Georgios Pinitas | c936917 | 2018-09-26 11:25:40 +0100 | [diff] [blame] | 1 | /* |
Manuel Bottini | 4c6bd51 | 2020-04-08 10:15:51 +0100 | [diff] [blame] | 2 | * Copyright (c) 2018-2020 ARM Limited. |
Georgios Pinitas | c936917 | 2018-09-26 11:25:40 +0100 | [diff] [blame] | 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 | */ |
Michalis Spyrou | f464337 | 2019-11-29 16:17:13 +0000 | [diff] [blame] | 24 | #ifndef ARM_COMPUTE_CLFUSEBATCHNORMALIZATIONKERNEL_H |
| 25 | #define ARM_COMPUTE_CLFUSEBATCHNORMALIZATIONKERNEL_H |
Georgios Pinitas | c936917 | 2018-09-26 11:25:40 +0100 | [diff] [blame] | 26 | |
| 27 | #include "arm_compute/core/CL/ICLKernel.h" |
| 28 | |
| 29 | namespace arm_compute |
| 30 | { |
| 31 | // Forward declarations |
| 32 | class ICLTensor; |
| 33 | |
| 34 | /** OpenCL kernel to fuse the batch normalization node to a preceding convolution node */ |
| 35 | class CLFuseBatchNormalizationKernel : public ICLKernel |
| 36 | { |
| 37 | public: |
| 38 | /** Default constructor */ |
| 39 | CLFuseBatchNormalizationKernel(); |
| 40 | /** Prevent instances of this class from being copied (As this class contains pointers) */ |
| 41 | CLFuseBatchNormalizationKernel(const CLFuseBatchNormalizationKernel &) = delete; |
| 42 | /** Prevent instances of this class from being copied (As this class contains pointers) */ |
| 43 | CLFuseBatchNormalizationKernel &operator=(const CLFuseBatchNormalizationKernel &) = delete; |
| 44 | /** Allow instances of this class to be moved */ |
| 45 | CLFuseBatchNormalizationKernel(CLFuseBatchNormalizationKernel &&) = default; |
| 46 | /** Allow instances of this class to be moved */ |
| 47 | CLFuseBatchNormalizationKernel &operator=(CLFuseBatchNormalizationKernel &&) = default; |
| 48 | /** Default destructor */ |
| 49 | ~CLFuseBatchNormalizationKernel() = default; |
| 50 | /** Set the source, destination of the kernel |
| 51 | * |
Manuel Bottini | 2732cca | 2019-05-28 11:44:41 +0100 | [diff] [blame] | 52 | * @param[in] input_weights Input weights tensor for convolution or depthwise convolution layer. Data type supported: F16/F32. Data layout supported: NCHW, NHWC |
| 53 | * @param[in] bn_mean Batch normalization layer mean tensor. Same as @p input_weights |
| 54 | * @param[in] bn_var Batch normalization layer variance tensor. Same as @p input_weights |
| 55 | * @param[out] fused_weights Output fused weights tensor. It can be a nullptr in case of in-place computation. Same as @p input_weights |
| 56 | * @param[out] fused_bias Output fused bias tensor. It can be a nullptr in case of in-place computation and input_bias != nullptr. Same as @p input_weights |
| 57 | * @param[in] input_bias (Optional) Input bias tensor for convolution or depthwise convolution layer. It can be a nullptr in case the bias tensor is not required. Same as @p input_weights |
| 58 | * @param[in] bn_beta (Optional) Batch normalization layer beta tensor. It can be a nullptr in case the beta tensor is not required. Same as @p input_weights |
Gian Marco Iodice | 761c8d0 | 2019-06-10 14:46:49 +0100 | [diff] [blame] | 59 | * @note if nullptr, bn_beta is set to 0.0 |
Manuel Bottini | 2732cca | 2019-05-28 11:44:41 +0100 | [diff] [blame] | 60 | * @param[in] bn_gamma (Optional) Batch normalization layer gamma tensor. It can be a nullptr in case the gamma tensor is not required. Same as @p input_weights |
