/* * Copyright (c) 2018-2021, 2023 Arm Limited. * * SPDX-License-Identifier: MIT * * Permission is hereby granted, free of charge, to any person obtaining a copy * of this software and associated documentation files (the "Software"), to * deal in the Software without restriction, including without limitation the * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or * sell copies of the Software, and to permit persons to whom the Software is * furnished to do so, subject to the following conditions: * * The above copyright notice and this permission notice shall be included in all * copies or substantial portions of the Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. */ #include "arm_compute/graph/mutators/NodeFusionMutator.h" #include "arm_compute/core/utils/DataTypeUtils.h" #include "arm_compute/graph/backends/BackendRegistry.h" #include "arm_compute/graph/GraphBuilder.h" #include "arm_compute/graph/Logger.h" #include "arm_compute/graph/nodes/FusedConvolutionBatchNormalizationNode.h" #include "arm_compute/graph/nodes/Nodes.h" #include "arm_compute/graph/Utils.h" #include "src/graph/mutators/MutatorUtils.h" #include "support/Cast.h" #include #include namespace arm_compute { namespace graph { namespace detail { void transfer_driving_nodes_and_remove_old_node(Graph &g, INode *new_node, INode *old_node, bool add_output_tensor) { if (new_node == nullptr || old_node == nullptr) { return; } // Get driving nodes of last fusable node std::vector last_driving_nodes = get_driving_nodes(*old_node); // Extract last fusable node accessor if any if (old_node->output(0) == nullptr) { return; } auto old_node_accessor = old_node->output(0)->extract_accessor(); // Remove node g.remove_node(old_node->id()); // Update fused node outputs for (auto &driving_node : last_driving_nodes) { g.add_connection(new_node->id(), 0, driving_node.node_id, driving_node.index); if (add_output_tensor) { configure_tensor(new_node->output(0)); } } // Update accessor to fused node new_node->output(0)->set_accessor(std::move(old_node_accessor)); } void fuse_convolution_with_batch_normalization(Graph &g, const Edge *output_edge) { ARM_COMPUTE_ERROR_ON(output_edge == nullptr); auto *conv_node = arm_compute::utils::cast::polymorphic_downcast(output_edge->producer()); auto *bn_node = arm_compute::utils::cast::polymorphic_downcast(output_edge->consumer()); // Not fusing if number of groups is greater than 1 if (conv_node->num_groups() > 1) { return; } ARM_COMPUTE_LOG_GRAPH_VERBOSE("Fusing convolution node with ID : " << output_edge->producer_id() << " with BatchNormalization Layer node with ID : " << output_edge->consumer_id() << std::endl); // Prevent fusion if fused node has an output accessor if (conv_node->output(0)->accessor() == nullptr) { const Target assigned_target = conv_node->assigned_target(); // Extract conv inputs const auto conv_input_id = conv_node->input_edge(0)->producer_id(); const auto conv_weights_id = conv_node->input_edge(1)->producer_id(); const auto conv_info = conv_node->convolution_info(); const auto conv_method = conv_node->convolution_method(); const auto num_groups = conv_node->num_groups(); const auto act_info = bn_node->fused_activation(); FastMathHint fast_math_hint = conv_node->fast_math_hint(); // Extract bn inputs const auto bn_mean_id = bn_node->input_edge(1)->producer_id(); const auto bn_var_id = bn_node->input_edge(2)->producer_id(); const auto epsilon = bn_node->epsilon(); // Create the fused node const NodeID fused_id = g.add_node( epsilon, conv_info, num_groups, conv_method, fast_math_hint, act_info); if (conv_node->input_edge(2) != nullptr) { auto conv_bias_id = conv_node->input_edge(2)->producer_id(); g.add_connection(conv_bias_id, 