From dcf4c87cf78a5f1667699c1a3511d09356938660 Mon Sep 17 00:00:00 2001 From: Giorgio Arena Date: Fri, 16 Apr 2021 12:41:45 +0100 Subject: CLDepthwiseConvolutionLayer rework - Part 1 Remove the reshaped variant for CLDepthwiseConvolutionLayer 3x3 NHWC Quantized - Remove kernel selection by GPUTarget - Remove unused quantized support from the NHWC kernel - Remove CLDepthwiseConvolutionLayerReshapeWeightsKernel - Remove OpenCL kernels for reshaped dwc 3x3 quantized and weights reshape - Remove the "_bifrost" suffix in common OpenCL kernel - Remove the ICLDepthwiseConvolutionLayer3x3Kernel common interface Resolve COMPMID-3864, COMPMID-3907 Change-Id: Icfac0fb6c00e214985beb05dad7c0cdbbee7d830 Signed-off-by: Giorgio Arena Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5447 Tested-by: Arm Jenkins Reviewed-by: Michele Di Giorgio Comments-Addressed: Arm Jenkins --- .../CLDepthwiseConvolutionLayer3x3NHWCKernel.h | 103 ++++++++++----------- 1 file changed, 50 insertions(+), 53 deletions(-) (limited to 'src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.h') diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.h b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.h index ce0bf5ceb3..ee47d98807 100644 --- a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.h +++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2020 Arm Limited. + * Copyright (c) 2018-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -24,7 +24,7 @@ #ifndef ARM_COMPUTE_CLDEPTHWISECONVOLUTIONNHWCKERNEL3x3_H #define ARM_COMPUTE_CLDEPTHWISECONVOLUTIONNHWCKERNEL3x3_H -#include "src/core/CL/kernels/ICLDepthwiseConvolutionLayer3x3Kernel.h" +#include "src/core/CL/ICLKernel.h" namespace arm_compute { @@ -32,81 +32,78 @@ class ICLTensor; /** Interface for the kernel to run a 3x3 depthwise convolution on a tensor when the data layout is NHWC. */ -class CLDepthwiseConvolutionLayer3x3NHWCKernel : public ICLDepthwiseConvolutionLayer3x3Kernel +class CLDepthwiseConvolutionLayer3x3NHWCKernel : public ICLKernel { public: /** Default constructor */ CLDepthwiseConvolutionLayer3x3NHWCKernel(); + /** Prevent instances of this class from being copied (As this class contains pointers) */ + CLDepthwiseConvolutionLayer3x3NHWCKernel(const CLDepthwiseConvolutionLayer3x3NHWCKernel &) = delete; + /** Prevent instances of this class from being copied (As this class contains pointers) */ + CLDepthwiseConvolutionLayer3x3NHWCKernel &operator=(const CLDepthwiseConvolutionLayer3x3NHWCKernel &) = delete; + /** Default Move Constructor. */ + CLDepthwiseConvolutionLayer3x3NHWCKernel(CLDepthwiseConvolutionLayer3x3NHWCKernel &&) = default; + /** Default move assignment operator */ + CLDepthwiseConvolutionLayer3x3NHWCKernel &operator=(CLDepthwiseConvolutionLayer3x3NHWCKernel &&) = default; /** Default move assignment operator. */ /** Initialize the function's source, destination, conv and border_size. * - * @param[in] input Source tensor. DataType supported: QASYMM8/QASYMM8_SIGNED/F16/F32. - * @param[in] weights Weights tensor. A 3D tensor with dimensions [IFM, 3, 3]. - * Data type supported: Same as @p input or QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL when @p input is QASYMM8/QASYMM8_SIGNED. - * @param[in] biases Biases tensor. A 1D tensor with dimensions [IFM]. Must be nullptr if not needed. - * Data type supported: Same as @p input, S32 when input is QASYMM8/QASYMM8_SIGNED. - * @param[out] output Destination tensor. Data type supported: Same as @p input. - * @param[in] conv_info Padding and stride information to use for the convolution. - * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1. - * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU are supported. - * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). - * @param[in] output_multipliers (Optional) Output multipliers tensor for quantized computations. In case of per-channel quantization, - * the number of multipliers must be equal to the number of filters (IFM). Supported data types: S32 - * @param[in] output_shifts (Optional) Output shifts tensor for quantized computations. In case of per-channel quantization, - * the number of multipliers must be equal to the number of filters (IFM). Supported data types: S32 + * @param[in] input Source tensor. DataType supported: F16/F32. + * @param[in] weights Weights tensor. A 3D tensor with dimensions [IFM, 3, 3]. + * Data type supported: Same as @p input. + * @param[in] biases Biases tensor. A 1D tensor with dimensions [IFM]. Must be nullptr if not needed. + * Data type supported: Same as @p input. + * @param[out] output Destination tensor. Data type supported: Same as @p input. + * @param[in] conv_info Padding and stride information to use for the convolution. + * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU are supported. + * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). */ void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, - unsigned int depth_multiplier = 1, ActivationLayerInfo act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U), - const ICLTensor *output_multipliers = nullptr, const ICLTensor *output_shifts = nullptr) override; + unsigned int depth_multiplier = 1, ActivationLayerInfo act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U)); /** Initialize the function's source, destination, conv and border_size. * - * @param[in] compile_context