From 3f217ec4ff11e20fe686beb9a28d0bbd80a56cd6 Mon Sep 17 00:00:00 2001 From: Isabella Gottardi Date: Mon, 12 Feb 2018 14:59:19 +0000 Subject: COMPMID-908 - Merge Activation layer with Convolution Layer (NEON. CL, GLES) Change-Id: Iab06d0768ecf805b841e601185608aae88cf9166 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/120874 Tested-by: Jenkins Reviewed-by: Anthony Barbier --- arm_compute/runtime/CL/functions/CLConvolutionLayer.h | 13 +++++++------ arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h | 13 ++++++++++--- arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h | 9 +++++++-- .../runtime/CL/functions/CLWinogradConvolutionLayer.h | 11 +++++++++-- 4 files changed, 33 insertions(+), 13 deletions(-) (limited to 'arm_compute/runtime/CL/functions') diff --git a/arm_compute/runtime/CL/functions/CLConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLConvolutionLayer.h index 5f383b28ab..7df765988f 100644 --- a/arm_compute/runtime/CL/functions/CLConvolutionLayer.h +++ b/arm_compute/runtime/CL/functions/CLConvolutionLayer.h @@ -24,10 +24,9 @@ #ifndef __ARM_COMPUTE_CLCONVOLUTIONLAYER_H__ #define __ARM_COMPUTE_CLCONVOLUTIONLAYER_H__ -#include "arm_compute/runtime/IFunction.h" - #include "arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h" #include "arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h" +#include "arm_compute/runtime/IFunction.h" #include "arm_compute/runtime/IMemoryManager.h" #include @@ -57,9 +56,10 @@ public: * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. * @param[in] weights_info Specifies if the weights tensor has been reshaped with CLWeightsReshapeKernel. Data type supported: Same as @p input. * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). + * @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 WeightsInfo &weights_info = WeightsInfo(), - const Size2D &dilation = Size2D(1U, 1U)); + const Size2D &dilation = Size2D(1U, 1U), const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref CLConvolutionLayer * * @param[in] input Source tensor. 3 lower dimensions represent a single input [width, height, IFM], @@ -72,11 +72,12 @@ public: * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. * @param[in] weights_info Specifies if the weights tensor has been reshaped with CLWeightsReshapeKernel. Data type supported: Same as @p input. * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). + * @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 WeightsInfo &weights_info = WeightsInfo(), const Size2D &dilation = Size2D(1U, 1U)); + const WeightsInfo &weights_info = WeightsInfo(), const Size2D &dilation = Size2D(1U, 1U), const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Static function to check if given info will return the convolution called by @ref CLConvolutionLayer * * @param[in] input Source tensor. 3 lower dimensions represent a single input [width, height, IFM], @@ -88,14 +89,14 @@ public: * Data types supported: Same as @p input. * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. * @param[in] weights_info Specifies if the weights tensor has been reshaped with CLWeightsReshapeKernel. Data type supported: Same as @p input. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. * @param[in] gpu_target Specifies the @p GPUTarget. * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). * * @return a status */ static ConvolutionMethod get_convolution_method(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info, - const WeightsInfo &weights_info, const GPUTarget gpu_target, const Size2D &dilation = Size2D(1U, 1U)); - + const WeightsInfo &weights_info, const ActivationLayerInfo &act_info, const GPUTarget gpu_target, const Size2D &dilation = Size2D(1U, 1U)); // Inherited methods overridden: void run() override; diff --git a/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h index 205c90c478..4363224a08 100644 --- a/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h +++ b/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017, 2018 ARM Limited. + * Copyright (c) 2017-2018 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -27,6 +27,7 @@ #include "arm_compute/core/CL/kernels/CLDirectConvolutionLayerKernel.h" #include "arm_compute/core/CL/kernels/CLFillBorderKernel.h" #include "arm_compute/core/Types.h" +#include "arm_compute/runtime/CL/functions/CLActivationLayer.h" #include "arm_compute/runtime/IFunction.h" #include @@ -53,8 +54,9 @@ public: * @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); + 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 CLDirectConvolutionLayer * * @param[in] input Source tensor. 