functions/CLConcatenateLayer.h index bfc8a39ac9..71e84e21b5 100644 --- a/arm_compute/runtime/CL/functions/CLConcatenateLayer.h +++ b/arm_compute/runtime/CL/functions/CLConcatenateLayer.h @@ -64,11 +64,22 @@ public: CLConcatenateLayer &operator=(CLConcatenateLayer &&); /** Initialise the kernel's inputs vector and output. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |F16 |F16 | + * |F32 |F32 | + * * @note Input and output tensor dimensions preconditions defer depending on the concatenation axis. * @note Preconditions can be found respectively at @ref opencl::kernels::ClWidthConcatenateKernel, * @ref opencl::kernels::ClHeightConcatenateKernel and @ref opencl::kernels::ClDepthConcatenateKernel. * - * @param[in,out] inputs_vector The vectors containing all the tensors to concatenate. Data types supported: All + * @param[in,out] inputs_vector The vectors containing all the tensors to concatenate. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32. * @param[out] output Output tensor. Data types supported: Same as @p input. * @param[in] axis Concatenation axis. Supported underlying concatenation axis are 0, 1, 2 and 3. */ @@ -80,7 +91,7 @@ public: * @ref opencl::kernels::ClHeightConcatenateKernel and @ref opencl::kernels::ClDepthConcatenateKernel. * * @param[in] compile_context The compile context to be used. - * @param[in,out] inputs_vector The vectors containing all the tensors to concatenate. Data types supported: All + * @param[in,out] inputs_vector The vectors containing all the tensors to concatenate. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32. * @param[out] output Output tensor. Data types supported: Same as @p input. * @param[in] axis Concatenation axis. Supported underlying concatenation axis are 0, 1, 2 and 3. */ @@ -91,7 +102,7 @@ public: * @note Preconditions can be found respectively at @ref opencl::kernels::ClWidthConcatenateKernel, * @ref opencl::kernels::ClHeightConcatenateKernel and @ref opencl::kernels::ClDepthConcatenateKernel. * - * @param[in] inputs_vector The vectors containing all the tensors info to concatenate. Data types supported: All. + * @param[in] inputs_vector The vectors containing all the tensors info to concatenate. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32. * @param[in] output Output tensor info. Data types supported: Same as @p input. * @param[in] axis Concatenation axis. Supported underlying concatenation axis are 0, 1, 2 and 3. * diff --git a/arm_compute/runtime/CL/functions/CLConvertFullyConnectedWeights.h b/arm_compute/runtime/CL/functions/CLConvertFullyConnectedWeights.h index 8892dbad6c..6c7d9e52e8 100644 --- a/arm_compute/runtime/CL/functions/CLConvertFullyConnectedWeights.h +++ b/arm_compute/runtime/CL/functions/CLConvertFullyConnectedWeights.h @@ -55,6 +55,15 @@ public: CLConvertFullyConnectedWeights &operator=(CLConvertFullyConnectedWeights &&) = default; /** Initialize the function. * + * Valid data layouts: + * - NHWC + * - NCHW + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |All |All | + * * @param[in] input Source weights tensor to convert. Must be 2 dimensional. Data types supported: All. * @param[out] output The converted weights tensor. Shape and Data Type: Same as @p input. * @param[in] original_input_shape Shape of the original input tensor (the one entering fully connected layer). diff --git a/arm_compute/runtime/CL/functions/CLCopy.h b/arm_compute/runtime/CL/functions/CLCopy.h index 795a183e1f..4fc4183d3e 100644 --- a/arm_compute/runtime/CL/functions/CLCopy.h +++ b/arm_compute/runtime/CL/functions/CLCopy.h @@ -53,6 +53,14 @@ public: CLCopy &operator=(CLCopy &&); /** Initialise the function's source and destination. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |All |All | + * * @param[in] input Source tensor. Data types supported: All. * @param[out] output Output tensor. Data types supported: Same as @p input. * @param[in] dst_window (Optional) Window to be used in case only copying into part of a tensor. Default is nullptr. diff --git a/arm_compute/runtime/CL/functions/CLDequantizationLayer.h b/arm_compute/runtime/CL/functions/CLDequantizationLayer.h index 7ecf094df3..4a5c3a3203 100644 --- a/arm_compute/runtime/CL/functions/CLDequantizationLayer.h +++ b/arm_compute/runtime/CL/functions/CLDequantizationLayer.h @@ -54,6 +54,23 @@ public: CLDequantizationLayer &operator=(CLDequantizationLayer &&) = default; /** Set the input and output tensors. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------------------|:--------------| + * |QASYMM8 |F16 | + * |QASYMM8 |F32 | + * |QASYMM8_SIGNED |F16 | + * |QASYMM8_SIGNED |F32 | + * |QSYMM8_PER_CHANNEL |F16 | + * |QSYMM8_PER_CHANNEL |F32 | + * |QSYMM8 |F16 | + * |QSYMM8 |F32 | + * |QSYMM16 |F16 | + * |QSYMM16 |F32 | + * * @param[in] input Source tensor with at least 3 dimensions. The dimensions over the third will be interpreted as batches. * Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL/QSYMM8/QSYMM16. * @param[out] output Destination tensor with the same dimensions of input. Data type supported: F16/F32. diff --git a/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h index 6e9e2161b9..3fd0c63782 100644 --- a/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h +++ b/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h @@ -55,6 +55,18 @@ public: CLDirectConvolutionLayer &operator=(CLDirectConvolutionLayer &&); /** Set the input and output tensors. * + * Valid data layouts: + * - NHWC + * - NCHW + * + * Valid data type configurations: + * |src0 |src1 |src2 |dst | + * |:--------------|:--------------|:------|:--------------| + * |F16 |F16 |F16 |F16 | + * |F32 |F32 |F32 |F32 | + * |QASYMM8 |QASYMM8 |S32 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED |S32 |QASYMM8_SIGNED | + * * @param[in] input Source tensor. 3 lower dimensions represent a single input [width, height, IFM], * while every optional dimension from 4 and above represent a batch of inputs. * Data types supported: QASYMM8_SIGNED/QASYMM8/F16/F32. diff --git a/arm_compute/runtime/CL/functions/CLFFT1D.h b/arm_compute/runtime/CL/functions/CLFFT1D.h index 731bad5c32..c7112dc737 100644 --- a/arm_compute/runtime/CL/functions/CLFFT1D.h +++ b/arm_compute/runtime/CL/functions/CLFFT1D.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019-2020 Arm Limited. + * Copyright (c) 2019-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -61,6 +61,15 @@ public: ~CLFFT1D(); /** Initialise the function's source, destinations and border mode. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------|:------| + * |F32 |F32 | + * |F16 |F16 | + * * @param[in] input Source tensor. Data types supported: F16/F32. * @param[out] output Destination tensor. Data types and data layouts supported: Same as @p input. * @param[in] config FFT related configuration diff --git a/arm_compute/runtime/CL/functions/CLFFT2D.h b/arm_compute/runtime/CL/functions/CLFFT2D.h index adc8e46cb2..3d20327bf1 100644 --- a/arm_compute/runtime/CL/functions/CLFFT2D.h +++ b/arm_compute/runtime/CL/functions/CLFFT2D.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019-2020 Arm Limited. + * Copyright (c) 2019-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -58,6 +58,15 @@ public: ~CLFFT2D(); /** Initialise the function's source, destinations and border mode. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------|:------| + * |F32 |F32 | + * |F16 |F16 | + * * @param[in] input Source tensor. Data types supported: F16/F32. * @param[out] output Destination tensor. Data types and data layouts supported: Same as @p input. * @param[in] config FFT related configuration diff --git a/arm_compute/runtime/CL/functions/CLFFTConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLFFTConvolutionLayer.h index 5085f5a66c..f873cb0b86 100644 --- a/arm_compute/runtime/CL/functions/CLFFTConvolutionLayer.h +++ b/arm_compute/runtime/CL/functions/CLFFTConvolutionLayer.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019-2020 Arm Limited. + * Copyright (c) 2019-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -71,6 +71,15 @@ public: CLFFTConvolutionLayer &operator=(CLFFTConvolutionLayer &&) = default; /** Set the input and output tensors. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------|:------| + * |F32 |F32 | + * |F16 |F16 | + * * @note: This function only works with any square kernel size and unit strides for both NCHW and NHWC data layout * * @param[in] input Source tensor. 