1 /* 2 * Copyright (c) 2017-2021 Arm Limited. 3 * 4 * SPDX-License-Identifier: MIT 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to 8 * deal in the Software without restriction, including without limitation the 9 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or 10 * sell copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in all 14 * copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 22 * SOFTWARE. 23 */ 24 #ifndef ARM_COMPUTE_CL_WEIGHTSRESHAPE_KERNEL_H 25 #define ARM_COMPUTE_CL_WEIGHTSRESHAPE_KERNEL_H 26 27 #include "src/core/common/Macros.h" 28 #include "src/gpu/cl/ClCompileContext.h" 29 #include "src/gpu/cl/IClKernel.h" 30 31 namespace arm_compute 32 { 33 namespace opencl 34 { 35 namespace kernels 36 { 37 /** OpenCL kernel to perform reshaping on the weights used by convolution and locally connected layer 38 * 39 * Rearranges each 3-dimensional kernel to a single row leading to a matrix with linearized kernels. 40 * In combination with the @ref opencl::kernels::ClIm2ColKernel can transform a convolution to a matrix multiplication. 41 * 42 * For example assuming a 3D weight kernel of 3x3 dimensions and depth of 2 we have: 43 * @f[ 44 * \left( \begin{array}{ccc} 45 * a000 & a001 & a002 \\ 46 * a010 & a011 & a012 \\ 47 * a020 & a021 & a022 \\ 48 * \end{array} \right) 49 * \left( \begin{array}{ccc} 50 * a100 & a101 & a102 \\ 51 * a110 & a111 & a112 \\ 52 * a120 & a121 & a122 \\ 53 * \end{array} \right) 54 * \rightarrow 55 * \left( \begin{array}{ccccccccc} 56 * a000 & a001 & a002 & a010 & a011 & a012 & a020 & a021 & a022 & a100 & a101 & a102 & a110 & a111 & a112 & a120 & a121 & a122 \\ 57 * \end{array} \right) 58 * @f] 59 */ 60 class ClWeightsReshapeKernel : public IClKernel 61 { 62 public: 63 ClWeightsReshapeKernel(); 64 ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClWeightsReshapeKernel); 65 /** Set the input and output of the kernel. 66 * 67 * @param[in] compile_context The compile context to be used. 68 * @param[in] src The input tensor info to convert. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM] if shared, 69 * and 5D tensor with dimensions [kernel_x, kernel_y, IFM, OFM, num_patches] if unshared. Data types supported: All 70 * @param[in] biases The shared biases tensor info to append. Bias is 1D tensor with dimensions [OFM] if shared and 2D tensor with 71 * dimensions [OFM, num_patches] if unshared. Data types supported: F16/F32, for quantized types this must be nullptr. 72 * @warning Appending biases to weights reshaped matrix is not supported for quantized asymmetric types. 73 * @param[out] dst The output tensor info. Should be a 2D Tensor if there are no groups and the weights are not shared; a 3D Tensor otherwise. 74 * Data types supported: Same as @p input 75 * @param[in] num_groups (Optional) Number of groups when performing a grouped convolution. num_groups != 1 is only supported for NCHW data layout 76 * Number of groups greater than one are only supported for NCHW data layout, and the number of weights must be a multiple of it. 77 */ 78 void configure(const ClCompileContext &compile_context, const ITensorInfo *src, const ITensorInfo *biases, ITensorInfo *dst, unsigned int num_groups = 1); 79 /** Static function to check if given info will lead to a valid configuration 80 * 81 * Similar to ClWeightsReshapeKernel::configure() 82 * 83 * @return a status 84 */ 85 static Status validate(const ITensorInfo *src, const ITensorInfo *biases, const ITensorInfo *dst, unsigned int num_groups = 1); 86 87 // Inherited methods overridden: 88 void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override; 89 }; 90 } // namespace kernels 91 } // namespace opencl 92 } // namespace arm_compute 93 #endif /*ARM_COMPUTE_CL_WEIGHTSRESHAPE_KERNEL_H */