0, fused_id, 2); } // Add connections from the conv/batch_norm inputs to the fused node g.add_connection(conv_input_id, 0, fused_id, 0); g.add_connection(conv_weights_id, 0, fused_id, 1); g.add_connection(bn_mean_id, 0, fused_id, 3); g.add_connection(bn_var_id, 0, fused_id, 4); if (bn_node->input_edge(3) != nullptr) { const auto bn_beta_id = bn_node->input_edge(3)->producer_id(); g.add_connection(bn_beta_id, 0, fused_id, 5); } if (bn_node->input_edge(4) != nullptr) { const auto bn_gamma_id = bn_node->input_edge(4)->producer_id(); g.add_connection(bn_gamma_id, 0, fused_id, 6); } auto fused_node = g.node(fused_id); auto bn_node_name = bn_node->name(); transfer_driving_nodes_and_remove_old_node(g, fused_node, bn_node, true); fused_node->set_assigned_target(assigned_target); fused_node->set_common_node_parameters(NodeParams{conv_node->name() + "+" + bn_node_name, assigned_target}); // Remove convolution node g.remove_node(conv_node->id()); } else { ARM_COMPUTE_LOG_GRAPH_VERBOSE( "Prevented fusion of convolution with batch normalization due to the presence of an output accessor\n"); } } void fuse_depthwise_convolution_with_batch_normalization(Graph &g, const Edge *output_edge) { ARM_COMPUTE_ERROR_ON(output_edge == nullptr); auto *depth_conv_node = arm_compute::utils::cast::polymorphic_downcast(output_edge->producer()); auto *bn_node = arm_compute::utils::cast::polymorphic_downcast(output_edge->consumer()); ARM_COMPUTE_LOG_GRAPH_VERBOSE("Fusing depthwise convolution node with ID : " << output_edge->producer_id() << " with BatchNormalization Layer node with ID : " << output_edge->consumer_id() << std::endl); // Prevent fusion if fused node has an output accessor if (depth_conv_node->output(0)->accessor() == nullptr) { const Target assigned_target = depth_conv_node->assigned_target(); // Extract conv inputs const auto depth_conv_input_id = depth_conv_node->input_edge(0)->producer_id(); const auto conv_weights_id = depth_conv_node->input_edge(1)->producer_id(); const auto conv_info = depth_conv_node->convolution_info(); const auto depth_conv_method = depth_conv_node->depthwise_convolution_method(); const auto depth_multiplier = depth_conv_node->depth_multiplier(); const auto act_info = bn_node->fused_activation(); // Extract bn inputs const auto bn_mean_id = bn_node->input_edge(1)->producer_id(); const auto bn_var_id = bn_node->input_edge(2)->producer_id(); const auto bn_beta_id = bn_node->input_edge(3)->producer_id(); const auto bn_gamma_id = bn_node->input_edge(4)->producer_id(); const auto epsilon = bn_node->epsilon(); // Create the fused node const NodeID fused_id = g.add_node( epsilon, conv_info, depth_multiplier, depth_conv_method, act_info); if (depth_conv_node->input_edge(2) != nullptr) { const auto conv_bias_id = depth_conv_node->input_edge(2)->producer_id(); g.add_connection(conv_bias_id, 0, fused_id, 2); } // Add connections from the conv/batch_norm inputs to the fused node g.add_connection(depth_conv_input_id, 0, fused_id, 0); g.add_connection(conv_weights_id, 0, fused_id, 1); g.add_connection(bn_mean_id, 0, fused_id, 3); g.add_connection(bn_var_id, 0, fused_id, 4); g.add_connection(bn_beta_id, 0, fused_id, 5); g.add_connection(bn_gamma_id, 0, fused_id, 6); auto fused_node = g.node(fused_id); auto bn_node_name = bn_node->name(); transfer_driving_nodes_and_remove_old_node(g, fused_node, bn_node, true); fused_node->set_assigned_target(assigned_target); fused_node->set_common_node_parameters( NodeParams{depth_conv_node->name() + "+" + bn_node_name, assigned_target}); // Remove convolution node g.remove_node(depth_conv_node->id()); } else { ARM_COMPUTE_LOG_GRAPH_VERBOSE("Prevented fusion of depthwise convolution with batch normalization due