From dde9ec96f471127e5b6d8dfaeffce024b6326f1a Mon Sep 17 00:00:00 2001
From: Michele Di Giorgio <michele.digiorgio@arm.com>
Date: Tue, 13 Feb 2018 15:24:04 +0000
Subject: COMPMID-909: Enabling in-place computation for batchnormalization and
 activation at graph level

Change-Id: I84d4a212629b21794451ab5fb5c5b187b5e28f98
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/120127
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Jenkins <bsgcomp@arm.com>
---
 src/graph/Graph.cpp                 | 13 +++++++++++--
 src/graph/INode.cpp                 | 10 +++++++++-
 src/graph/SubGraph.cpp              |  6 +++++-
 src/graph/nodes/ActivationLayer.cpp |  3 ++-
 4 files changed, 27 insertions(+), 5 deletions(-)

(limited to 'src/graph')

diff --git a/src/graph/Graph.cpp b/src/graph/Graph.cpp
index 7af313acbb..98d95904dc 100644
--- a/src/graph/Graph.cpp
+++ b/src/graph/Graph.cpp
@@ -131,6 +131,11 @@ void Graph::Private::configure(GraphHints _next_hints)
         _previous_hints = _current_hints; // For the first node just assume the previous node was of the same type as this one
     }
 
+    if(_current_node->supports_in_place())
+    {
+        _current_output = _current_input;
+    }
+
     //Automatic output configuration ?
     if(_current_output == nullptr)
     {
@@ -152,8 +157,12 @@ void Graph::Private::configure(GraphHints _next_hints)
     _ctx.hints()                                 = _current_hints;
     std::unique_ptr<arm_compute::IFunction> func = _current_node->instantiate_node(_ctx, _current_input, _current_output);
 
-    // Allocate current input
-    _current_input->allocate();
+    // If the operation is done in-place, do not allocate or it will prevent following layers from performing the configuration
+    if(!_current_node->supports_in_place())
+    {
+        // Allocate current input
+        _current_input->allocate();
+    }
 
     // Map input if needed
     if(_current_input->target() == TargetHint::OPENCL)
diff --git a/src/graph/INode.cpp b/src/graph/INode.cpp
index 582f936351..c753f66b43 100644
--- a/src/graph/INode.cpp
+++ b/src/graph/INode.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -39,6 +39,14 @@ TargetHint INode::override_target_hint(TargetHint target_hint) const
     ARM_COMPUTE_ERROR_ON(target_hint == TargetHint::OPENCL && !opencl_is_available());
     return target_hint;
 }
+bool INode::supports_in_place() const
+{
+    return _supports_in_place;
+}
+void INode::set_supports_in_place(bool value)
+{
+    _supports_in_place = value;
+}
 GraphHints INode::node_override_hints(GraphHints hints) const
 {
     TargetHint target_hint = hints.target_hint();
diff --git a/src/graph/SubGraph.cpp b/src/graph/SubGraph.cpp
index f62b2617c5..b1cbb9cc95 100644
--- a/src/graph/SubGraph.cpp
+++ b/src/graph/SubGraph.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -67,6 +67,10 @@ std::unique_ptr<Graph> SubGraph::construct(const GraphContext &ctx, std::unique_
     }
     graph->add_tensor_object(std::move(_input));
 
+    // Make sure first and last nodes of the subgraph always do operations out-of-place
+    _nodes.front()->set_supports_in_place(false);
+    _nodes.back()->set_supports_in_place(false);
+
     // Construct nodes
     for(auto &node : _nodes)
     {
diff --git a/src/graph/nodes/ActivationLayer.cpp b/src/graph/nodes/ActivationLayer.cpp
index 54f30ef777..546c42a1e5 100644
--- a/src/graph/nodes/ActivationLayer.cpp
+++ b/src/graph/nodes/ActivationLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -33,6 +33,7 @@ using namespace arm_compute::graph;
 ActivationLayer::ActivationLayer(const ActivationLayerInfo activation_info)
     : _activation_info(activation_info)
 {
+    set_supports_in_place(true);
 }
 
 std::unique_ptr<arm_compute::IFunction> ActivationLayer::instantiate_node(GraphContext &ctx, ITensorObject *input, ITensorObject *output)
-- 
cgit v1.2.1