3 lower dimensions represent a single input [width, height, IFM], diff --git a/arm_compute/runtime/CL/functions/CLFill.h b/arm_compute/runtime/CL/functions/CLFill.h index 9a27d158a6..a01e0c3188 100644 --- a/arm_compute/runtime/CL/functions/CLFill.h +++ b/arm_compute/runtime/CL/functions/CLFill.h @@ -52,6 +52,14 @@ public: CLFill &operator=(CLFill &&); /** Initialize the kernel's tensor and filling value * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |All |All | + * * @param[in,out] tensor Input tensor to fill. Supported data types: All. * @param[in] constant_value The value used to fill the planes of the tensor * @param[in] window Window to be used in case setting only part of a tensor. Default is nullptr. diff --git a/arm_compute/runtime/CL/functions/CLFloor.h b/arm_compute/runtime/CL/functions/CLFloor.h index 7c7607002d..87cd5b44c7 100644 --- a/arm_compute/runtime/CL/functions/CLFloor.h +++ b/arm_compute/runtime/CL/functions/CLFloor.h @@ -54,6 +54,15 @@ public: CLFloor &operator=(CLFloor &&); /** Set the source, destination of the kernel * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------|:------| + * |F32 |F32 | + * |F16 |F16 | + * * @param[in] input Source tensor. Data type supported: F16/F32. * @param[out] output Destination tensor. Same as @p input */ diff --git a/arm_compute/runtime/CL/functions/CLPReluLayer.h b/arm_compute/runtime/CL/functions/CLPReluLayer.h index 7b6667044e..b123d0e46e 100644 --- a/arm_compute/runtime/CL/functions/CLPReluLayer.h +++ b/arm_compute/runtime/CL/functions/CLPReluLayer.h @@ -53,6 +53,17 @@ public: CLPReluLayer &operator=(CLPReluLayer &&); /** Set the input and output tensor. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |F16 |F16 | + * |F32 |F32 | + * * @note If the output tensor is a nullptr or is equal to the input, the activation function will be performed in-place * * @param[in] input Source tensor. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32. diff --git a/arm_compute/runtime/CL/functions/CLPermute.h b/arm_compute/runtime/CL/functions/CLPermute.h index e59cca2a67..8e15da2287 100644 --- a/arm_compute/runtime/CL/functions/CLPermute.h +++ b/arm_compute/runtime/CL/functions/CLPermute.h @@ -53,6 +53,15 @@ public: CLPermute &operator=(CLPermute &&) = default; /** Set the input and output tensors. * + * Valid data layouts: + * - NHWC + * - NCHW + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |All |All | + * * @note Arbitrary permutation vectors are supported with rank not greater than 4 * * @param[in] input The input tensor to permute. Data types supported: All. diff --git a/arm_compute/runtime/CL/functions/CLPixelWiseMultiplication.h b/arm_compute/runtime/CL/functions/CLPixelWiseMultiplication.h index 9f70b05a35..14422383ad 100644 --- a/arm_compute/runtime/CL/functions/CLPixelWiseMultiplication.h +++ b/arm_compute/runtime/CL/functions/CLPixelWiseMultiplication.h @@ -52,19 +52,23 @@ public: CLPixelWiseMultiplication &operator=(CLPixelWiseMultiplication &&); /** Initialise the kernel's inputs, output and convertion policy. * - * Valid configurations (Input1,Input2) -> Output : + * Valid data layouts: + * - All * - * - (U8,U8) -> U8 - * - (U8,U8) -> S16 - * - (U8,S16) -> S16 - * - (S16,U8) -> S16 - * - (S16,S16) -> S16 - * - (F16,F16) -> F16 - * - (F32,F32) -> F32 - * - (QASYMM8,QASYMM8) -> QASYMM8 - * - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED - * - (QSYMM16,QSYMM16) -> QSYMM16 - * - (QSYMM16,QSYMM16) -> S32 + * Valid data type configurations: + * |src0 |src1 |dst | + * |:--------------|:--------------|:--------------| + * |QASYMM8 |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |QSYMM16 |QSYMM16 |QASYMM16 | + * |QSYMM16 |QSYMM16 |S32 | + * |U8 |U8 |U8 | + * |U8 |U8 |S16 | + * |U8 |S16 |S16 | + * |S16 |U8 |S16 | + * |S16 |S16 |S16 | + * |F16 |F16 |F16 | + * |F32 |S32 |F32 | * * @param[in, out] input1 An input tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32. * The input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0. @@ -81,20 +85,6 @@ public: ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Initialise the kernel's inputs, output and convertion policy. * - * Valid configurations (Input1,Input2) -> Output : - * - * - (U8,U8) -> U8 - * - (U8,U8) -> S16 - * - (U8,S16) -> S16 - * - (S16,U8) -> S16 - * - (S16,S16) -> S16 - * - (F16,F16) -> F16 - * - (F32,F32) -> F32 - * - (QASYMM8,QASYMM8) -> QASYMM8 - * - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED - * - (QSYMM16,QSYMM16) -> QSYMM16 - * - (QSYMM16,QSYMM16) -> S32 - * * @param[in] compile_context The compile context to be used. * @param[in, out] input1 An input tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32. * The input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0. @@ -111,21 +101,6 @@ public: ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref CLPixelWiseMultiplication * - * Valid configurations (Input1,Input2) -> Output : - * - * - (U8,U8) -> U8 - * - (U8,U8) -> S16 - * - (U8,S16) -> S16 - * - (S16,U8) -> S16 - * - (S16,S16) -> S16 - * - (F16,F16) -> F16 - * - (F32,F32) -> F32 - * - (QASYMM8,QASYMM8) -> QASYMM8 - * - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED - * - (QSYMM16,QSYMM16) -> QSYMM16 - * - (QSYMM16,QSYMM16) -> S32 - * - * * @param[in] input1 An input tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32. * @param[in] input2 An input tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32. * @param[in] output The output tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32. diff --git a/arm_compute/runtime/CL/functions/CLPoolingLayer.h b/arm_compute/runtime/CL/functions/CLPoolingLayer.h index ec1c4eb768..902feca234 100644 --- a/arm_compute/runtime/CL/functions/CLPoolingLayer.h +++ b/arm_compute/runtime/CL/functions/CLPoolingLayer.h @@ -54,6 +54,18 @@ public: CLPoolingLayer &operator=(CLPoolingLayer &&) = default; /** Set the input and output tensors. * + * Valid data layouts: + * - NHWC + * - NCHW + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |F16 |F16 | + * |F32 |F32 | + * * @param[in,out] input Source tensor. (Written to only when padding != 0) Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32. * @param[out] output Destination tensor. Data types supported: Same as @p input. * @param[in] pool_info Contains pooling operation information described in @ref PoolingLayerInfo. diff --git a/arm_compute/runtime/CL/functions/CLQuantizationLayer.h b/arm_compute/runtime/CL/functions/CLQuantizationLayer.h index c5dad53513..6a44a226d4 100644 --- a/arm_compute/runtime/CL/functions/CLQuantizationLayer.h +++ b/arm_compute/runtime/CL/functions/CLQuantizationLayer.h @@ -59,6 +59,25 @@ public: CLQuantizationLayer &operator=(CLQuantizationLayer &&) = default; /** Set the input and output tensors. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------------------|:--------------| + * |QASYMM8 |QASYMM8 | + * |QASYMM8 |QASYMM8_SIGNED | + * |QASYMM8 |QASYMM16 | + * |QASYMM8_SIGNED |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |QASYMM8_SIGNED |QASYMM16 | + * |F16 |QASYMM8 | + * |F16 |QASYMM8_SIGNED | + * |F16 |QASYMM16 | + * |F32 |QASYMM8 | + * |F32 |QASYMM8_SIGNED | + * |F32 |QASYMM16 | + * * @param[in] input Source tensor. The dimensions over the third will be interpreted as batches. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/32. * @param[out] output Destination tensor with the same dimensions of input. Data types supported: QASYMM8/QASYMM8_SIGNED/QASYMM16. * diff --git a/arm_compute/runtime/CL/functions/CLReshapeLayer.h b/arm_compute/runtime/CL/functions/CLReshapeLayer.h index 60ed81680e..7346b65e9b 100644 --- a/arm_compute/runtime/CL/functions/CLReshapeLayer.h +++ b/arm_compute/runtime/CL/functions/CLReshapeLayer.h @@ -52,6 +52,14 @@ public: CLReshapeLayer &operator=(CLReshapeLayer &&); /** Initialise the kernel's inputs and outputs * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |All |All | + * * @param[in] input First tensor input. Data type supported: All * @param[out] output Output tensor. Data type supported: Same as @p input */ diff --git a/arm_compute/runtime/CL/functions/CLScale.h b/arm_compute/runtime/CL/functions/CLScale.h index 8a67f74bb6..ddb4a23531 100644 --- a/arm_compute/runtime/CL/functions/CLScale.h +++ b/arm_compute/runtime/CL/functions/CLScale.h @@ -55,6 +55,20 @@ public: /** Initialize the function's source, destination, interpolation type and border_mode. * + * Valid data layouts: + * - NHWC + * - NCHW + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |F16 |F16 | + * |F32 |F32 | + * |U8 |U8 | + * |S16 |S16 | + * * @param[in,out] input Source tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/F32. (Written to only for @p border_mode != UNDEFINED) * @param[out] output Destination tensor. Data types supported: Same as @p input * All but the lowest two dimensions must be the same size as in the input tensor, i.e. scaling is only performed within the XY-plane. diff --git a/arm_compute/runtime/CL/functions/CLSlice.h b/arm_compute/runtime/CL/functions/CLSlice.h index f17e77236d..7a7689c528 100644 --- a/arm_compute/runtime/CL/functions/CLSlice.h +++ b/arm_compute/runtime/CL/functions/CLSlice.