to the " "presence of an output accessor\n"); } } template void fuse_node_with_activation(Graph &g, const Edge *output_edge, const std::set &supported_fused_activations) { ARM_COMPUTE_ERROR_ON(output_edge == nullptr); auto *n_node = arm_compute::utils::cast::polymorphic_downcast(output_edge->producer()); auto *act_node = arm_compute::utils::cast::polymorphic_downcast(output_edge->consumer()); ARM_COMPUTE_ERROR_ON(act_node->output(0) == nullptr || n_node->output(0) == nullptr); // Check if activation is supported for fusion if (supported_fused_activations.count(act_node->activation_info().activation()) == 0) { return; } // EltwiseLayerNode can only be fused when dataype is float if (n_node->type() == NodeType::EltwiseLayer && !is_data_type_float(n_node->output(0)->desc().data_type)) { return; } ARM_COMPUTE_LOG_GRAPH_VERBOSE("Fusing node with ID : " << output_edge->producer_id() << " with Activation Layer node with ID : " << output_edge->consumer_id() << std::endl); // Prevent fusion if fused node has an output accessor if (n_node->output(0)->accessor() == nullptr) { // Set activation info to fused node n_node->set_fused_activation(act_node->activation_info()); transfer_driving_nodes_and_remove_old_node(g, n_node, act_node, false); } else { ARM_COMPUTE_LOG_GRAPH_VERBOSE( "Prevented fusion of node with activation due to the presence of an output accessor\n"); } } template void fuse_pad_with_convolution(Graph &g, const Edge *output_edge) { auto *pad_node = arm_compute::utils::cast::polymorphic_downcast(output_edge->producer()); auto *conv_node = arm_compute::utils::cast::polymorphic_downcast(output_edge->consumer()); const Edge *input_edge = pad_node->input_edge(0); if (input_edge != nullptr && input_edge->tensor() != nullptr && pad_node->output(0)->accessor() == nullptr && pad_node->pad_value().get() == 0.0) { const DataLayout layout = input_edge->tensor()->desc().layout; const PaddingList padding_list = pad_node->padding(); const unsigned int height_index = get_dimension_idx(layout, DataLayoutDimension::HEIGHT); const unsigned int width_index = get_dimension_idx(layout, DataLayoutDimension::WIDTH); const PaddingInfo pad_w = width_index < padding_list.size() ? padding_list[width_index] : PaddingInfo(0, 0); const PaddingInfo pad_h = height_index < padding_list.size() ? padding_list[height_index] : PaddingInfo(0, 0); if (is_padding_in_height_or_width(layout, padding_list)) { // Add paddings to the convolution node const PadStrideInfo conv_info = conv_node->convolution_info(); const PadStrideInfo new_conv_info(conv_info.stride().first, conv_info.stride().second, conv_info.pad_left() + pad_w.first, conv_info.pad_right() + pad_w.second, conv_info.pad_top() + pad_h.first, conv_info.pad_bottom() + pad_h.second, conv_info.round()); conv_node->set_convolution_info(new_conv_info); // Update drivers of the convolution node std::vector pad_driver_nodes = get_driver_nodes(*pad_node); g.remove_node(pad_node->id()); // Update fused node inputs for (auto &driver_node : pad_driver_nodes) { g.add_connection(driver_node.node_id, driver_node.index, conv_node->id(), 0); } } } } template void fuse_layer(Graph &g, std::function const &prec, const F fuse_fcn, Args &&...optional_arguments) { // Note that fused nodes may be added to the end of the node list. // Instead of only looping over the original list of nodes, we loop over the current node list which could be growing. // This is intentional as it probes the newly added fused nodes for further fusing opportunities. for (unsigned int i = 0; i < g.nodes().size(); ++i) { auto node = g.node(i); // Check if the node is of type N1 and not a branching node if (node && node->type() == N1::node_type && node->output_edges().size() == 