3 lower dimensions represent a single input [width, height, IFM], @@ -65,10 +67,12 @@ public: * @param[in] 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); + 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; @@ -76,6 +80,9 @@ public: private: CLDirectConvolutionLayerKernel _direct_conv_kernel; CLFillBorderKernel _input_border_handler; + CLActivationLayer _activationlayer_function; + + bool _is_activationlayer_enabled; }; } #endif /* __ARM_COMPUTE_CLDIRECTCONVOLUTIONLAYER_H__ */ diff --git a/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h index 91d35ca0b2..9bfbb536ef 100644 --- a/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h +++ b/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h @@ -36,6 +36,7 @@ #include "arm_compute/core/Types.h" #include "arm_compute/runtime/CL/CLMemoryGroup.h" #include "arm_compute/runtime/CL/CLTensor.h" +#include "arm_compute/runtime/CL/functions/CLActivationLayer.h" #include "arm_compute/runtime/CL/functions/CLGEMM.h" #include "arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h" #include "arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h" @@ -125,9 +126,10 @@ public: * @param[in] weights_info Specifies if the weights tensor has been reshaped with CLWeightsReshapeKernel. If this is not part of the fully connected layer the weights * tensor has also been transposed with CLGEMMTranspose1xWKernel. Data type supported: Same as @p input. * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. */ void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, const WeightsInfo &weights_info = WeightsInfo(), - const Size2D &dilation = Size2D(1U, 1U)); + const Size2D &dilation = Size2D(1U, 1U), const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMConvolutionLayer. * * @param[in] input Source tensor. 3 lower dimensions represent a single input [width, height, IFM], @@ -142,11 +144,12 @@ public: * @param[in] weights_info Specifies if the weights tensor has been reshaped with CLWeightsReshapeKernel. If this is not part of the fully connected layer the weights * tensor has also been transposed with CLGEMMTranspose1xWKernel. Data type supported: Same as @p input. * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). + * @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 WeightsInfo &weights_info = WeightsInfo(), const Size2D &dilation = Size2D(1U, 1U)); + const WeightsInfo &weights_info = WeightsInfo(), const Size2D &dilation = Size2D(1U, 1U), const ActivationLayerInfo &act_info = ActivationLayerInfo()); // Inherited methods overridden: void run() override; @@ -179,6 +182,7 @@ private: CLGEMMLowpMatrixMultiplyCore _mm_gemmlowp; CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint _gemmlowp_output_stage; CLCol2ImKernel _col2im_kernel; + CLActivationLayer _activationlayer_function; const ICLTensor *_original_weights; @@ -189,6 +193,7 @@ private: bool _is_quantized; bool _is_first_run; + bool _is_activationlayer_enabled; }; } #endif /* __ARM_COMPUTE_CLGEMMCONVOLUTIONLAYER_H__ */ diff --git a/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h index 14de169236..2cf1f77fb4 100644 --- a/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h +++ b/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h @@ -27,6 +27,7 @@ #include "arm_compute/core/CL/kernels/CLWinogradFilterTransformKernel.h" #include "arm_compute/core/CL/kernels/CLWinogradOutputTransformKernel.h" #include "arm_compute/core/Types.h" +#include "arm_compute/runtime/CL/functions/CLActivationLayer.h" #include "arm_compute/runtime/CL/functions/CLGEMM.h" #include "arm_compute/runtime/CL/functions/CLWinogradInputTransform.h" #include "arm_compute/runtime/IFunction.h" @@ -60,8 +61,10 @@ public: * @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); + 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 CLWinogradConvolutionLayer * * @note: This function only works with 3x3 kernels and unit strides @@ -74,10 +77,12 @@ public: * @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); + 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; @@ -88,10 +93,12 @@ private: CLWinogradInputTransform _input_transform; CLWinogradFilterTransformKernel _filter_transform; CLWinogradOutputTransformKernel _output_transform; + CLActivationLayer _activationlayer_function; CLTensor _input0; CLTensor _input1; CLTensor _batched_mm_output; bool _is_first_run; + bool _is_activationlayer_enabled; }; } #endif /* __ARM_COMPUTE_CLWINOGRADCONVOLUTIONLAYER_H__ */ -- cgit v1.2.1