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2020 Arm Limited. + * Copyright (c) 2018-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -91,6 +91,14 @@ public: CLSlice &operator=(CLSlice &&); /** Configure kernel * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |All |All | + * * @note Supported tensor rank: up to 4 * @note Start indices must be non-negative. 0 <= starts[i] * @note End coordinates can be negative, which represents the number of elements before the end of that dimension. diff --git a/arm_compute/runtime/CL/functions/CLStridedSlice.h b/arm_compute/runtime/CL/functions/CLStridedSlice.h index fdbef81f7d..6fab0c0186 100644 --- a/arm_compute/runtime/CL/functions/CLStridedSlice.h +++ b/arm_compute/runtime/CL/functions/CLStridedSlice.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2020 Arm Limited. + * Copyright (c) 2018-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -54,6 +54,14 @@ public: CLStridedSlice &operator=(CLStridedSlice &&); /** Configure kernel * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |All |All | + * * @note Supported tensor rank: up to 4 * * @param[in] input Source tensor. Data type supported: All. diff --git a/arm_compute/runtime/CL/functions/CLTranspose.h b/arm_compute/runtime/CL/functions/CLTranspose.h index 43cebeba90..a866aeabaa 100644 --- a/arm_compute/runtime/CL/functions/CLTranspose.h +++ b/arm_compute/runtime/CL/functions/CLTranspose.h @@ -53,6 +53,14 @@ public: CLTranspose &operator=(CLTranspose &&) = default; /** Initialise the kernel's inputs and output * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |All |All | + * * @param[in] input Input tensor. Data types supported: All. * @param[out] output Output tensor. Data type supported: Same as @p input */ diff --git a/arm_compute/runtime/NEON/functions/NEActivationLayer.h b/arm_compute/runtime/NEON/functions/NEActivationLayer.h index ffda8406aa..b39a8d7701 100644 --- a/arm_compute/runtime/NEON/functions/NEActivationLayer.h +++ b/arm_compute/runtime/NEON/functions/NEActivationLayer.h @@ -62,6 +62,18 @@ public: /** [NEActivationLayer snippet] **/ /** Set the input and output tensor. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |QSYMM16 |QSYMM16 | + * |F16 |F16 | + * |F32 |F32 | + * * @note If the output tensor is a nullptr or is equal to the input, the activation function will be performed in-place * * @param[in, out] input Source tensor. In case of @p output tensor = nullptr, this tensor will store the result diff --git a/arm_compute/runtime/NEON/functions/NEArithmeticAddition.h b/arm_compute/runtime/NEON/functions/NEArithmeticAddition.h index 8f9fd27906..734e3502dd 100644 --- a/arm_compute/runtime/NEON/functions/NEArithmeticAddition.h +++ b/arm_compute/runtime/NEON/functions/NEArithmeticAddition.h @@ -51,19 +51,25 @@ public: NEArithmeticAddition &operator=(NEArithmeticAddition &&); /** Initialise the kernel's inputs, output and conversion policy. * - * Valid configurations (Input1,Input2) -> Output : + * Valid data layouts: + * - NHWC + * - NCHW * - * - (U8,U8) -> U8 - * - (U8,U8) -> S16 - * - (S16,U8) -> S16 - * - (U8,S16) -> S16 - * - (S16,S16) -> S16 - * - (S32,S32) -> S32 - * - (F16,F16) -> F16 - * - (F32,F32) -> F32 - * - (QASYMM8,QASYMM8) -> QASYMM8 - * - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED - * - (QSYMM16,QSYMM16) -> QSYMM16 + * Valid data type configurations: + * |src0 |src1 |dst | + * |:--------------|:--------------|:--------------| + * |QASYMM8 |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |QSYMM16 |QSYMM16 |QASYMM16 | + * |QSYMM16 |QSYMM16 |S32 | + * |U8 |U8 |U8 | + * |U8 |U8 |S16 | + * |U8 |S16 |S16 | + * |S16 |U8 |S16 | + * |S16 |S16 |S16 | + * |S32 |S32 |S32 | + * |F16 |F16 |F16 | + * |F32 |F32 |F32 | * * @param[in] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/S32/F32 * @param[in] input2 Second tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/S32/F32 diff --git a/arm_compute/runtime/NEON/functions/NEConcatenateLayer.h b/arm_compute/runtime/NEON/functions/NEConcatenateLayer.h index 6aa724ab0c..dd1c709d76 100644 --- a/arm_compute/runtime/NEON/functions/NEConcatenateLayer.h +++ b/arm_compute/runtime/NEON/functions/NEConcatenateLayer.h @@ -55,6 +55,17 @@ public: NEConcatenateLayer &operator=(NEConcatenateLayer &&); /** Initialise the kernel's inputs vector and output. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |F16 |F16 | + * |F32 |F32 | + * * @note Input and output tensor dimensions preconditions defer depending on the concatenation axis. * @note Preconditions can be found respectively at @ref cpu::kernels::CpuConcatenateWidthKernel, @ref cpu::kernels::CpuConcatenateHeightKernel, * @ref cpu::kernels::CpuConcatenateDepthKernel and @ref cpu::kernels::CpuConcatenateBatchKernel. diff --git a/arm_compute/runtime/NEON/functions/NEConvertFullyConnectedWeights.h b/arm_compute/runtime/NEON/functions/NEConvertFullyConnectedWeights.h index f9ce66db13..218877d421 100644 --- a/arm_compute/runtime/NEON/functions/NEConvertFullyConnectedWeights.h +++ b/arm_compute/runtime/NEON/functions/NEConvertFullyConnectedWeights.h @@ -51,6 +51,15 @@ public: ~NEConvertFullyConnectedWeights(); /** Initialize the function. * + * Valid data layouts: + * - NHWC + * - NCHW + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |All |All | + * * @param[in] input Source weights tensor to convert. Must be 2 dimensional. Data types supported: All. * @param[out] output The converted weights tensor. Shape and Data Type: Same as @p input. * @param[in] original_input_shape Shape of the original input tensor (the one entering fully connected layer). diff --git a/arm_compute/runtime/NEON/functions/NECopy.h b/arm_compute/runtime/NEON/functions/NECopy.h index 56f8bd9803..ee02c259f4 100644 --- a/arm_compute/runtime/NEON/functions/NECopy.h +++ b/arm_compute/runtime/NEON/functions/NECopy.h @@ -53,6 +53,14 @@ public: NECopy &operator=(NECopy &&); /** Initialise the function's source and destination. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |All |All | + * * @param[in] input Source tensor. Data types supported: All * @param[out] output Output tensor. Data types supported: Same as @p input. * diff --git a/arm_compute/runtime/NEON/functions/NEDequantizationLayer.h b/arm_compute/runtime/NEON/functions/NEDequantizationLayer.h index a18566e6ca..dfec835f45 100644 --- a/arm_compute/runtime/NEON/functions/NEDequantizationLayer.h +++ b/arm_compute/runtime/NEON/functions/NEDequantizationLayer.h @@ -53,6 +53,23 @@ public: NEDequantizationLayer &operator=(NEDequantizationLayer &&) = default; /** Configure the kernel. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------------------|:--------------| + * |QASYMM8 |F16 | + * |QASYMM8 |F32 | + * |QASYMM8_SIGNED |F16 | + * |QASYMM8_SIGNED |F32 | + * |QSYMM8_PER_CHANNEL |F16 | + * |QSYMM8_PER_CHANNEL |F32 | + * |QSYMM8 |F16 | + * |QSYMM8 |F32 | + * |QSYMM16 |F16 | + * |QSYMM16 |F32 | + * * @param[in] input Source tensor. Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL/QSYMM8/QSYMM16. * @param[out] output Destination tensor with the same dimensions of input. Data type supported: F16/F32. */ diff --git a/arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h b/arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h index fc4017e635..82cabed6c9 100644 --- a/arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h +++ b/arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h @@ -58,6 +58,16 @@ public: ~NEDirectConvolutionLayer(); /** Set the input, weights, biases and output tensors. * + * Valid data layouts: + * - NHWC + * - NCHW + * + * Valid data type configurations: + * |src0 |src1 |src2 |dst | + * |:------|:------|:------|:------| + * |F16 |F16 |F16 |F16 | + * |F32 |F32 |F32 |F32 | + * * @note: DirectConvolution only works in the following configurations: * 1x1 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = F16/F32 * 3x3 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = F16/F32 diff --git a/arm_compute/runtime/NEON/functions/NEFFT1D.h b/arm_compute/runtime/NEON/functions/NEFFT1D.h index a533aa7f48..9654b1e604 100644 --- a/arm_compute/runtime/NEON/functions/NEFFT1D.h +++ b/arm_compute/runtime/NEON/functions/NEFFT1D.h @@ -63,6 +63,14 @@ public: ~NEFFT1D(); /** Initialise the function's source and destinations. