1 /* 2 * Copyright (c) 2018-2021 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_CLRNN_LAYER_H 25 #define ARM_COMPUTE_CLRNN_LAYER_H 26 27 #include "arm_compute/runtime/CL/ICLSimpleFunction.h" 28 #include "arm_compute/runtime/CL/functions/CLActivationLayer.h" 29 #include "arm_compute/runtime/CL/functions/CLCopy.h" 30 #include "arm_compute/runtime/CL/functions/CLElementwiseOperations.h" 31 #include "arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h" 32 #include "arm_compute/runtime/CL/functions/CLGEMM.h" 33 34 #include <memory> 35 36 namespace arm_compute 37 { 38 class ICLTensor; 39 40 /** Basic function to run @ref CLRNNLayer */ 41 class CLRNNLayer : public IFunction 42 { 43 public: 44 /** Default constructor */ 45 CLRNNLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr); 46 /** Prevent instances of this class from being copied */ 47 CLRNNLayer(const CLRNNLayer &) = delete; 48 /** Prevent instances of this class from being copied */ 49 CLRNNLayer &operator=(const CLRNNLayer &) = delete; 50 /** Default destructor */ 51 ~CLRNNLayer(); 52 /** Initialize the function 53 * 54 * Valid data layouts: 55 * - NHWC 56 * - NCHW 57 * 58 * Valid data type configurations: 59 * |src0 |src1 |src2 |src3 |dst0 |dst1 | 60 * |:------|:------|:------|:------|:------|:------| 61 * |F16 |F16 |F16 |F16 |F16 |F16 | 62 * |F32 |F32 |F32 |F32 |F32 |F32 | 63 * 64 * @param[in] input Input is a 2-D tensor of shape [input_size, batch_size]. Data types supported: F16/F32 65 * @param[in] weights Weights tensor of shape [input_size, num_units] that multiplies the input. Data types supported: Same as @p input 66 * @param[in] recurrent_weights Weights tensor of shape [num_units, num_units] that multiplies the current 'state'. Data types supported: Same as @p input 67 * @param[in] bias Bias vector of shape [num_units]. Data types supported: Same as @p input 68 * @param[out] output Output tensor of shape [num_units, batch_size]. Data types supported: Same as @p input 69 * @param[in,out] hidden_state Output tensor of shape [num_units, batch_size]. Data types supported: Same as @p input 70 * @param[in] info Activation layer parameter. 71 */ 72 void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *recurrent_weights, const ICLTensor *bias, ICLTensor *hidden_state, ICLTensor *output, ActivationLayerInfo &info); 73 /** Initialize the function 74 * 75 * @param[in] compile_context The compile context to be used. 76 * @param[in] input Input is a 2-D tensor of shape [input_size, batch_size]. Data types supported: F16/F32 77 * @param[in] weights Weights tensor of shape [input_size, num_units] that multiplies the input. Data types supported: Same as @p input 78 * @param[in] recurrent_weights Weights tensor of shape [num_units, num_units] that multiplies the current 'state'. Data types supported: Same as @p input 79 * @param[in] bias Bias vector of shape [num_units]. Data types supported: Same as @p input 80 * @param[out] output Output tensor of shape [num_units, batch_size]. Data types supported: Same as @p input 81 * @param[in,out] hidden_state Output tensor of shape [num_units, batch_size]. Data types supported: Same as @p input 82 * @param[in] info Activation layer parameter. 83 */ 84 void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *weights, const ICLTensor *recurrent_weights, const ICLTensor *bias, ICLTensor *hidden_state, 85 ICLTensor *output, ActivationLayerInfo &info); 86 /** Initialize the function 87 * 88 * @param[in] input Input is a 2-D tensor of shape [input_size, batch_size]. Data types supported: F16/F32 89 * @param[in] weights Weights tensor of shape [input_size, num_units] that multiplies the input. Data types supported: Same as @p input 90 * @param[in] recurrent_weights Weights tensor of shape [num_units, num_units] that multiplies the current 'state'. Data types supported: Same as @p input 91 * @param[in] bias Bias vector of shape [num_units]. Data types supported: Same as @p input 92 * @param[in] output Output tensor of shape [num_units, batch_size]. Data types supported: Same as @p input 93 * @param[in] hidden_state Output tensor of shape [num_units, batch_size]. Data types supported: Same as @p input 94 * @param[in] info Activation layer parameter. 95 * 96 * @return a status 97 */ 98 static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *recurrent_weights, const ITensorInfo *bias, const ITensorInfo *hidden_state, const ITensorInfo *output, 99 const ActivationLayerInfo &info); 100 101 // Inherited methods overridden: 102 void run() override; 103 void prepare() override; 104 105 private: 106 MemoryGroup _memory_group; 107 CLGEMM _gemm_state_f; 108 CLArithmeticAddition _add_kernel; 109 CLActivationLayer _activation; 110 CLFullyConnectedLayer _fully_connected_kernel; 111 CLCopy _copy; 112 CLTensor _fully_connected_out; 113 CLTensor _gemm_output; 114 CLTensor _add_output; 115 bool _is_prepared; 116 }; 117 } 118 #endif /* ARM_COMPUTE_CLRNN_LAYER_H */ 119