commit 2a2db590fd179dcb8e1a575293cd2b887e2dc246
author Georgios Pinitas <georgios.pinitas@arm.com> Wed Aug 15 12:14:46 2018 +0100
committer Anthony Barbier <anthony.barbier@arm.com> Fri Nov 02 16:54:54 2018 +0000
tree 5e10da7cb6777f3020b84a2389b279ceef2be5ee
parent c1961b51df2e15a01a5950139e81bbd47fbfa627

COMPMID-1505: Add native grouping support at graph level

Change-Id: Iedc91b0aee743b59af5140c8acb8124548da3163
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/144362
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Reviewed-by: Michele DiGiorgio <michele.digiorgio@arm.com>

src/graph/mutators/DepthConcatSubTensorMutator.cpp

diff --git a/src/graph/mutators/DepthConcatSubTensorMutator.cpp b/src/graph/mutators/DepthConcatSubTensorMutator.cpp
index 241c07b..937528d 100644
--- a/src/graph/mutators/DepthConcatSubTensorMutator.cpp
+++ b/src/graph/mutators/DepthConcatSubTensorMutator.cpp
@@ -26,6 +26,7 @@
 #include "arm_compute/graph/Graph.h"
 #include "arm_compute/graph/Logger.h"
 #include "arm_compute/graph/Utils.h"
+#include "arm_compute/graph/algorithms/TopologicalSort.h"
 #include "arm_compute/graph/backends/BackendRegistry.h"
 #include "arm_compute/graph/nodes/ConcatenateLayerNode.h"
 
@@ -43,16 +44,26 @@
 
 void DepthConcatSubTensorMutator::mutate(Graph &g)
 {
-    // Should be in reverse order of execution
-    for(auto &node : arm_compute::utils::iterable::reverse_iterate(g.nodes()))
+    // Early exit if no Concatenation layers exist in graph
+    if(g.nodes(NodeType::ConcatenateLayer).empty())
     {
-        if(node && node->type() == NodeType::ConcatenateLayer && node->output(0) != nullptr)
+        return;
+    }
+
+    // Perform topological sort
+    std::vector<NodeID> topological_sorted_node_ids = dfs(g);
+
+    // Should be in reverse order of execution
+    for(auto &node_id : arm_compute::utils::iterable::reverse_iterate(topological_sorted_node_ids))
+    {
+        INode *node = g.node(node_id);
+        if(node != nullptr && node->type() == NodeType::ConcatenateLayer && node->output(0) != nullptr)
         {
             // Get output tensor
             auto output_tensor = node->output(0);
 
             // Check concatenation axis (Sub-tensor optimization is support for concatenation axis >=2)
-            auto *concat_node = arm_compute::utils::cast::polymorphic_downcast<ConcatenateLayerNode *>(node.get());
+            auto *concat_node = arm_compute::utils::cast::polymorphic_downcast<ConcatenateLayerNode *>(node);
             if(output_tensor == nullptr || get_dimension_idx(output_tensor->desc(), concat_node->concatenation_axis()) < 2)
             {
                 continue;
@@ -84,7 +95,7 @@
                     depth += input_shape.z();
                 }
 
-                auto *dc_node = arm_compute::utils::cast::polymorphic_downcast<ConcatenateLayerNode *>(node.get());
+                auto *dc_node = arm_compute::utils::cast::polymorphic_downcast<ConcatenateLayerNode *>(node);
                 dc_node->set_enabled(false);
             }
         }