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------|:------| + * |F32 |F32 | + * * @param[in] input Source tensor. Data types supported: F32. Number of channels supported: 1 (real tensor) or 2 (complex tensor). * @param[out] output Destination tensor. Data types and data layouts supported: Same as @p input. * Number of channels supported: 1 (real tensor) or 2 (complex tensor).If @p input is real, @p output must be complex. diff --git a/arm_compute/runtime/NEON/functions/NEFFT2D.h b/arm_compute/runtime/NEON/functions/NEFFT2D.h index ce84a85105..57f38d1942 100644 --- a/arm_compute/runtime/NEON/functions/NEFFT2D.h +++ b/arm_compute/runtime/NEON/functions/NEFFT2D.h @@ -58,6 +58,14 @@ public: ~NEFFT2D(); /** Initialise the function's source and destinations * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------|:------| + * |F32 |F32 | + * * @param[in] input Source tensor. Data types supported: F32. * @param[out] output Destination tensor. Data types and data layouts supported: Same as @p input. * @param[in] config FFT related configuration diff --git a/arm_compute/runtime/NEON/functions/NEFFTConvolutionLayer.h b/arm_compute/runtime/NEON/functions/NEFFTConvolutionLayer.h index 213fa6093b..c5f4d45b6b 100644 --- a/arm_compute/runtime/NEON/functions/NEFFTConvolutionLayer.h +++ b/arm_compute/runtime/NEON/functions/NEFFTConvolutionLayer.h @@ -73,6 +73,14 @@ public: ~NEFFTConvolutionLayer(); /** Set the input and output tensors. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------|:------| + * |F32 |F32 | + * * @note: This function only works with any square kernel size and unit strides for both NCHW and NHWC data layout * * @param[in] input Source tensor. 3 lower dimensions represent a single input [width, height, IFM], diff --git a/arm_compute/runtime/NEON/functions/NEFill.h b/arm_compute/runtime/NEON/functions/NEFill.h index ba5d020496..e923ce33e1 100644 --- a/arm_compute/runtime/NEON/functions/NEFill.h +++ b/arm_compute/runtime/NEON/functions/NEFill.h @@ -53,6 +53,14 @@ public: NEFill &operator=(NEFill &&); /** Initialize the function * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------|:------| + * |All |All | + * * @param[in,out] tensor Source tensor. Data types supported: All * @param[in] constant_value Constant value to use to fill tensor. */ diff --git a/arm_compute/runtime/NEON/functions/NEFloor.h b/arm_compute/runtime/NEON/functions/NEFloor.h index 9560eb9169..4d47b068db 100644 --- a/arm_compute/runtime/NEON/functions/NEFloor.h +++ b/arm_compute/runtime/NEON/functions/NEFloor.h @@ -54,6 +54,15 @@ public: ~NEFloor(); /** Set the source, destination of the kernel * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------|:------| + * |F32 |F32 | + * |F16 |F16 | + * * @param[in] input Source tensor. Data type supported: F16/F32. * @param[out] output Destination tensor. Same as @p input */ diff --git a/arm_compute/runtime/NEON/functions/NEPReluLayer.h b/arm_compute/runtime/NEON/functions/NEPReluLayer.h index b07febfe7f..81d5fd162c 100644 --- a/arm_compute/runtime/NEON/functions/NEPReluLayer.h +++ b/arm_compute/runtime/NEON/functions/NEPReluLayer.h @@ -55,6 +55,17 @@ public: NEPReluLayer &operator=(NEPReluLayer &&); /** Set the input and output tensor. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |F16 |F16 | + * |F32 |F32 | + * * @param[in] input Source tensor. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32. * @param[in] alpha Source alpha tensor. Data types supported: same of @p input. * @param[out] output Destination tensor. Data type supported: same as @p input diff --git a/arm_compute/runtime/NEON/functions/NEPermute.h b/arm_compute/runtime/NEON/functions/NEPermute.h index 2508458a3d..c863fde0ac 100644 --- a/arm_compute/runtime/NEON/functions/NEPermute.h +++ b/arm_compute/runtime/NEON/functions/NEPermute.h @@ -54,6 +54,15 @@ public: NEPermute &operator=(NEPermute &&) = default; /** Configure the permute function * + * Valid data layouts: + * - NHWC + * - NCHW + * + * Valid data type configurations: + * |src |dst | + * |:------|:------| + * |All |All | + * * @note Arbitrary permutation vectors are supported with rank not greater than 4 * * @param[in] input The input tensor to permute. Data types supported: All diff --git a/arm_compute/runtime/NEON/functions/NEPixelWiseMultiplication.h b/arm_compute/runtime/NEON/functions/NEPixelWiseMultiplication.h index 6f4cce3cde..f8074e791a 100644 --- a/arm_compute/runtime/NEON/functions/NEPixelWiseMultiplication.h +++ b/arm_compute/runtime/NEON/functions/NEPixelWiseMultiplication.h @@ -52,6 +52,24 @@ public: NEPixelWiseMultiplication &operator=(NEPixelWiseMultiplication &&) = default; /** Initialise the kernel's inputs, output and convertion policy. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src0 |src1 |dst | + * |:--------------|:--------------|:--------------| + * |QASYMM8 |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |QSYMM16 |QSYMM16 |QASYMM16 | + * |QSYMM16 |QSYMM16 |S32 | + * |U8 |U8 |U8 | + * |U8 |U8 |S16 | + * |U8 |S16 |S16 | + * |S16 |U8 |S16 | + * |S16 |S16 |S16 | + * |F16 |F16 |F16 | + * |F32 |S32 |F32 | + * * @note For @p scale equal to 1/255 only round to nearest even (implemented as round half up) is supported. * For all other scale values only round to zero (implemented as round towards minus infinity) is supported. * diff --git a/arm_compute/runtime/NEON/functions/NEPoolingLayer.h b/arm_compute/runtime/NEON/functions/NEPoolingLayer.h index cb136ebca9..851dc0ca32 100644 --- a/arm_compute/runtime/NEON/functions/NEPoolingLayer.h +++ b/arm_compute/runtime/NEON/functions/NEPoolingLayer.h @@ -59,6 +59,18 @@ public: ~NEPoolingLayer(); /** Set the input and output tensors. * + * Valid data layouts: + * - NHWC + * - NCHW + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |F16 |F16 | + * |F32 |F32 | + * * @note F16 is supported for pool sizes 2 and 3 only * * @param[in, out] input Source tensor. (Written to only when padding != 0) Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32. diff --git a/arm_compute/runtime/NEON/functions/NEQuantizationLayer.h b/arm_compute/runtime/NEON/functions/NEQuantizationLayer.h index 9e2d9ecf24..a7fadfc7cd 100644 --- a/arm_compute/runtime/NEON/functions/NEQuantizationLayer.h +++ b/arm_compute/runtime/NEON/functions/NEQuantizationLayer.h @@ -52,6 +52,25 @@ public: NEQuantizationLayer &operator=(NEQuantizationLayer &&) = default; /** Set the input and output tensors. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------------------|:--------------| + * |QASYMM8 |QASYMM8 | + * |QASYMM8 |QASYMM8_SIGNED | + * |QASYMM8 |QASYMM16 | + * |QASYMM8_SIGNED |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |QASYMM8_SIGNED |QASYMM16 | + * |F16 |QASYMM8 | + * |F16 |QASYMM8_SIGNED | + * |F16 |QASYMM16 | + * |F32 |QASYMM8 | + * |F32 |QASYMM8_SIGNED | + * |F32 |QASYMM16 | + * * @param[in] input Source tensor. The dimensions over the third will be interpreted as batches. Data types supported: QASYMM8/QASYMM8_SIGNED/F32/F16. * @param[out] output Destination tensor with the same dimensions of input. Data types supported: QASYMM8/QASYMM8_SIGNED/QASYMM16 */ diff --git a/arm_compute/runtime/NEON/functions/NEReshapeLayer.h b/arm_compute/runtime/NEON/functions/NEReshapeLayer.h index b8c0a841bc..3e6e33f797 100644 --- a/arm_compute/runtime/NEON/functions/NEReshapeLayer.h +++ b/arm_compute/runtime/NEON/functions/NEReshapeLayer.h @@ -52,6 +52,14 @@ public: NEReshapeLayer &operator=(NEReshapeLayer &&); /** Initialise the kernel's inputs and outputs * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------|:------| + * |All |All | + * * @param[in] input Input tensor. Data type supported: All * @param[out] output Output tensor. Data type supported: Same as @p input */ diff --git a/arm_compute/runtime/NEON/functions/NEScale.h b/arm_compute/runtime/NEON/functions/NEScale.h index 45658a7cd3..233ee2969e 100644 --- a/arm_compute/runtime/NEON/functions/NEScale.h +++ b/arm_compute/runtime/NEON/functions/NEScale.h @@ -48,6 +48,20 @@ public: ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(NEScale); /** Initialize the function's source, destination, interpolation type and border_mode. * + * Valid data layouts: + * - NHWC + * - NCHW + * + * Valid data type configurations: + * |src |dst | + * |:--------------|:--------------| + * |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |F16 |F16 | + * |F32 |F32 | + * |U8 |U8 | + * |S16 |S16 | + * * @param[in, out] input Source tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/U8/S16/F16/F32. (Written to only for @p border_mode != UNDEFINED) * @param[out] output Destination tensor. Data type supported: Same as @p input. All but the lowest two dimensions must be the same size as in the input tensor, i.e. scaling is only performed within the XY-plane. * @param[in] info @ref ScaleKernelInfo to be used for configuration diff --git a/arm_compute/runtime/NEON/functions/NESlice.h b/arm_compute/runtime/NEON/functions/NESlice.h index 28628778cb..214ffa512c 100644 --- a/arm_compute/runtime/NEON/functions/NESlice.h +++ b/arm_compute/runtime/NEON/functions/NESlice.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2020 Arm Limited. + * Copyright (c) 2018-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -89,6 +89,14 @@ public: /** Configure kernel * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------|:------| + * |All |All | + * * @note Supported tensor rank: up to 4 * @note Start indices must be non-negative. 