1 /* 2 * Copyright (c) 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 25 #pragma once 26 27 #include "arm_gemm.hpp" 28 #include "common.hpp" 29 30 namespace arm_conv 31 { 32 namespace depthwise 33 { 34 using arm_gemm::Nothing; 35 36 enum class DepthwiseMethod 37 { 38 DEFAULT, 39 DEPTHFIRST, 40 PLANAR, 41 }; 42 43 struct KernelDescription 44 { 45 DepthwiseMethod method = DepthwiseMethod::DEFAULT; 46 std::string name = ""; 47 bool is_default = false; 48 uint64_t cycle_estimate = 0; 49 KernelDescriptionarm_conv::depthwise::KernelDescription50 KernelDescription( 51 DepthwiseMethod method, 52 std::string name, 53 bool is_default, 54 uint64_t cycle_estimate) 55 : method(method), name(name), is_default(is_default), cycle_estimate(cycle_estimate) 56 { 57 } 58 KernelDescriptionarm_conv::depthwise::KernelDescription59 KernelDescription() noexcept {}; 60 }; 61 62 class IDepthwiseCommon 63 { 64 public: 65 virtual ~IDepthwiseCommon() = default; 66 67 // Determine the amount of storage space required for the rearranged weights 68 // and bias. 69 virtual size_t get_storage_size(void) const = 0; 70 71 // Rearrange the weights and biases into a storage buffer. 72 // Accepts a pointer to a buffer into which to store the packed parameters, a 73 // pointer the bias vector (which may be nullptr in the case of no bias) and 74 // a pointer to the array of weights (stored in HWIO order). 75 virtual void pack_parameters( 76 void *buffer, 77 const void *biases, 78 const void *weights, 79 size_t ld_weight_col = 0, 80 size_t ld_weight_row = 0) = 0; 81 82 // Determine the amount of working space required 83 virtual size_t get_working_size(unsigned int n_threads, unsigned int n_input_channels) const = 0; 84 85 // Execute the convolution over the specified area of memory. 86 virtual void execute( 87 const void *input, // Pointer to input tensor 88 const void *parameters, // Packed parameters buffer 89 void *output, 90 void *working_space, 91 unsigned int thread_id, 92 unsigned int n_threads) const = 0; 93 94 virtual void execute( 95 const void *input, 96 size_t ld_input_col, 97 size_t ld_input_row, 98 size_t ld_input_batch, 99 const void *parameters, 100 void *output, 101 size_t ld_output_col, 102 size_t ld_output_row, 103 size_t ld_output_batch, 104 void *working_space, 105 unsigned int thread_id, 106 unsigned int n_threads) const = 0; 107 108 virtual void execute( 109 unsigned int batches, 110 unsigned int input_height, 111 unsigned int input_width, 112 unsigned int channels, 113 const PaddingValues &, 114 const void *input, 115 size_t ld_input_col, 116 size_t ld_input_row, 117 size_t ld_input_batch, 118 const void *parameters, 119 unsigned int output_height, 120 unsigned int output_width, 121 void *output, 122 size_t ld_output_col, 123 size_t ld_output_row, 124 size_t ld_output_batch, 125 void *working_space, 126 unsigned int thread_id, 127 unsigned int n_threads) const = 0; 128 }; 129 130 } // namespace depthwise 131 } // namespace arm_conv 132