From 8cdfdb83c89178b5cf654a5b27471950ab1b997e Mon Sep 17 00:00:00 2001 From: Frank Lei Date: Tue, 2 Jan 2018 16:49:33 +0800 Subject: APPBROWSER-366: Add DepthwiseConvolutionLayer(fp16 only) support. Change-Id: I051b7e56b60bf1a55cdf014539ef71346d3aee26 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/114737 Reviewed-by: Anthony Barbier Tested-by: Jenkins --- arm_compute/core/GLES_COMPUTE/GCKernels.h | 1 + .../kernels/GCDepthwiseConvolutionLayer3x3Kernel.h | 76 ++++++++++++++++++++++ 2 files changed, 77 insertions(+) create mode 100644 arm_compute/core/GLES_COMPUTE/kernels/GCDepthwiseConvolutionLayer3x3Kernel.h (limited to 'arm_compute/core/GLES_COMPUTE') diff --git a/arm_compute/core/GLES_COMPUTE/GCKernels.h b/arm_compute/core/GLES_COMPUTE/GCKernels.h index c6f4877fa1..5be44984b2 100644 --- a/arm_compute/core/GLES_COMPUTE/GCKernels.h +++ b/arm_compute/core/GLES_COMPUTE/GCKernels.h @@ -31,6 +31,7 @@ #include "arm_compute/core/GLES_COMPUTE/kernels/GCBatchNormalizationLayerKernel.h" #include "arm_compute/core/GLES_COMPUTE/kernels/GCCol2ImKernel.h" #include "arm_compute/core/GLES_COMPUTE/kernels/GCDepthConcatenateLayerKernel.h" +#include "arm_compute/core/GLES_COMPUTE/kernels/GCDepthwiseConvolutionLayer3x3Kernel.h" #include "arm_compute/core/GLES_COMPUTE/kernels/GCDirectConvolutionLayerKernel.h" #include "arm_compute/core/GLES_COMPUTE/kernels/GCDropoutLayerKernel.h" #include "arm_compute/core/GLES_COMPUTE/kernels/GCFillBorderKernel.h" diff --git a/arm_compute/core/GLES_COMPUTE/kernels/GCDepthwiseConvolutionLayer3x3Kernel.h b/arm_compute/core/GLES_COMPUTE/kernels/GCDepthwiseConvolutionLayer3x3Kernel.h new file mode 100644 index 0000000000..e10769db5e --- /dev/null +++ b/arm_compute/core/GLES_COMPUTE/kernels/GCDepthwiseConvolutionLayer3x3Kernel.h @@ -0,0 +1,76 @@ +/* + * Copyright (c) 2017 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_GCDEPTHWISECONVOLUTIONKERNEL3x3_H__ +#define __ARM_COMPUTE_GCDEPTHWISECONVOLUTIONKERNEL3x3_H__ + +#include "arm_compute/core/GLES_COMPUTE/IGCKernel.h" + +namespace arm_compute +{ +class IGCTensor; + +/** Interface for the kernel to run a 3x3 depthwise convolution on a tensor. + */ +class GCDepthwiseConvolutionLayer3x3Kernel : public IGCKernel +{ +public: + /** Default constructor */ + GCDepthwiseConvolutionLayer3x3Kernel(); + /** Prevent instances of this class from being copied (As this class contains pointers) */ + GCDepthwiseConvolutionLayer3x3Kernel(const GCDepthwiseConvolutionLayer3x3Kernel &) = delete; + /** Prevent instances of this class from being copied (As this class contains pointers) */ + GCDepthwiseConvolutionLayer3x3Kernel &operator=(const GCDepthwiseConvolutionLayer3x3Kernel &) = delete; + /** Default Move Constructor. */ + GCDepthwiseConvolutionLayer3x3Kernel(GCDepthwiseConvolutionLayer3x3Kernel &&) = default; + /** Default move assignment operator. */ + GCDepthwiseConvolutionLayer3x3Kernel &operator=(GCDepthwiseConvolutionLayer3x3Kernel &&) = default; + /** Initialize the function's source, destination, conv and border_size. + * + * @param[in] input Source tensor. DataType supported: F16. + * @param[in] weights Weights tensor. A 3D tensor with dimensions [3, 3, IFM]. Data type supported: Same as @p input. + * @param[in] biases (Optional) 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. + */ + void configure(const IGCTensor *input, const IGCTensor *weights, const IGCTensor *biases, IGCTensor *output, const PadStrideInfo &conv_info); + + // Inherited methods overridden: + void run(const Window &window) override; + BorderSize border_size() const override; + +private: + BorderSize _border_size; + const IGCTensor *_input; + IGCTensor *_output; + const IGCTensor *_weights; + const IGCTensor *_biases; + unsigned int _conv_stride_x; + unsigned int _conv_stride_y; + unsigned int _conv_pad_left; + unsigned int _conv_pad_top; + gles::NDRange _lws; +}; +} // namespace arm_compute +#endif /*__ARM_COMPUTE_GCDEPTHWISECONVOLUTIONKERNEL3x3_H__ */ -- cgit v1.2.1