0 <= starts[i] * @note End coordinates can be negative, which represents the number of elements before the end of that dimension. diff --git a/arm_compute/runtime/NEON/functions/NEStridedSlice.h b/arm_compute/runtime/NEON/functions/NEStridedSlice.h index f9c94f5301..7ba6a52a58 100644 --- a/arm_compute/runtime/NEON/functions/NEStridedSlice.h +++ b/arm_compute/runtime/NEON/functions/NEStridedSlice.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2020 Arm Limited. + * Copyright (c) 2018-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -95,6 +95,14 @@ public: /** Configure kernel * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------|:------| + * |All |All | + * * @note Supported tensor rank: up to 4 * * @param[in] input Source tensor. Data type supported: All diff --git a/arm_compute/runtime/NEON/functions/NETranspose.h b/arm_compute/runtime/NEON/functions/NETranspose.h index 78916f67b7..581fe74309 100644 --- a/arm_compute/runtime/NEON/functions/NETranspose.h +++ b/arm_compute/runtime/NEON/functions/NETranspose.h @@ -54,6 +54,14 @@ public: NETranspose &operator=(NETranspose &&) = default; /** Initialise the kernel's inputs and output * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src |dst | + * |:------|:------| + * |All |All | + * * @param[in] input Input tensor. Data types supported: All * @param[out] output Output tensor. Data type supported: Same as @p input */ diff --git a/arm_compute/runtime/OperatorList.h b/arm_compute/runtime/OperatorList.h new file mode 100644 index 0000000000..8c43c68b90 --- /dev/null +++ b/arm_compute/runtime/OperatorList.h @@ -0,0 +1,1028 @@ +/* + * Copyright (c) 2021 Arm Limited. + * + * SPDX-License-Identifier: MIT + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to + * deal in the Software without restriction, including without limitation the + * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or + * sell copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in all + * copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#ifndef ARM_COMPUTE_OPERATOR_LIST_H +#define ARM_COMPUTE_OPERATOR_LIST_H + +/** ActivationLayer + * + * Description: + * Function to simulate an activation layer with the specified activation function. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_ELU + * ANEURALNETWORKS_HARD_SWISH + * ANEURALNETWORKS_LOGISTIC + * ANEURALNETWORKS_RELU + * ANEURALNETWORKS_RELU1 + * ANEURALNETWORKS_RELU6 + * ANEURALNETWORKS_TANH + * + */ + +/** ArgMinMaxLayer (not ported) + * + * Description: + * Function to calculate the index of the minimum or maximum values in a tensor based on an axis. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_ARGMAX + * ANEURALNETWORKS_ARGMIN + * + */ + +/** ArithmeticAddition (no CL) + * + * Description: + * Function to add 2 tensors. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_ADD + * + */ + +/** ArithmeticSubtraction (no CL) + * + * Description: + * Function to substract 2 tensors. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_SUB + * + */ + +/** BatchNormalizationLayer (not ported) + * + * Description: + * @f[ out_i = \gamma * (\frac{in_i - \mu_{B}}{\sqrt{\sigma^2_{B} + \epsilon}}) + \beta \equiv BN_{\gamma,\beta}(in_i) @f] + * + * Equivalent Android NNAPI Op: + * None + * + */ + +/** BatchToSpaceLayer (not ported) + * + * Description: + * Rearranges (permutes) data from batch into blocks of spatial data, followed by cropping. It is the reverse transformation of SpaceToBatch (from TF website) + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_BATCH_TO_SPACE_ND + * + */ + +/** BitwiseAnd (not ported) + * + * Description: + * Function to performe bitwise AND between 2 tensors. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_LOGICAL_AND + * + */ + +/** BitwiseNot (not ported) + * + * Description: + * Function to performe bitwise NOT. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_LOGICAL_NOT + * + */ + +/** BitwiseOr (not ported) + * + * Description: + * Function to performe bitwise OR between 2 tensors. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_LOGICAL_OR + * + */ + +/** BitwiseXor (not ported) + * + * Description: + * Function to performe bitwise XOR between 2 tensors. + * + * Equivalent Android NNAPI Op: + * None + * + */ + +/** BoundingBoxTransform (not ported) + * + * Description: + * Function to . + * + * Equivalent Android NNAPI Op: + * ? + * + */ + +/** Cast (not ported) + * + * Description: + * Function to cast a tensor. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_CAST + * + */ + +/** ChannelShuffelLayer (not ported) + * + * Description: + * Function to cast a tensor. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_CHANNEL_SHUFFLE + * + */ + +/** Comparison (not ported) (only CL) + * + * Description: + * Function to cast a tensor. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_EQUAL + * ANEURALNETWORKS_GREATER + * ANEURALNETWORKS_GREATER_EQUAL + * ANEURALNETWORKS_LESS + * ANEURALNETWORKS_LESS_EQUAL + * ANEURALNETWORKS_NOT_EQUAL + * + */ + +/** ConcatenateLayer + * + * Description: + * Function to concatenate tensors along a given axis. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_CONCATENATION + * + */ + +/** ConvertFullyConnectedWeights + * + * Description: + * Function to tranpose the wieghts for the fully connected layer. + * + * Equivalent Android NNAPI Op: + * None + * + */ + +/** ConvolutionLayer (not ported) + * + * Description: + * Function to compute a convolution layer. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_CONV_2D + * + */ + +/** Copy + * + * Description: + * Function to copy a tensor. + * + * Equivalent Android NNAPI Op: + * None + * + */ + +/** Crop (only CL) + * + * Description: + * Function to . + * + * Equivalent Android NNAPI Op: + * ? + * + */ + +/** CropResize (not ported) + * + * Description: + * Function to . + * + * Equivalent Android NNAPI Op: + * ? + * + */ + +/** DeconvolutionLayer (not ported) + * + * Description: + * Function to . + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_TRANSPOSE_CONV_2D + * + */ + +/** DeconvolutionLayerUpsample (only CL) (not ported) + * + * Description: + * Function to . + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_TRANSPOSE_CONV_2D + * + */ + +/** DepthConverterLayer (not ported) + * + * Description: + * Function to . + * + * Equivalent Android NNAPI Op: + * None + * + */ + +/** DepthToSpaceLayer (not ported) + * + * Description: + * Function to . + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_DEPTH_TO_SPACE + * + */ + +/** DepthwiseConvolutionLayer (not ported) + * + * Description: + * Function to perform depthwise separable convolution + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_DEPTHWISE_CONV_2D + * + */ + +/** DequantizationLayer + * + * Description: + * Function to dequantize the values in a tensor + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_DEQUANTIZE + * + */ + +/** DetectionPostProcessLayer (not ported) (no CL) + * + * Description: + * Function to generate the detection output based on center size encoded boxes, class prediction and anchors by doing non maximum suppression (NMS) + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_DETECTION_POSTPROCESSING + * + */ + +/** DirectConvolutionLayer + * + * Description: + * Function to + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_CONV_2D + * + */ + +/** DirectDeconvolutionLayer (only CL) + * + * Description: + * Function to + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_TRANSPOSE_CONV_2D + * + */ + +/** ElementWiseOperations (skip) + * + * Description: + * Function to perform in Cpu: + * - Div + * - Max + * - Min + * - Pow + * - SquaredDiff + * - Comparisons (Equal, greater, greater_equal, less, less_equal, not_equal) + * + * Function to perform in CL: + * - Add + * - Sub + * - Div + * - Max + * - Min + * - Pow + * - SquaredDiff + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_MAXIMUM + * ANEURALNETWORKS_MINIMUM + * ANEURALNETWORKS_POW + * ANEURALNETWORKS_DIV + * ANEURALNETWORKS_ADD (only CL) + * ANEURALNETWORKS_SUB (only CL) + * ANEURALNETWORKS_EQUAL (no CL) + * ANEURALNETWORKS_GREATER (no CL) + * ANEURALNETWORKS_GREATER_EQUAL (no CL) + * ANEURALNETWORKS_LESS (no CL) + * ANEURALNETWORKS_LESS_EQUAL (no CL) + * ANEURALNETWORKS_NOT_EQUAL (no CL) + * + */ + +/** ElementWiseOperationUnary (skip) + * + * Description: + * Function to perform: + * - Rsqrt + * - Exp + * - Neg + * - Log + * - Abs + * - Round + * - Sin + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_ABS + * ANEURALNETWORKS_EXP + * ANEURALNETWORKS_LOG + * ANEURALNETWORKS_NEG + * ANEURALNETWORKS_RSQRT + * ANEURALNETWORKS_SIN + * + */ + +/** FFT1D + * + * Description: + * Fast Fourier Transform 1D + * + * Equivalent Android NNAPI