Gian Marco Iodice | 761c8d0 | 2019-06-10 14:46:49 +0100 | [diff] [blame] | 61 | * @note if nullptr, bn_gamma is set to 1.0 |
Georgios Pinitas | c936917 | 2018-09-26 11:25:40 +0100 | [diff] [blame] | 62 | * @param[in] epsilon (Optional) Batch normalization layer epsilon parameter. Defaults to 0.001f. |
Manuel Bottini | 2732cca | 2019-05-28 11:44:41 +0100 | [diff] [blame] | 63 | * @param[in] fbn_type (Optional) Fused batch normalization type. Defaults to CONVOLUTION. |
Georgios Pinitas | c936917 | 2018-09-26 11:25:40 +0100 | [diff] [blame] | 64 | */ |
Manuel Bottini | 2732cca | 2019-05-28 11:44:41 +0100 | [diff] [blame] | 65 | void configure(const ICLTensor *input_weights, const ICLTensor *bn_mean, const ICLTensor *bn_var, ICLTensor *fused_weights, ICLTensor *fused_bias, |
| 66 | const ICLTensor *input_bias = nullptr, const ICLTensor *bn_beta = nullptr, const ICLTensor *bn_gamma = nullptr, |
| 67 | float epsilon = 0.001f, FuseBatchNormalizationType fbn_type = FuseBatchNormalizationType::CONVOLUTION); |
Manuel Bottini | 4c6bd51 | 2020-04-08 10:15:51 +0100 | [diff] [blame] | 68 | /** Set the source, destination of the kernel |
| 69 | * |
| 70 | * @param[in] compile_context The compile context to be used. |
| 71 | * @param[in] input_weights Input weights tensor for convolution or depthwise convolution layer. Data type supported: F16/F32. Data layout supported: NCHW, NHWC |
| 72 | * @param[in] bn_mean Batch normalization layer mean tensor. Same as @p input_weights |
| 73 | * @param[in] bn_var Batch normalization layer variance tensor. Same as @p input_weights |
| 74 | * @param[out] fused_weights Output fused weights tensor. It can be a nullptr in case of in-place computation. Same as @p input_weights |
| 75 | * @param[out] fused_bias Output fused bias tensor. It can be a nullptr in case of in-place computation and input_bias != nullptr. Same as @p input_weights |
| 76 | * @param[in] input_bias (Optional) Input bias tensor for convolution or depthwise convolution layer. It can be a nullptr in case the bias tensor is not required. Same as @p input_weights |
| 77 | * @param[in] bn_beta (Optional) Batch normalization layer beta tensor. It can be a nullptr in case the beta tensor is not required. Same as @p input_weights |
| 78 | * @note if nullptr, bn_beta is set to 0.0 |
| 79 | * @param[in] bn_gamma (Optional) Batch normalization layer gamma tensor. It can be a nullptr in case the gamma tensor is not required. Same as @p input_weights |
| 80 | * @note if nullptr, bn_gamma is set to 1.0 |
| 81 | * @param[in] epsilon (Optional) Batch normalization layer epsilon parameter. Defaults to 0.001f. |
| 82 | * @param[in] fbn_type (Optional) Fused batch normalization type. Defaults to CONVOLUTION. |
| 83 | */ |
| 84 | void configure(CLCompileContext &compile_context, const ICLTensor *input_weights, const ICLTensor *bn_mean, const ICLTensor *bn_var, ICLTensor *fused_weights, ICLTensor *fused_bias, |
| 85 | const ICLTensor *input_bias = nullptr, const ICLTensor *bn_beta = nullptr, const ICLTensor *bn_gamma = nullptr, |
| 86 | float epsilon = 0.001f, FuseBatchNormalizationType fbn_type = FuseBatchNormalizationType::CONVOLUTION); |
Georgios Pinitas | c936917 | 2018-09-26 11:25:40 +0100 | [diff] [blame] | 87 | /** Static function to check if given info will lead to a valid configuration of @ref CLFuseBatchNormalizationKernel |
