1 /* 2 * Copyright (c) 2022 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_INDIRECT_CONV2D_ADDRESS_PRECALCULATION_KERNEL_H 25 #define ARM_COMPUTE_CL_INDIRECT_CONV2D_ADDRESS_PRECALCULATION_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 // Forward declarations 34 struct DirectConvComputeKernelInfo; 35 36 namespace opencl 37 { 38 namespace kernels 39 { 40 /** Interface for the direct convolution kernel. */ 41 class ClIndirectConv2dAddressPrecalculationKernel : public IClKernel 42 { 43 public: 44 ClIndirectConv2dAddressPrecalculationKernel(); 45 ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClIndirectConv2dAddressPrecalculationKernel); 46 /** Set the src, weights, biases and dst tensors info. 47 * 48 * @note: When M0 is 5,6,7, the kernel rounds up M0 to the nearest power of two. Therefore, eight. The reason behind 49 * this implementation detail is because we can exploit native opencl stores in the kernel. 50 * 51 * @param[in] compile_context The compile context to be used. 52 * @param[in] src The src tensor info to convolve. 3 lower dimensions represent a single src [IFM, width, height], 53 * while every optional dimension from 4 and above represent a batch of inputs. Data types supported: F16/F32. 54 * @param[in] weights Weights tensor info. Weights are 4D tensor with dimensions [IFM, kernel_x, kernel_y, OFM]. 55 * The 1st dimension must be the same as the src's volume 1st dimension. 56 * Data type supported:Same as @p src. 57 * @param[out] dst Output tensor info where to store the precalculated offsets. Data types supported: S32. 58 * The output is a 3D tensor with the following dimensions: [M0 x Kw x Kh, ceil(M/M0), batch-size], where: 59 * Kw=Kernel width, Kh=Kernel height, M0=number of rows processed by each workitem, and M=dst_width x dst_height. 60 * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. 61 * @param[in] desc Direct convolution descriptor used to build the NHWC direct/indirect convolution kernel. 62 */ 63 void configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *weights, ITensorInfo *dst, 64 const PadStrideInfo &conv_info, const DirectConvComputeKernelInfo &desc); 65 /** Static function to check if given info will lead to a valid configuration 66 * 67 * Similar to ClIndirectConv2dAddressPreCalculationKernel::configure() 68 * 69 * @return a status 70 */ 71 static Status validate(const ITensorInfo *src, const ITensorInfo *weights, const ITensorInfo *dst, 72 const PadStrideInfo &conv_info, const DirectConvComputeKernelInfo &desc); 73 74 // Inherited methods overridden: 75 void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override; 76 }; 77 } // namespace kernels 78 } // namespace opencl 79 } // namespace arm_compute 80 #endif /* ARM_COMPUTE_CL_INDIRECT_CONV2D_ADDRESS_PRECALCULATION_KERNEL_H */ 81