Op: + * None + * + */ + +/** FFT2D + * + * Description: + * Fast Fourier Transform 2D + * + * Equivalent Android NNAPI Op: + * None + * + */ + +/** FFTConvolutionLayer + * + * Description: + * Fast Fourier Transform Convolution + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_CONV_2D + * + */ + +/** Fill + * + * Description: + * Set the values of a tensor with a given value + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_FILL + * + */ + +/** FillBorder (not ported) + * + * Description: + * + * + * Equivalent Android NNAPI Op: + * ? + * + */ + +/** FlattenLayer (not ported) + * + * Description: + * Reshape a tensor to be 1D + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_RESHAPE + * + */ + +/** Floor + * + * Description: + * Round the value to the lowest number + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_FLOOR + * + */ + +/** FullyConnectedLayer (not ported) + * + * Description: + * Function to perform a fully connected / dense layer + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_FULLY_CONNECTED + * + */ + +/** FuseBatchNormalization (not ported) + * + * Description: + * Function to . + * + * Equivalent Android NNAPI Op: + * None + * + */ + +/** Gather (not ported) + * + * Description: + * Function to . + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_GATHER + * + */ + +/** GEMM (not ported) + * + * Description: + * General Matrix Multiplication. + * + * Equivalent Android NNAPI Op: + * None + * + */ + +/** GEMMConv2D (not ported) (no CL) + * + * Description: + * General Matrix Multiplication. + * + * Equivalent Android NNAPI Op: + * None + * + */ + +/** GEMMConvolutionLayer (not ported) + * + * Description: + * General Matrix Multiplication. + * + * Equivalent Android NNAPI Op: + * None + * + */ + +/** GEMMDeconvolutionLayer (not ported) (only CL) + * + * Description: + * General Matrix Multiplication. + * + * Equivalent Android NNAPI Op: + * None + * + */ + +/** GEMMLowpMatrixMultiplyCore (not ported) + * + * Description: + * General Matrix Multiplication. + * + * Equivalent Android NNAPI Op: + * None + * + */ + +/** GEMMLowpOutputStage (not ported) + * + * Description: + * General Matrix Multiplication. + * + * Equivalent Android NNAPI Op: + * None + * + */ + +/** GenerateProposalsLayer (not ported) + * + * Description: + * Function to generate proposals for a RPN (Region Proposal Network). + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_GENERATE_PROPOSALS + * + */ + +/** InstanceNormalizationLayer (not ported) + * + * Description: + * Function to perform a Instance normalization on a given axis. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_INSTANCE_NORMALIZATION + * + */ + +/** L2NormalizationLayer (not ported) + * + * Description: + * Function to perform a L2 normalization on a given axis. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_L2_NORMALIZATION + * + */ + +/** Logical (no CL) + * + * Description: + * Function to perform: + * - Logical AND + * - Logical OR + * - Logical NOT + * + * Equivalent Android NNAPI Op: + * None? + * + */ + +/** LogicalAnd (only CL) + * + * Description: + * Function to perform Logical AND + * + * Equivalent Android NNAPI Op: + * None? + * + */ + +/** LogicalOr (only CL) + * + * Description: + * Function to perform Logical OR + * + * Equivalent Android NNAPI Op: + * None? + * + */ + +/** LogicalNot (only CL) + * + * Description: + * Function to perform Logical NOT + * + * Equivalent Android NNAPI Op: + * None? + * + */ + +/** LSTMLayer (not ported) + * + * Description: + * Function to perform LSTM + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_LSTM + * + */ + +/** LSTMLayerQuantized (not ported) + * + * Description: + * Function to perform LSTM + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_QUANTIZED_LSTM + * ANEURALNETWORKS_QUANTIZED_16BIT_LSTM ? + * + */ + +/** MaxUnpoolingLayer (not ported) + * + * Description: + * Function to perform MaxUnpooling + * + * Equivalent Android NNAPI Op: + * ? + * + */ + +/** MeanStdDevNormalizationLayer (not ported) + * + * Description: + * Function to execute mean and standard deviation normalization. + * + * Equivalent Android NNAPI Op: + * None ? + * + */ + +/** MeanStdDevNormalizationLayer (not ported) + * + * Description: + * Function to execute mean and standard deviation normalization. + * + * Equivalent Android NNAPI Op: + * None ? + * + */ + +/** NormalizationLayer (not ported) + * + * Description: + * Function to compute normalization layer. + * + * Equivalent Android NNAPI Op: + * None ? + * + */ + +/** PadLayer (not ported) + * + * Description: + * Function to pad a tensor. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_PAD + * ANEURALNETWORKS_PAD_V2 + * + */ + +/** Permute + * + * Description: + * Function to transpose an ND tensor. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_TRANSPOSE + * + */ + +/** PixelWiseMultiplication + * + * Description: + * Function to performe a multiplication. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_MUL + * + */ + +/** PoolingLayer + * + * Description: + * Function to performe pooling with the specified pooling operation. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_AVERAGE_POOL_2D + * ANEURALNETWORKS_L2_POOL_2D + * ANEURALNETWORKS_MAX_POOL_2D + * + */ + +/** PReluLayer + * + * Description: + * Function to compute the activation layer with the PRELU activation function. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_PRELU + * + */ + +/** PriorBoxLayer (not ported) + * + * Description: + * Function to compute the activation layer with the PRELU activation function. + * + * Equivalent Android NNAPI Op: + * ? + * + */ + +/** QLSTMLayer (not ported) + * + * Description: + * Function to perform LSTM + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_QUANTIZED_LSTM + * ANEURALNETWORKS_QUANTIZED_16BIT_LSTM ? + * + */ + +/** QuantizationLayer + * + * Description: + * Function to perform quantization layer + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_QUANTIZE + * + */ + +/** Range (not ported) + * + * Description: + * Function to . + * + * Equivalent Android NNAPI Op: + * none? + * + */ + +/** RecudeMean (not ported) + * + * Description: + * Function to performe reduce mean operation. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_MEAN + * + */ + +/** RecudeOperation (not ported) + * + * Description: + * Function to performe reduce mean operation. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_REDUCE_ALL + * ANEURALNETWORKS_REDUCE_ANY + * ANEURALNETWORKS_REDUCE_MAX + * ANEURALNETWORKS_REDUCE_MIN + * ANEURALNETWORKS_REDUCE_PROD + * ANEURALNETWORKS_REDUCE_SUM + * + */ + +/** RecudeOperation (not ported) + * + * Description: + * Function to performe reduce with the following operations + * - ARG_IDX_MAX: Index of the max value + * - ARG_IDX_MIN: Index of the min value + * - MEAN_SUM: Mean of sum + * - PROD: Product + * - SUM_SQUARE: Sum of squares + * - SUM: Sum + * - MIN: Min + * - MAX: Max + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_REDUCE_ALL + * ANEURALNETWORKS_REDUCE_ANY + * ANEURALNETWORKS_REDUCE_MAX + * ANEURALNETWORKS_REDUCE_MIN + * ANEURALNETWORKS_REDUCE_PROD + * ANEURALNETWORKS_REDUCE_SUM + * + */ + +/** ReorgLayer (not ported) + * + * Description: + * Function to performe reorg + * + * Equivalent Android NNAPI Op: + * None? + * + */ + +/** ReshapeLayer + * + * Description: + * Fucntion to reshape a tensor + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_RESHAPE + * ANEURALNETWORKS_SQUEEZE + * + */ + +/** ReverseLayer (not ported) + * + * Description: + * Fucntion to . + * + * Equivalent Android NNAPI Op: + * None? + * + */ + +/** RNNLayer (not ported) + * + * Description: + * Fucntion to perform RNN . + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_RNN + * + */ + +/** ROIAligmentLayer (not ported) + * + * Description: + * Fucntion to perform RNN . + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_ROI_ALIGN + * + */ + +/** ROIPoolingLayer (not ported) + * + * Description: + * Fucntion to perform RNN . + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_ROI_POOLING + * + */ + +/** Scale + * + * Description: + * Fucntion to perform resize a tensor using to interpolate: + * - Bilenear + * - Nearest neighbor + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_RESIZE_BILINEAR + * ANEURALNETWORKS_RESIZE_NEAREST_NEIGHBOR + * + */ + +/** Select (not ported) + * + * Description: + * Fucntion to select values from 2 tensors depending on an input tensor of booleans. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_SELECT + * + */ + +/** Slice + * + * Description: + * Function to perform tensor slicing. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_SLICE + * + */ + +/** SoftmaxLayer (skip) + * + * Description: + * Function to compute a SoftmaxLayer and a Log SoftmaxLayer. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_LOG_SOFTMAX + * ANEURALNETWORKS_SOFTMAX + * + */ + +/** SpaceToBatchLayer (not ported) + * + * Description: + * Function to divide a tensor spatially. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_SPACE_TO_BATCH_ND + * + */ + +/** SpaceToDepthLayer (not ported) + * + * Description: + * Function to rearrange blocks of spatial data into depth. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_SPACE_TO_DEPTH + * + */ + +/** Split (not ported) + * + * Description: + * Function to split a tensor along a given axis. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_SPLIT + * + */ + +/** StackLayer (not ported) + * + * Description: + * Function to stack tensors along an axis. + * + * Equivalent Android NNAPI Op: + * none + * + */ + +/** StridedSlice + * + * Description: + * Function to extract a strided slice of a tensor. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_STRIDED_SLICE + * + */ + +/** Tile (not ported) + * + * Description: + * Function to construct a tensor by tiling a given tensor. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_TILE + * + */ + +/** Transpose + * + * Description: + * Function to transpose an 2D tensor. + * + * Equivalent Android NNAPI Op: + * ANEURALNETWORKS_TRANSPOSE + * + */ + +/** Unstack (not ported) + * + * Description: + * Function to unpack a rank-R tensor into rank-(R-1) tensors. + * + * Equivalent Android NNAPI Op: + * none + * + */ + +/** WinogradConvolutionLayer (not ported) + * + * Description: + * Function to. + * + * Equivalent Android NNAPI Op: + * None + * + */ + +/** WinogradInputTransform (not ported) (only CL) + * + * Description: + * Function to. + * + * Equivalent Android NNAPI Op: + * None + * + */ + +#endif /* ARM_COMPUTE_OPERATOR_LIST_H */
\ No newline at end of file diff --git a/docs/09_operators_list.dox b/docs/09_operators_list.dox new file mode 100644 index 0000000000..82a127bbd3 --- /dev/null +++ b/docs/09_operators_list.dox @@ -0,0 +1,784 @@ +/// +/// Copyright (c) 2021 Arm Limited. +/// +/// SPDX-License-Identifier: MIT +/// +/// Permission is hereby granted, free of charge, to any person obtaining a copy +/// of this software and associated documentation files (the "Software"), to +/// deal in the Software without restriction, including without limitation the +/// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +/// sell copies of the Software, and to permit persons to whom the Software is +/// furnished to do so, subject to the following conditions: +/// +/// The above copyright notice and this permission notice shall be included in all +/// copies or substantial portions of the Software. +/// +/// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +/// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +/// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +/// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +/// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +/// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +/// SOFTWARE. +/// +namespace arm_compute +{ +/** +@page operators_list Supported Operators + +@tableofcontents + +@section S9_1_operators_list Supported Operators + +Compute Library supports operators that are listed in below table. + +Compute Library supports a wide list of data-types, information can been directly found in the documentation of each kernel/function. +The main data-types that the Machine Learning functions support are the following: + <ul> + <li>BFLOAT16: 16-bit non-standard brain floating point + <li>QASYMM8: 8-bit unsigned asymmetric quantized + <li>QASYMM8_SIGNED: 8-bit signed asymmetric quantized + <li>QSYMM8_PER_CHANNEL: 8-bit signed symmetric quantized (Used for the weights) + <li>QSYMM8: 8-bit unsigned symmetric quantized + <li>QSYMM16: 16-bit unsigned symmetric quantized + <li>F32: 32-bit single precision floating point + <li>F16: 16-bit half precision floating point + <li>S32: 32-bit signed integer + <li>U8: 8-bit unsigned char + <li>All: include all above data types + </ul> + +Compute Library supports the following data layouts (fast changing dimension from right to left): + <ul> + <li>NHWC: The native layout of Compute Library that delivers the best performance where channels are in the fastest changing dimension + <li>NCHW: Legacy layout where width is in the fastest changing dimension + <li>All: include all above data layouts + </ul> +where N = batches, C = channels, H = height, W = width + +<table> +<caption id="multi_row"></caption> +<tr> + <th>Function + <th>Description + <th>Equivalent Android NNAPI Op + <th>Backends + <th>Data Layouts + <th>Data Types +<tr> + <td rowspan="2">ActivationLayer + <td rowspan="2" style="width:200px;"> Function to simulate an activation layer with the specified activation function. + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_ELU + <li>ANEURALNETWORKS_HARD_SWISH + <li>ANEURALNETWORKS_LOGISTIC + <li>ANEURALNETWORKS_RELU + <li>ANEURALNETWORKS_RELU1 + <li>ANEURALNETWORKS_RELU6 + <li>ANEURALNETWORKS_TANH + </ul> + <td>NEActivationLayer + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>QASYMM8<td>QASYMM8 + <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED + <tr><td>QSYMM16<td>QSYMM16 + <tr><td>F16<td>F16 + <tr><td>F32<td>F32 + </table> +<tr> + <td>CLActivationLayer + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>QASYMM8<td>QASYMM8 + <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED + <tr><td>QSYMM16<td>QSYMM16 + <tr><td>F16<td>F16 + <tr><td>F32<td>F32 + </table> +<tr> + <td rowspan="2">ConcatenateLayer + <td rowspan="2" style="width:200px;"> Function to concatenate tensors along a given axis. + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_CONCATENATION + </ul> + <td>NEConcatenateLayer + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>QASYMM8<td>QASYMM8 + <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED + <tr><td>F16<td>F16 + <tr><td>F32<td>F32 + </table> +<tr> + <td>CLConcatenateLayer + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>QASYMM8<td>QASYMM8 + <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED + <tr><td>F16<td>F16 + <tr><td>F32<td>F32 + </table> +<tr> + <td rowspan="2">ConvertFullyConnectedWeights + <td rowspan="2" style="width:200px;"> Function to tranpose the wieghts for the fully connected layer. + <td rowspan="2"> + <ul> + <li>None + </ul> + <td>NEConvertFullyConnectedWeights + <td> + <ul> + <li>NHWC + <li>NCHW + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>All<td>All + </table> +<tr> + <td>CLConvertFullyConnectedWeights + <td> + <ul> + <li>NHWC + <li>NCHW + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>All<td>All + </table> +<tr> + <td rowspan="2">Copy + <td rowspan="2" style="width:200px;"> Function to copy a tensor. + <td rowspan="2"> + <ul> + <li>None + </ul> + <td>NECopy + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>All<td>All + </table> +<tr> + <td>CLCopy + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>All<td>All + </table> +<tr> + <td rowspan="2">DequantizationLayer + <td rowspan="2" style="width:200px;"> Function to dequantize the values in a tensor + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_DEQUANTIZE + </ul> + <td>NEDequantizationLayer + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>QASYMM8<td>F16 + <tr><td>QASYMM8<td>F32 + <tr><td>QASYMM8_SIGNED<td>F16 + <tr><td>QASYMM8_SIGNED<td>F32 + <tr><td>QSYMM8_PER_CHANNEL<td>F16 + <tr><td>QSYMM8_PER_CHANNEL<td>F32 + <tr><td>QSYMM8<td>F16 + <tr><td>QSYMM8<td>F32 + <tr><td>QSYMM16<td>F16 + <tr><td>QSYMM16<td>F32 + </table> +<tr> + <td>CLDequantizationLayer + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>QASYMM8<td>F16 + <tr><td>QASYMM8<td>F32 + <tr><td>QASYMM8_SIGNED<td>F16 + <tr><td>QASYMM8_SIGNED<td>F32 + <tr><td>QSYMM8_PER_CHANNEL<td>F16 + <tr><td>QSYMM8_PER_CHANNEL<td>F32 + <tr><td>QSYMM8<td>F16 + <tr><td>QSYMM8<td>F32 + <tr><td>QSYMM16<td>F16 + <tr><td>QSYMM16<td>F32 + </table> +<tr> + <td rowspan="2">DirectConvolutionLayer + <td rowspan="2" style="width:200px;"> Function to + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_CONV_2D + </ul> + <td>NEDirectConvolutionLayer + <td> + <ul> + <li>NHWC + <li>NCHW + </ul> + <td> + <table> + <tr><th>src0<th>src1<th>src2<th>dst + <tr><td>F16<td>F16<td>F16<td>F16 + <tr><td>F32<td>F32<td>F32<td>F32 + </table> +<tr> + <td>CLDirectConvolutionLayer + <td> + <ul> + <li>NHWC + <li>NCHW + </ul> + <td> + <table> + <tr><th>src0<th>src1<th>src2<th>dst + <tr><td>F16<td>F16<td>F16<td>F16 + <tr><td>F32<td>F32<td>F32<td>F32 + <tr><td>QASYMM8<td>QASYMM8<td>S32<td>QASYMM8 + <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>S32<td>QASYMM8_SIGNED + </table> +<tr> + <td rowspan="2">FFT1D + <td rowspan="2" style="width:200px;"> Fast Fourier Transform 1D + <td