1) { const auto output_edge_id = *node->output_edges().begin(); const auto output_edge = g.edge(output_edge_id); // Check if following node is a type N2 node if ((output_edge != nullptr) && (output_edge->consumer() != nullptr) && (output_edge->consumer()->type() == N2::node_type) && prec(*output_edge->producer())) { fuse_fcn(g, output_edge, optional_arguments...); } } } } template void fuse_layer(Graph &g, std::function const &prec, const F fuse_fcn, Args &&...optional_arguments) { // Note that fused nodes may be added to the end of the node list. // Instead of only looping over the original list of nodes, we loop over the current node list which could be growing. // This is intentional as it probes the newly added fused nodes for further fusing opportunities. for (unsigned int i = 0; i < g.nodes().size(); ++i) { auto node = g.node(i); // Check if the node is of type N1 and not a branching node if (node && node->type() == N1::node_type && node->output_edges().size() == 1) { const auto output_edge_id = *node->output_edges().begin(); const auto output_edge = g.edge(output_edge_id); // Check if it's the correct target if ((output_edge != nullptr) && (output_edge->consumer() != nullptr) && prec(*output_edge->producer())) { fuse_fcn(g, output_edge, i, optional_arguments...); } } } } } // namespace detail const char *NodeFusionMutator::name() { return "NodeFusionMutator"; } IGraphMutator::MutationType NodeFusionMutator::type() const { return IGraphMutator::MutationType::Backend; } void NodeFusionMutator::mutate(Graph &g) { // Supported activations when fusing const std::set supported_fused_activations = { Activation::ABS, Activation::BOUNDED_RELU, Activation::ELU, Activation::HARD_SWISH, Activation::IDENTITY, Activation::LEAKY_RELU, Activation::LINEAR, Activation::LOGISTIC, Activation::LU_BOUNDED_RELU, Activation::RELU, Activation::SOFT_RELU, Activation::SQRT, Activation::SQUARE, Activation::TANH}; // Preconditions auto empty_prec = [](INode &) { return true; }; auto cl_target_prec = [](INode &n) { return n.assigned_target() == Target::CL; }; auto qs8_prec = [&g](INode &n) { ARM_COMPUTE_ERROR_ON(n.output(0) == nullptr); const auto output_edge_id = *n.output_edges().begin(); const auto output_edge = g.edge(output_edge_id); // To perform fusion the two nodes must have same output quantization information const bool same_qinfo = n.output(0)->desc().quant_info == output_edge->producer()->output(0)->desc().quant_info; const bool output_qasymm8 = n.output(0)->desc().data_type == DataType::QASYMM8; return (output_qasymm8 && same_qinfo) || !output_qasymm8; }; // Fusion mutations detail::fuse_layer(g, empty_prec, detail::fuse_pad_with_convolution); detail::fuse_layer( g, empty_prec, detail::fuse_pad_with_convolution); detail::fuse_layer( g, empty_prec, detail::fuse_node_with_activation, supported_fused_activations); detail::fuse_layer( g, empty_prec, detail::fuse_node_with_activation, supported_fused_activations); detail::fuse_layer( g, qs8_prec, detail::fuse_node_with_activation, supported_fused_activations); detail::fuse_layer( g, empty_prec, detail::fuse_node_with_activation, supported_fused_activations); detail::fuse_layer( g, cl_target_prec, detail::fuse_node_with_activation, supported_fused_activations); // The fusion of BatchNormalizationLayer must occur after the fusion of ActivationLayer. Because FusedConvolutionBatchNormalizationNode assumes the BatchNormalization is already fused with activation, if any detail::fuse_layer( g, empty_prec, detail::fuse_convolution_with_batch_normalization); detail::fuse_layer( g, empty_prec, detail::fuse_depthwise_convolution_with_batch_normalization); } } // namespace graph } // namespace arm_compute