ramelg01 | b75d624 | 2021-11-26 19:12:40 +0000 | [diff] [blame] | 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 | #include "arm_compute/graph/nodes/FusedConvolutionBatchNormalizationWithPostOpsNode.h" |
| 25 | |
| 26 | #include "arm_compute/core/Utils.h" |
| 27 | #include "arm_compute/graph/Graph.h" |
| 28 | #include "arm_compute/graph/INodeVisitor.h" |
| 29 | #include "arm_compute/graph/Utils.h" |
| 30 | |
| 31 | namespace arm_compute |
| 32 | { |
| 33 | namespace graph |
| 34 | { |
| 35 | FusedConvolutionBatchNormalizationWithPostOpsNode::FusedConvolutionBatchNormalizationWithPostOpsNode(float epsilon, PadStrideInfo info, |
| 36 | unsigned int num_groups, |
| 37 | ConvolutionMethod method, |
| 38 | FastMathHint fast_math_hint) |
| 39 | : _epsilon(epsilon), _info(std::move(info)), _num_groups(num_groups), _method(method), _fast_math_hint(fast_math_hint) |
| 40 | { |
| 41 | _input_edges.resize(8, EmptyEdgeID); |
| 42 | _outputs.resize(1, NullTensorID); |
| 43 | } |
| 44 | |
| 45 | void FusedConvolutionBatchNormalizationWithPostOpsNode::set_convolution_method(ConvolutionMethod method) |
| 46 | { |
| 47 | _method = method; |
| 48 | } |
| 49 | |
| 50 | float FusedConvolutionBatchNormalizationWithPostOpsNode::epsilon() const |
| 51 | { |
| 52 | return _epsilon; |
| 53 | } |
| 54 | |
| 55 | ConvolutionMethod FusedConvolutionBatchNormalizationWithPostOpsNode::convolution_method() const |
| 56 | { |
| 57 | return _method; |
| 58 | } |
| 59 | |
| 60 | void FusedConvolutionBatchNormalizationWithPostOpsNode::set_fast_math_hint(FastMathHint hint) |
| 61 | { |
| 62 | _fast_math_hint = hint; |
| 63 | } |
| 64 | |
| 65 | FastMathHint FusedConvolutionBatchNormalizationWithPostOpsNode::fast_math_hint() const |
| 66 | { |
| 67 | return _fast_math_hint; |
| 68 | } |
| 69 | |
| 70 | PadStrideInfo FusedConvolutionBatchNormalizationWithPostOpsNode::convolution_info() const |
| 71 | { |
| 72 | return _info; |
| 73 | } |
| 74 | |
| 75 | unsigned int FusedConvolutionBatchNormalizationWithPostOpsNode::num_groups() const |
| 76 | { |
| 77 | return _num_groups; |
| 78 | } |
| 79 | |
| 80 | TensorDescriptor FusedConvolutionBatchNormalizationWithPostOpsNode::compute_output_descriptor(const TensorDescriptor &input_descriptor, |
| 81 | const TensorDescriptor &weights_descriptor, |
| 82 | const PadStrideInfo &info) |
| 83 | { |
| 84 | unsigned int output_width = 0; |
| 85 | unsigned int output_height = 0; |
| 86 | |
| 87 | const unsigned int input_width = get_dimension_size(input_descriptor, DataLayoutDimension::WIDTH); |
| 88 | const unsigned int input_height = get_dimension_size(input_descriptor, DataLayoutDimension::HEIGHT); |
| 89 | const unsigned int kernel_width = get_dimension_size(weights_descriptor, DataLayoutDimension::WIDTH); |
| 90 | const unsigned int kernel_height = get_dimension_size(weights_descriptor, DataLayoutDimension::HEIGHT); |
| 91 | |
| 92 | std::tie(output_width, output_height) = scaled_dimensions(input_width, input_height, kernel_width, kernel_height, info); |
| 93 | |
| 94 | const DataLayout data_layout = input_descriptor.layout; |
| 95 | TensorDescriptor output_descriptor = input_descriptor; |
| 96 | output_descriptor.shape.set(get_dimension_idx(data_layout, DataLayoutDimension::WIDTH), output_width); |
| 97 | output_descriptor.shape.set(get_dimension_idx(data_layout, DataLayoutDimension::HEIGHT), output_height); |
| 98 | output_descriptor.shape.set(get_dimension_idx(data_layout, DataLayoutDimension::CHANNEL), weights_descriptor.shape[3]); |
| 99 | |
| 100 | return output_descriptor; |
| 101 | } |
| 102 | |
| 103 | bool FusedConvolutionBatchNormalizationWithPostOpsNode::forward_descriptors() |
| 104 | { |
| 105 | if((input_id(0) != NullTensorID) && (input_id(1) != NullTensorID) && (output_id(0) != NullTensorID)) |
| 106 | { |
| 107 | Tensor *dst = output(0); |
| 108 | ARM_COMPUTE_ERROR_ON(dst == nullptr); |
| 109 | dst->desc() = configure_output(0); |
| 110 | return true; |
| 111 | } |
| 112 | return false; |
| 113 | } |
| 114 | |
| 115 | TensorDescriptor FusedConvolutionBatchNormalizationWithPostOpsNode::configure_output(size_t idx) const |
| 116 | { |
| 117 | ARM_COMPUTE_UNUSED(idx); |
| 118 | const Tensor *src = input(0); |
| 119 | const Tensor *weights = input(1); |
| 120 | |
| 121 | ARM_COMPUTE_ERROR_ON(src == nullptr || weights == nullptr); |
| 122 | |
| 123 | TensorDescriptor output_info = compute_output_descriptor(src->desc(), weights->desc(), _info); |
| 124 | |
| 125 | return output_info; |
| 126 | } |
| 127 | |
| 128 | NodeType FusedConvolutionBatchNormalizationWithPostOpsNode::type() const |
| 129 | { |
| 130 | return FusedConvolutionBatchNormalizationWithPostOpsNode::node_type; |
| 131 | } |
| 132 | |
| 133 | void FusedConvolutionBatchNormalizationWithPostOpsNode::accept(INodeVisitor &v) |
| 134 | { |
| 135 | v.visit(*this); |
| 136 | } |
| 137 | } // namespace graph |
| 138 | } // namespace arm_compute |