diff options
Diffstat (limited to 'arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h')
| -rw-r--r-- | arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h | 89 |
1 files changed, 63 insertions, 26 deletions
diff --git a/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h b/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h index e2ed0e0abc..aabe42f928 100644 --- a/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h +++ b/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2020 ARM Limited. + * Copyright (c) 2017-2021, 2023 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -24,15 +24,14 @@ #ifndef ARM_COMPUTE_NEDECONVOLUTIONLAYER_H #define ARM_COMPUTE_NEDECONVOLUTIONLAYER_H -#include "arm_compute/runtime/CPP/functions/CPPUpsample.h" -#include "arm_compute/runtime/NEON/functions/NEConvolutionLayer.h" -#include "arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h" -#include "arm_compute/runtime/NEON/functions/NEReverse.h" - #include "arm_compute/core/Types.h" +#include "arm_compute/runtime/CPP/functions/CPPUpsample.h" #include "arm_compute/runtime/IFunction.h" #include "arm_compute/runtime/IMemoryManager.h" #include "arm_compute/runtime/MemoryGroup.h" +#include "arm_compute/runtime/NEON/functions/NEConvolutionLayer.h" +#include "arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h" +#include "arm_compute/runtime/NEON/functions/NEReverse.h" #include "arm_compute/runtime/Tensor.h" #include <memory> @@ -64,11 +63,10 @@ namespace arm_compute * The weights used by Deconvolution are supposed to be the same as the ones used for Convolution. Therefore, it will be necessary to use the weights in the * reverse order to perform an actual convolution. This is achieved by using @ref NEReverse. * - * This function calls the following NEON kernels/functions: + * This function calls the following kernels/functions: * * -# @ref CPPUpsample * -# @ref NEConvolutionLayer - * -# @ref NEPermute * -# @ref NEReverse * */ @@ -77,39 +75,77 @@ class NEDeconvolutionLayer : public IFunction public: /** Constructor */ NEDeconvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr); - /** Prevent instances of this class from being copied (As this class contains pointers) */ NEDeconvolutionLayer(const NEDeconvolutionLayer &) = delete; + /** Default move constructor */ + NEDeconvolutionLayer(NEDeconvolutionLayer &&) = default; /** Prevent instances of this class from being copied (As this class contains pointers) */ NEDeconvolutionLayer &operator=(const NEDeconvolutionLayer &) = delete; - /** Allow instances of this class to be moved */ - NEDeconvolutionLayer(NEDeconvolutionLayer &&) = default; - /** Allow instances of this class to be moved */ + /** Default move assignment operator */ NEDeconvolutionLayer &operator=(NEDeconvolutionLayer &&) = default; /** Default destructor */ - virtual ~NEDeconvolutionLayer() = default; + ~NEDeconvolutionLayer() = default; /** Set the input, weights, biases and output tensors. * - * @param[in,out] input Input tensor. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: F32/F16/QASYMM8/QASYMM8_SIGNED. - * @param[in] weights The 4d weights with dimensions [width, height, IFM, OFM]. Data type supported: Same as @p input. - * @param[in] bias Optional, ignored if NULL. The biases have one dimension. Data type supported: Data types supported: S32 for QASYMM8 and QASYMM8_SIGNED input, F32 for F32 input, F16 for F16 input. - * @param[out] output Output tensor. The output has the same number of dimensions as the @p input. - * @param[in] info Contains padding and policies to be used in the deconvolution, this is decribed in @ref PadStrideInfo. + * 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 |QSYMM8_PER_CHANNEL |S32 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED |S32 |QASYMM8_SIGNED | + * |QASYMM8_SIGNED |QSYMM8_PER_CHANNEL |S32 |QASYMM8_SIGNED | + * + * @param[in,out] input Input tensor. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. + * Data types supported: F32/F16/QASYMM8/QASYMM8_SIGNED. + * @param[in] weights The 4d weights with dimensions [width, height, IFM, OFM]. + * Data type supported: Same as @p input, also could be QSYMM8_PER_CHANNEL if input is QASYMM8/QASYMM8_SIGNED. + * @param[in] bias Optional, ignored if NULL. The biases have one dimension. + * Data type supported: Data types supported: S32 for QASYMM8/QASYMM8_SIGNED input, F32 for F32 input, F16 for F16 input. + * @param[out] output Output tensor. The output has the same number of dimensions as the @p input. + * @param[in] info Contains padding and policies to be used in the deconvolution, this is described in @ref PadStrideInfo. + * @param[in] enable_fast_math (Optional) Enable fast math computation. In case this flag were set, the function could dispatch the fastest implementation + * available which may introduce a drop of accuracy as well. Default is false + * @param[in] weights_info (Optional) Specifies the weight format. Default is unspecified. This parameter can be used to specify the weight format that is optimal for + * the GEMM convolution. * */ - void configure(ITensor *input, const ITensor *weights, const ITensor *bias, ITensor *output, const PadStrideInfo &info); + void configure(ITensor *input, + const ITensor *weights, + const ITensor *bias, + ITensor *output, + const PadStrideInfo &info, + bool enable_fast_math = false, + const WeightsInfo &weights_info = WeightsInfo()); /** Static function to check if given info will lead to a valid configuration of @ref NEDeconvolutionLayer * - * @param[in] input Input tensor info. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: F32/F16/QASYMM8/QASYMM8_SIGNED. - * @param[in] weights The 4d weights info with dimensions [width, height, IFM, OFM]. Data type supported: Same as @p input. - * @param[in] bias (Optional) The biases have one dimension. Data type supported: Data types supported: S32 for QASYMM8 and QASYMM8_SIGNED input, F32 for F32 input, F16 for F16 input. - * @param[in] output Output tensor info. The output has the same number of dimensions as the @p input. - * @param[in] info Contains padding and policies to be used in the deconvolution, this is decribed in @ref PadStrideInfo. + * @param[in] input Input tensor info. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. + * Data types supported: F32/F16/QASYMM8/QASYMM8_SIGNED. + * @param[in] weights The 4d weights info with dimensions [width, height, IFM, OFM]. + * Data type supported: Same as @p input, also could be QSYMM8_PER_CHANNEL if input is QASYMM8/QASYMM8_SIGNED. + * @param[in] bias (Optional) The biases have one dimension. Data type supported: Data types supported: S32 for QASYMM8/QASYMM8_SIGNED input, F32 for F32 input, F16 for F16 input. + * @param[in] output Output tensor info. The output has the same number of dimensions as the @p input. + * @param[in] info Contains padding and policies to be used in the deconvolution, this is described in @ref PadStrideInfo. + * @param[in] enable_fast_math (Optional) Enable fast math computation. In case this flag were set, the function could dispatch the fastest implementation + * available which may introduce a drop of accuracy as well. Default is false + * @param[in] weights_info (Optional) Specifies the weight format. Default is unspecified. This parameter can be used to specify the weight format that is optimal for + * the GEMM convolution. * * @return a status */ - static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *bias, const ITensorInfo *output, const PadStrideInfo &info); + static Status validate(const ITensorInfo *input, + const ITensorInfo *weights, + const ITensorInfo *bias, + const ITensorInfo *output, + const PadStrideInfo &info, + bool enable_fast_math = false, + const WeightsInfo &weights_info = WeightsInfo()); // Inherited methods overridden: void run() override; @@ -127,6 +163,7 @@ private: ITensor *_input; PadStrideInfo _info; bool _is_prepared; + bool _do_upsampling; }; -} // arm_compute +} // namespace arm_compute #endif /* ARM_COMPUTE_NEDECONVOLUTIONLAYER_H */ |