rowspan="2"> + <ul> + <li>None + </ul> + <td>NEFFT1D + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>F32<td>F32 + </table> +<tr> + <td>CLFFT1D + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>F32<td>F32 + <tr><td>F16<td>F16 + </table> +<tr> + <td rowspan="2">FFT2D + <td rowspan="2" style="width:200px;"> Fast Fourier Transform 2D + <td rowspan="2"> + <ul> + <li>None + </ul> + <td>NEFFT2D + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>F32<td>F32 + </table> +<tr> + <td>CLFFT2D + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>F32<td>F32 + <tr><td>F16<td>F16 + </table> +<tr> + <td rowspan="2">FFTConvolutionLayer + <td rowspan="2" style="width:200px;"> Fast Fourier Transform Convolution + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_CONV_2D + </ul> + <td>NEFFTConvolutionLayer + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>F32<td>F32 + </table> +<tr> + <td>CLFFTConvolutionLayer + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>F32<td>F32 + <tr><td>F16<td>F16 + </table> +<tr> + <td rowspan="2">Fill + <td rowspan="2" style="width:200px;"> Set the values of a tensor with a given value + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_FILL + </ul> + <td>NEFill + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>All<td>All + </table> +<tr> + <td>CLFill + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>All<td>All + </table> +<tr> + <td rowspan="2">Floor + <td rowspan="2" style="width:200px;"> Round the value to the lowest number + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_FLOOR + </ul> + <td>NEFloor + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>F32<td>F32 + <tr><td>F16<td>F16 + </table> +<tr> + <td>CLFloor + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>F32<td>F32 + <tr><td>F16<td>F16 + </table> +<tr> + <td rowspan="2">Permute + <td rowspan="2" style="width:200px;"> Function to transpose an ND tensor. + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_TRANSPOSE + </ul> + <td>NEPermute + <td> + <ul> + <li>NHWC + <li>NCHW + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>All<td>All + </table> +<tr> + <td>CLPermute + <td> + <ul> + <li>NHWC + <li>NCHW + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>All<td>All + </table> +<tr> + <td rowspan="2">PixelWiseMultiplication + <td rowspan="2" style="width:200px;"> Function to performe a multiplication. + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_MUL + </ul> + <td>NEPixelWiseMultiplication + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src0<th>src1<th>dst + <tr><td>QASYMM8<td>QASYMM8<td>QASYMM8 + <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>QASYMM8_SIGNED + <tr><td>QSYMM16<td>QSYMM16<td>QASYMM16 + <tr><td>QSYMM16<td>QSYMM16<td>S32 + <tr><td>U8<td>U8<td>U8 + <tr><td>U8<td>U8<td>S16 + <tr><td>U8<td>S16<td>S16 + <tr><td>S16<td>U8<td>S16 + <tr><td>S16<td>S16<td>S16 + <tr><td>F16<td>F16<td>F16 + <tr><td>F32<td>S32<td>F32 + </table> +<tr> + <td>CLPixelWiseMultiplication + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src0<th>src1<th>dst + <tr><td>QASYMM8<td>QASYMM8<td>QASYMM8 + <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED<td>QASYMM8_SIGNED + <tr><td>QSYMM16<td>QSYMM16<td>QASYMM16 + <tr><td>QSYMM16<td>QSYMM16<td>S32 + <tr><td>U8<td>U8<td>U8 + <tr><td>U8<td>U8<td>S16 + <tr><td>U8<td>S16<td>S16 + <tr><td>S16<td>U8<td>S16 + <tr><td>S16<td>S16<td>S16 + <tr><td>F16<td>F16<td>F16 + <tr><td>F32<td>S32<td>F32 + </table> +<tr> + <td rowspan="2">PoolingLayer + <td rowspan="2" style="width:200px;"> Function to performe pooling with the specified pooling operation. + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_AVERAGE_POOL_2D + <li>ANEURALNETWORKS_L2_POOL_2D + <li>ANEURALNETWORKS_MAX_POOL_2D + </ul> + <td>NEPoolingLayer + <td> + <ul> + <li>NHWC + <li>NCHW + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>QASYMM8<td>QASYMM8 + <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED + <tr><td>F16<td>F16 + <tr><td>F32<td>F32 + </table> +<tr> + <td>CLPoolingLayer + <td> + <ul> + <li>NHWC + <li>NCHW + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>QASYMM8<td>QASYMM8 + <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED + <tr><td>F16<td>F16 + <tr><td>F32<td>F32 + </table> +<tr> + <td rowspan="2">PReluLayer + <td rowspan="2" style="width:200px;"> Function to compute the activation layer with the PRELU activation function. + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_PRELU + </ul> + <td>NEPReluLayer + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>QASYMM8<td>QASYMM8 + <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED + <tr><td>F16<td>F16 + <tr><td>F32<td>F32 + </table> +<tr> + <td>CLPReluLayer + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>QASYMM8<td>QASYMM8 + <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED + <tr><td>F16<td>F16 + <tr><td>F32<td>F32 + </table> +<tr> + <td rowspan="2">QuantizationLayer + <td rowspan="2" style="width:200px;"> Function to perform quantization layer + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_QUANTIZE + </ul> + <td>NEQuantizationLayer + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>QASYMM8<td>QASYMM8 + <tr><td>QASYMM8<td>QASYMM8_SIGNED + <tr><td>QASYMM8<td>QASYMM16 + <tr><td>QASYMM8_SIGNED<td>QASYMM8 + <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED + <tr><td>QASYMM8_SIGNED<td>QASYMM16 + <tr><td>F16<td>QASYMM8 + <tr><td>F16<td>QASYMM8_SIGNED + <tr><td>F16<td>QASYMM16 + <tr><td>F32<td>QASYMM8 + <tr><td>F32<td>QASYMM8_SIGNED + <tr><td>F32<td>QASYMM16 + </table> +<tr> + <td>CLQuantizationLayer + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>QASYMM8<td>QASYMM8 + <tr><td>QASYMM8<td>QASYMM8_SIGNED + <tr><td>QASYMM8<td>QASYMM16 + <tr><td>QASYMM8_SIGNED<td>QASYMM8 + <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED + <tr><td>QASYMM8_SIGNED<td>QASYMM16 + <tr><td>F16<td>QASYMM8 + <tr><td>F16<td>QASYMM8_SIGNED + <tr><td>F16<td>QASYMM16 + <tr><td>F32<td>QASYMM8 + <tr><td>F32<td>QASYMM8_SIGNED + <tr><td>F32<td>QASYMM16 + </table> +<tr> + <td rowspan="2">ReshapeLayer + <td rowspan="2" style="width:200px;"> Fucntion to reshape a tensor + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_RESHAPE + <li>ANEURALNETWORKS_SQUEEZE + </ul> + <td>NEReshapeLayer + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>All<td>All + </table> +<tr> + <td>CLReshapeLayer + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>All<td>All + </table> +<tr> + <td rowspan="2">Scale + <td rowspan="2" style="width:200px;"> Fucntion to perform resize a tensor using to interpolate: - Bilenear - Nearest neighbor + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_RESIZE_BILINEAR + <li>ANEURALNETWORKS_RESIZE_NEAREST_NEIGHBOR + </ul> + <td>NEScale + <td> + <ul> + <li>NHWC + <li>NCHW + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>QASYMM8<td>QASYMM8 + <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED + <tr><td>F16<td>F16 + <tr><td>F32<td>F32 + <tr><td>U8<td>U8 + <tr><td>S16<td>S16 + </table> +<tr> + <td>CLScale + <td> + <ul> + <li>NHWC + <li>NCHW + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>QASYMM8<td>QASYMM8 + <tr><td>QASYMM8_SIGNED<td>QASYMM8_SIGNED + <tr><td>F16<td>F16 + <tr><td>F32<td>F32 + <tr><td>U8<td>U8 + <tr><td>S16<td>S16 + </table> +<tr> + <td rowspan="2">Slice + <td rowspan="2" style="width:200px;"> Function to perform tensor slicing. + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_SLICE + </ul> + <td>NESlice + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>All<td>All + </table> +<tr> + <td>CLSlice + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>All<td>All + </table> +<tr> + <td rowspan="2">StridedSlice + <td rowspan="2" style="width:200px;"> Function to extract a strided slice of a tensor. + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_STRIDED_SLICE + </ul> + <td>NEStridedSlice + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>All<td>All + </table> +<tr> + <td>CLStridedSlice + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>All<td>All + </table> +<tr> + <td rowspan="2">Transpose + <td rowspan="2" style="width:200px;"> Function to transpose an 2D tensor. + <td rowspan="2"> + <ul> + <li>ANEURALNETWORKS_TRANSPOSE + </ul> + <td>NETranspose + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>All<td>All + </table> +<tr> + <td>CLTranspose + <td> + <ul> + <li>All + </ul> + <td> + <table> + <tr><th>src<th>dst + <tr><td>All<td>All + </table> +</table> + +*/ +} // namespace
\ No newline at end of file diff --git a/docs/Doxyfile b/docs/Doxyfile index 9d711486c0..6fb5de7020 100644 --- a/docs/Doxyfile +++ b/docs/Doxyfile @@ -777,6 +777,7 @@ INPUT = ./docs/00_introduction.dox \ ./docs/06_functions_list.dox \ ./docs/07_errata.dox \ ./docs/08_api.dox \ + ./docs/09_operators_list.dox \ ./docs/ComputeLibrary.dir \ ./arm_compute/ \ ./src/ \ |