The compile context to be used. - * @param[in] input Source tensor. DataType supported: QASYMM8/QASYMM8_SIGNED/F16/F32. - * @param[in] weights Weights tensor. A 3D tensor with dimensions [IFM, 3, 3]. - * Data type supported: Same as @p input or QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL when @p input is QASYMM8/QASYMM8_SIGNED. - * @param[in] biases Biases tensor. A 1D tensor with dimensions [IFM]. Must be nullptr if not needed. - * Data type supported: Same as @p input, S32 when input is QASYMM8/QASYMM8_SIGNED. - * @param[out] output Destination tensor. Data type supported: Same as @p input. - * @param[in] conv_info Padding and stride information to use for the convolution. - * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1. - * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU are supported. - * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). - * @param[in] output_multipliers (Optional) Output multipliers tensor for quantized computations. In case of per-channel quantization, - * the number of multipliers must be equal to the number of filters (IFM). Supported data types: S32 - * @param[in] output_shifts (Optional) Output shifts tensor for quantized computations. In case of per-channel quantization, - * the number of multipliers must be equal to the number of filters (IFM). Supported data types: S32 + * @param[in] compile_context The compile context to be used. + * @param[in] input Source tensor. DataType supported: F16/F32. + * @param[in] weights Weights tensor. A 3D tensor with dimensions [IFM, 3, 3]. + * Data type supported: Same as @p input. + * @param[in] biases Biases tensor. A 1D tensor with dimensions [IFM]. Must be nullptr if not needed. + * Data type supported: Same as @p input. + * @param[out] output Destination tensor. Data type supported: Same as @p input. + * @param[in] conv_info Padding and stride information to use for the convolution. + * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU are supported. + * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). */ void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, - unsigned int depth_multiplier = 1, ActivationLayerInfo act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U), - const ICLTensor *output_multipliers = nullptr, const ICLTensor *output_shifts = nullptr) override; + unsigned int depth_multiplier = 1, ActivationLayerInfo act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U)); /** Static function to check if given info will lead to a valid configuration of @ref CLDepthwiseConvolutionLayer3x3NHWCKernel * - * @param[in] input Source tensor info. DataType supported: QASYMM8/QASYMM8_SIGNED/F16/F32. - * @param[in] weights Weights tensor info. A 3D tensor with dimensions [IFM, 3, 3]. - * Data type supported: Same as @p input or QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL when @p input is QASYMM8/QASYMM8_SIGNED. - * @param[in] biases Biases tensor info. A 1D tensor with dimensions [IFM]. Must be nullptr if not needed. - * Data type supported: Same as @p input, S32 when input is QASYMM8/QASYMM8_SIGNED. - * @param[in] output Destination tensor info. Data type supported: Same as @p input. - * @param[in] conv_info Padding and stride information to use for the convolution. - * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1. - * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU are supported. - * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). - * @param[in] output_multipliers (Optional) Output multipliers tensor info for quantized computations. In case of per-channel quantization, - * the number of multipliers must be equal to the number of filters (IFM). Supported data types: S32 - * @param[in] output_shifts (Optional) Output shifts tensor for quantized computations. In case of per-channel quantization, - * the number of multipliers must be equal to the number of filters (IFM). Supported data types: S32 + * @param[in] input Source tensor info. DataType supported: F16/F32. + * @param[in] weights Weights tensor info. A 3D tensor with dimensions [IFM, 3, 3]. + * Data type supported: Same as @p input. + * @param[in] biases Biases tensor info. A 1D tensor with dimensions [IFM]. Must be nullptr if not needed. + * Data type supported: Same as @p input. + * @param[in] output Destination tensor info. Data type supported: Same as @p input. + * @param[in] conv_info Padding and stride information to use for the convolution. + * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU are supported. + * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). * * @return a status */ static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info, - unsigned int depth_multiplier = 1, ActivationLayerInfo act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U), - const ITensorInfo *output_multipliers = nullptr, const ITensorInfo *output_shifts = nullptr); + unsigned int depth_multiplier = 1, ActivationLayerInfo act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U)); // Inherited methods overridden: void run(const Window &window, cl::CommandQueue &queue) override; - BorderSize border_size() const override; private: + const ICLTensor *_input; + ICLTensor *_output; + const ICLTensor *_weights; + const ICLTensor *_biases; + unsigned int _num_planes_processed_per_iteration; }; } // namespace arm_compute -- cgit v1.2.1