diff options
Diffstat (limited to 'arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h')
| -rw-r--r-- | arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h | 193 |
1 files changed, 42 insertions, 151 deletions
diff --git a/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h b/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h index 188117f674..9fd0b4aaef 100644 --- a/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h +++ b/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2020 ARM Limited. + * Copyright (c) 2017-2021, 2023 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -24,113 +24,19 @@ #ifndef ARM_COMPUTE_CLFULLYCONNECTEDLAYER_H #define ARM_COMPUTE_CLFULLYCONNECTEDLAYER_H -#include "arm_compute/runtime/CL/ICLSimpleFunction.h" - -#include "arm_compute/core/CL/kernels/CLTransposeKernel.h" +#include "arm_compute/function_info/FullyConnectedLayerInfo.h" #include "arm_compute/runtime/CL/CLTensor.h" -#include "arm_compute/runtime/CL/functions/CLConvertFullyConnectedWeights.h" -#include "arm_compute/runtime/CL/functions/CLFlattenLayer.h" -#include "arm_compute/runtime/CL/functions/CLGEMM.h" -#include "arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h" +#include "arm_compute/runtime/IFunction.h" #include "arm_compute/runtime/IWeightsManager.h" #include "arm_compute/runtime/MemoryGroup.h" namespace arm_compute { -/** Basic function to reshape the weights of Fully Connected layer with OpenCL. This function calls the following kernels: - * - * -# @ref CLTransposeKernel - * - * @note The fully connected layer accepts "weights" tensors only with 2 dimensions. - */ -class CLFullyConnectedLayerReshapeWeights : public ICLSimpleFunction -{ -public: - /** Set the input and output tensors. - * - * @param[in] input Weights tensor. The weights must be 2 dimensional. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32. - * @param[out] output Destination tensor which stores the transposed input tensor. Data type supported: Same as @p input. - */ - void configure(const ICLTensor *input, ICLTensor *output); - /** Set the input and output tensors. - * - * @param[in] compile_context The compile context to be used. - * @param[in] input Weights tensor. The weights must be 2 dimensional. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32. - * @param[out] output Destination tensor which stores the transposed input tensor. Data type supported: Same as @p input. - */ - void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output); - /** Static function to check if given info will lead to a valid configuration of @ref CLFullyConnectedLayerReshapeWeights - * - * @param[in] input Weights tensor. The weights must be 2 dimensional. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32. - * @param[in] output Destination tensor which stores the transposed input tensor. Data type supported: Same as @p input. - * - * @return a status - */ - static Status validate(const ITensorInfo *input, const ITensorInfo *output); -}; - -namespace weights_transformations -{ -/** Basic function to manage the reshape weights generated from @ref CLFullyConnectedLayerReshapeWeights */ -class CLFullyConnectedLayerReshapeWeightsManaged : public ITransformWeights -{ -public: - //Inherited method override - void run() override - { - _output.allocator()->allocate(); - _func.run(); - _reshape_run = true; - } - - //Inherited method override - void release() override - { - _output.allocator()->free(); - } - - //Inherited method override - ICLTensor *get_weights() override - { - return &_output; - } - - //Inherited method override - uint32_t uid() override - { - return _uid; - } - - /** Configures the @ref CLFullyConnectedLayerReshapeWeights function - * - * @param[in] input Source tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/F16/F32. - */ - void configure(const ICLTensor *input) - { - configure(CLKernelLibrary::get().get_compile_context(), input); - } - /** Configures the @ref CLFullyConnectedLayerReshapeWeights function - * - * @param[in] compile_context The compile context to be used. - * @param[in] input Source tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/F16/F32. - */ - void configure(const CLCompileContext &compile_context, const ICLTensor *input) - { - _func.configure(compile_context, input, &_output); - } - -private: - static constexpr uint32_t _uid = 0x0; - CLTensor _output{}; - CLFullyConnectedLayerReshapeWeights _func{}; -}; -} // namespace weights_transformations - /** Basic function to compute a Fully Connected layer on OpenCL. This function calls the following OpenCL kernels: * - * -# @ref CLIm2ColKernel (called when the input comes from a convolutional layer) - * -# @ref CLFullyConnectedLayerReshapeWeights (if @p are_weights_reshaped is set to false and transpose_weights is set to true ) (called once) - * -# @ref CLGEMMMatrixMultiplyKernel or @ref CLGEMMLowpMatrixMultiplyCore (if quantized asymmetric) + * -# @ref opencl::kernels::ClIm2ColKernel (called when the input comes from a convolutional layer) + * -# @ref CLTranspose (if @p are_weights_reshaped is set to false and transpose_weights is set to true ) (called once) + * -# @ref opencl::ClGemm or @ref CLGEMMLowpMatrixMultiplyCore (if quantized asymmetric) * * @note The fully connected layer accepts "weights" tensors only with 2 dimensions. */ @@ -138,7 +44,10 @@ class CLFullyConnectedLayer : public IFunction { public: /** Constructor */ - CLFullyConnectedLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr, IWeightsManager *weights_manager = nullptr); + CLFullyConnectedLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr, + IWeightsManager *weights_manager = nullptr); + /** Default destructor */ + ~CLFullyConnectedLayer(); /** Prevent instances of this class from being copied (As this class contains pointers) */ CLFullyConnectedLayer(const CLFullyConnectedLayer &) = delete; /** Default move constructor */ @@ -149,21 +58,17 @@ public: CLFullyConnectedLayer &operator=(CLFullyConnectedLayer &&) = default; /** Set the input and output tensors. * - * @param[in] input Source tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/F16/F32. - * @param[in] weights Weights tensor. The weights must be 2 dimensional. - * If this function is called after a Convolution Layer, the (transposed) weights will have as many rows as the product of the first 3 input's dimensions. - * If it is called after another FullyConnected Layer, the (transposed) weights will have as many rows as the input's first dimension. - * Data type supported: Same as @p input. - * @param[in] biases Bias tensor. Can be nullptr. Data type supported:Same as @p input. - * @param[out] output Destination tensor. Its shape should be equal to the output of a matrix multiplication between: - * - The output of im2col on the input and the (transposed) 2D weights, if the function is called after a Convolution Layer - * - The input tensor and the (transposed) 2D weights, if the function is called after another FullyConnected Layer. - * Data type supported: Same as @p input. - * @param[in] fc_info (Optional) Fully connected layer additional info - */ - void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, - FullyConnectedLayerInfo fc_info = FullyConnectedLayerInfo()); - /** 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] compile_context The compile context to be used. * @param[in] input Source tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/F16/F32. @@ -178,25 +83,31 @@ public: * Data type supported: Same as @p input. * @param[in] fc_info (Optional) Fully connected layer additional info */ - void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, + void configure(const CLCompileContext &compile_context, + const ICLTensor *input, + const ICLTensor *weights, + const ICLTensor *biases, + ICLTensor *output, + FullyConnectedLayerInfo fc_info = FullyConnectedLayerInfo()); + /** Set the input and output tensors. + * + * Similar to @ref CLFullyConnectedLayer + */ + void configure(const ICLTensor *input, + const ICLTensor *weights, + const ICLTensor *biases, + ICLTensor *output, FullyConnectedLayerInfo fc_info = FullyConnectedLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref CLFullyConnectedLayer * - * @param[in] input Source tensor info. Data type supported: QASYMM8/QASYMM8_SIGNED/F16/F32. - * @param[in] weights Weights tensor info. The weights must be 2 dimensional. - * If this function is called after a Convolution Layer, the (transposed) weights will have as many rows as the product of the first 3 input's dimensions. - * If it is called after another FullyConnected Layer, the (transposed) weights will have as many rows as the input's first dimension. - * Data type supported: Same as @p input. - * @param[in] biases Bias tensor info. Can be nullptr. Data type supported:Same as @p input. - * @param[out] output Destination tensor info. Its shape should be equal to the output of a matrix multiplication between: - * - The output of im2col on the input and the (transposed) 2D weights, if the function is called after a Convolution Layer - * - The input tensor and the (transposed) 2D weights, if the function is called after another FullyConnected Layer. - * Data type supported: Same as @p input. - * @param[in] fc_info (Optional) Fully connected layer additional info + * Similar to @ref CLFullyConnectedLayer * * @return a status */ - static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, + static Status validate(const ITensorInfo *input, + const ITensorInfo *weights, + const ITensorInfo *biases, + const ITensorInfo *output, FullyConnectedLayerInfo fc_info = FullyConnectedLayerInfo()); //Inherited methods override @@ -204,28 +115,8 @@ public: void prepare() override; private: - void configure_fc_fc(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *weights, const ICLTensor *bias, ICLTensor *output, const FullyConnectedLayerInfo &fc_info); - void configure_conv_fc(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *weights, const ICLTensor *bias, ICLTensor *output, const FullyConnectedLayerInfo &fc_info); - void configure_mm(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *weights, const ICLTensor *bias, ICLTensor *output, const FullyConnectedLayerInfo &fc_info); - - MemoryGroup _memory_group; - IWeightsManager *_weights_manager; - CLConvertFullyConnectedWeights _convert_weights; - weights_transformations::CLConvertFullyConnectedWeightsManaged _convert_weights_managed; - weights_transformations::CLFullyConnectedLayerReshapeWeightsManaged _reshape_weights_managed_function; - CLFlattenLayer _flatten_layer; - CLFullyConnectedLayerReshapeWeights _reshape_weights_function; - CLGEMM _mm_gemm; - CLGEMMLowpMatrixMultiplyCore _mm_gemmlowp; - CLTensor _flatten_output; - CLTensor _converted_weights_output; - CLTensor _reshape_weights_output; - bool _are_weights_converted; - bool _are_weights_reshaped; - bool _is_fc_after_conv; - bool _is_quantized; - bool _is_prepared; - const ICLTensor *_original_weights; + struct Impl; + std::unique_ptr<Impl> _impl; }; } // namespace arm_compute #endif /* ARM_COMPUTE_CLFULLYCONNECTEDLAYER_H */ |