| 88 | * |
Manuel Bottini | 2732cca | 2019-05-28 11:44:41 +0100 | [diff] [blame] | 89 | * @param[in] input_weights Input weights tensor info for convolution or depthwise convolution layer. Data type supported: F16/F32. Data layout supported: NCHW, NHWC |
| 90 | * @param[in] bn_mean Batch normalization layer mean tensor info. Same as @p input_weights |
| 91 | * @param[in] bn_var Batch normalization layer variance tensor info. Same as @p input_weights |
| 92 | * @param[in] fused_weights Output fused weights tensor info. It can be a nullptr in case of in-place computation. Same as @p input_weights |
| 93 | * @param[in] fused_bias Output fused bias tensor info. It can be a nullptr in case of in-place computation and input_bias != nullptr. Same as @p input_weights |
| 94 | * @param[in] input_bias (Optional) Input bias tensor info for convolution or depthwise convolution layer. It can be a nullptr in case the bias tensor is not required. Same as @p input_weights |
| 95 | * @param[in] bn_beta (Optional) Batch normalization layer beta tensor info. It can be a nullptr in case the beta tensor is not required. Same as @p input_weights |
| 96 | * @note if nullptr, bn_beta is set to 0.0 |
| 97 | * @param[in] bn_gamma (Optional) Batch normalization layer gamma tensor info. It can be a nullptr in case the gamma tensor is not required. Same as @p input_weights |
| 98 | * @note if nullptr, bn_gamma is set to 1.0 |
| 99 | * @param[in] epsilon (Optional) Batch normalization layer epsilon parameter. Defaults to 0.001f. |
| 100 | * @param[in] fbn_type (Optional) Fused batch normalization type. Defaults to CONVOLUTION. |
Georgios Pinitas | c936917 | 2018-09-26 11:25:40 +0100 | [diff] [blame] | 101 | * |
| 102 | * @return a status |
| 103 | */ |
Manuel Bottini | 2732cca | 2019-05-28 11:44:41 +0100 | [diff] [blame] | 104 | static Status validate(const ITensorInfo *input_weights, const ITensorInfo *bn_mean, const ITensorInfo *bn_var, |
Georgios Pinitas | c936917 | 2018-09-26 11:25:40 +0100 | [diff] [blame] | 105 | const ITensorInfo *fused_weights, const ITensorInfo *fused_bias, |
Manuel Bottini | 2732cca | 2019-05-28 11:44:41 +0100 | [diff] [blame] | 106 | const ITensorInfo *input_bias = nullptr, const ITensorInfo *bn_beta = nullptr, const ITensorInfo *bn_gamma = nullptr, |
| 107 | float epsilon = 0.001f, FuseBatchNormalizationType fbn_type = FuseBatchNormalizationType::CONVOLUTION); |
Georgios Pinitas | c936917 | 2018-09-26 11:25:40 +0100 | [diff] [blame] | 108 | |
| 109 | // Inherited methods overridden: |
| 110 | void run(const Window &window, cl::CommandQueue &queue) override; |
| 111 | |
| 112 | private: |
Michalis Spyrou | f464337 | 2019-11-29 16:17:13 +0000 | [diff] [blame] | 113 | const ICLTensor *_input_weights; |
| 114 | const ICLTensor *_input_bias; |
| 115 | const ICLTensor *_bn_mean; |
| 116 | const ICLTensor *_bn_var; |
| 117 | const ICLTensor *_bn_gamma; |
| 118 | const ICLTensor *_bn_beta; |
| 119 | ICLTensor *_fused_weights; |
| 120 | ICLTensor *_fused_bias; |
| 121 | float _epsilon; |
| 122 | bool _run_in_place_weights; |
| 123 | bool _run_in_place_bias; |
Georgios Pinitas | c936917 | 2018-09-26 11:25:40 +0100 | [diff] [blame] | 124 | }; |
| 125 | } // namespace arm_compute |
Michalis Spyrou | f464337 | 2019-11-29 16:17:13 +0000 | [diff] [blame] | 126 | #endif /*ARM_COMPUTE_CLFUSEBATCHNORMALIZATIONKERNEL_H */ |