diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-03-26 17:23:28 +0000 |
---|---|---|
committer | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2019-04-11 09:34:26 +0000 |
commit | 8be9148814b88e5b0cabd5a4d2b1f4ff470a8c1c (patch) | |
tree | 760658b8c7b8917379467bd3fc119a5502faa850 /arm_compute/runtime/CL/functions/CLFFTConvolutionLayer.h | |
parent | a50e702289af66944e860eafc7f3b32f6c5f30be (diff) | |
download | ComputeLibrary-8be9148814b88e5b0cabd5a4d2b1f4ff470a8c1c.tar.gz |
COMPMID-1959: Implements 2D FFT on OpenCL
Change-Id: I73cf3984a5463acc854c8a59dc2bd9a5234cd99c
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/936
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Diffstat (limited to 'arm_compute/runtime/CL/functions/CLFFTConvolutionLayer.h')
-rw-r--r-- | arm_compute/runtime/CL/functions/CLFFTConvolutionLayer.h | 154 |
1 files changed, 154 insertions, 0 deletions
diff --git a/arm_compute/runtime/CL/functions/CLFFTConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLFFTConvolutionLayer.h new file mode 100644 index 0000000000..0fd2cf3cb1 --- /dev/null +++ b/arm_compute/runtime/CL/functions/CLFFTConvolutionLayer.h @@ -0,0 +1,154 @@ +/* + * Copyright (c) 2019 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_CLFFTCONVOLUTIONLAYER_H__ +#define __ARM_COMPUTE_CLFFTCONVOLUTIONLAYER_H__ + +#include "arm_compute/runtime/IFunction.h" + +#include "arm_compute/core/Types.h" +#include "arm_compute/runtime/CL/functions/CLActivationLayer.h" +#include "arm_compute/runtime/CL/functions/CLElementwiseOperations.h" +#include "arm_compute/runtime/CL/functions/CLFFT2D.h" +#include "arm_compute/runtime/CL/functions/CLPadLayer.h" +#include "arm_compute/runtime/CL/functions/CLPermute.h" +#include "arm_compute/runtime/CL/functions/CLPixelWiseMultiplication.h" +#include "arm_compute/runtime/CL/functions/CLReductionOperation.h" +#include "arm_compute/runtime/CL/functions/CLReshapeLayer.h" +#include "arm_compute/runtime/CL/functions/CLReverse.h" +#include "arm_compute/runtime/CL/functions/CLSlice.h" + +namespace arm_compute +{ +// Forward declarations +class ICLTensor; + +/** Basic function to execute FFT-based convolution on OpenCL. This function calls the following OpenCL functions/kernels: + * + * -# @ref CLPermute Permute input if NHWC(only NCHW is supported). + * -# @ref CLPadLayer Pad input. + * -# @ref CLFFT2D Forward transform to the frequency domain. + * -# @ref CLComplexPixelWiseMultiplication Complex element-wise product of input and the weights. + * -# @ref CLReductionOperation Reduction across channels. + * -# @ref CLFFT2D Inverse transform back to the time domain. + * -# @ref CLStridedSlice Extract valid output. + * -# @ref CLArithmeticAddition Add bias. + * -# @ref CLActivationLayer Perform activation. + * -# @ref CLPermute Permute output if NHWC(only NCHW is supported). + */ +class CLFFTConvolutionLayer : public IFunction +{ +public: + /** Default constructor */ + CLFFTConvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr); + /** Prevent instances of this class from being copied (As this class contains pointers) */ + CLFFTConvolutionLayer(const CLFFTConvolutionLayer &) = delete; + /** Default move constructor */ + CLFFTConvolutionLayer(CLFFTConvolutionLayer &&) = default; + /** Prevent instances of this class from being copied (As this class contains pointers) */ + CLFFTConvolutionLayer &operator=(const CLFFTConvolutionLayer &) = delete; + /** Default move assignment operator */ + CLFFTConvolutionLayer &operator=(CLFFTConvolutionLayer &&) = default; + /** Set the input and output tensors. + * + * @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], + * while every optional dimension from 4 and above represent a batch of inputs. + * Data types supported: F32. + * @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. Data type supported:Same as @p input. + * @param[in] biases Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM].Data type supported: Same as @p input + * @param[out] output Destination tensor. 3 lower dimensions represent a single output [width, height, OFM], while the rest represent batch of outputs. + * Data types supported: Same as @p input. + * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. + */ + void configure(ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); + /** Static function to check if given info will lead to a valid configuration of @ref CLFFTConvolutionLayer + * + * @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], + * while every optional dimension from 4 and above represent a batch of inputs. + * Data types supported: F32. + * @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. Data type supported:Same as @p input. + * @param[in] biases Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM].Data type supported: Same as @p input + * @param[out] output Destination tensor. 3 lower dimensions represent a single output [width, height, OFM], while the rest represent batch of outputs. + * Data types supported: Same as @p input. + * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. + * + * @return a status + */ + static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); + + // Inherited methods overridden: + void run() override; + void prepare() override; + +private: + CLMemoryGroup _memory_group; + CLReverse _flip_weights_func; + CLPermute _permute_input_func; + CLPermute _permute_output_func; + CLPermute _permute_weights_func; + CLPermute _permute_bias_func; + CLPadLayer _pad_input_func; + CLPadLayer _pad_weights_func; + CLFFT2D _transform_input_func; + CLFFT2D _transform_weights_func; + CLFFT2D _itransform_output_func; + CLComplexPixelWiseMultiplication _prod_func; + CLReductionOperation _reduce_func; + CLSlice _extract_output_func; + CLArithmeticAddition _bias_add_func; + CLActivationLayer _activation_layer_func; + + CLTensor _permuted_input; + CLTensor _permuted_weights; + CLTensor _permuted_bias; + CLTensor _permuted_output; + CLTensor _padded_input; + CLTensor _padded_weights; + CLTensor _flip_axis; + CLTensor _flipped_weights; + CLTensor _transformed_input; + CLTensor _transformed_weights; + CLTensor _input_weights_product; + CLTensor _output_product; + CLTensor _output_reduced; + CLTensor _itransformed_output; + CLTensor _reshaped_output; + CLTensor _bias_output; + + const ICLTensor *_original_weights; + const ICLTensor *_original_bias; + bool _is_activationlayer_enabled; + bool _needs_permute; + bool _has_bias; + bool _is_prepared; +}; +} // namespace arm_compute +#endif /* __ARM_COMPUTE_CLFFTCONVOLUTIONLAYER_H__ */ |