From 8b2a7d3aa119e7f1d6a03690d05eb27c5d178b9f Mon Sep 17 00:00:00 2001
From: Giorgio Arena <giorgio.arena@arm.com>
Date: Tue, 11 Feb 2020 17:21:31 +0000
Subject: COMPMID-3101 Fuse activation with floating point elementwise
 operation layers in CL

Signed-off-by: Giorgio Arena <giorgio.arena@arm.com>
Change-Id: I1693f8664ba7c0dc8c076bbe7365cef1e667bd25
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2718
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
---
 src/graph/mutators/NodeFusionMutator.cpp | 18 +++++++++++++-----
 1 file changed, 13 insertions(+), 5 deletions(-)

(limited to 'src/graph/mutators/NodeFusionMutator.cpp')

diff --git a/src/graph/mutators/NodeFusionMutator.cpp b/src/graph/mutators/NodeFusionMutator.cpp
index 273e6ce746..ae53b8ff75 100644
--- a/src/graph/mutators/NodeFusionMutator.cpp
+++ b/src/graph/mutators/NodeFusionMutator.cpp
@@ -294,13 +294,20 @@ IGraphMutator::MutationType NodeFusionMutator::type() const
 void NodeFusionMutator::mutate(Graph &g)
 {
     // Supported activations when fusing
-    const std::set<Activation> supported_fused_activations = { Activation::RELU, Activation::BOUNDED_RELU, Activation::LU_BOUNDED_RELU };
+    const std::set<Activation> supported_fused_activations_conv    = { Activation::RELU, Activation::BOUNDED_RELU, Activation::LU_BOUNDED_RELU };
+    const std::set<Activation> supported_fused_activations_eltwise = { Activation::RELU, Activation::BOUNDED_RELU, Activation::LU_BOUNDED_RELU,
+                                                                       Activation::TANH, Activation::LOGISTIC
+                                                                     };
 
     // 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);
@@ -315,10 +322,11 @@ void NodeFusionMutator::mutate(Graph &g)
     };
 
     // Fusion mutations
-    detail::fuse_layer<BatchNormalizationLayerNode, ActivationLayerNode>(g, empty_prec, detail::fuse_node_with_activation<BatchNormalizationLayerNode>, supported_fused_activations);
-    detail::fuse_layer<ConvolutionLayerNode, ActivationLayerNode>(g, empty_prec, detail::fuse_node_with_activation<ConvolutionLayerNode>, supported_fused_activations);
-    detail::fuse_layer<DepthwiseConvolutionLayerNode, ActivationLayerNode>(g, qs8_prec, detail::fuse_node_with_activation<DepthwiseConvolutionLayerNode>, supported_fused_activations);
-    detail::fuse_layer<FullyConnectedLayerNode, ActivationLayerNode>(g, empty_prec, detail::fuse_node_with_activation<FullyConnectedLayerNode>, supported_fused_activations);
+    detail::fuse_layer<BatchNormalizationLayerNode, ActivationLayerNode>(g, empty_prec, detail::fuse_node_with_activation<BatchNormalizationLayerNode>, supported_fused_activations_conv);
+    detail::fuse_layer<ConvolutionLayerNode, ActivationLayerNode>(g, empty_prec, detail::fuse_node_with_activation<ConvolutionLayerNode>, supported_fused_activations_conv);
+    detail::fuse_layer<DepthwiseConvolutionLayerNode, ActivationLayerNode>(g, qs8_prec, detail::fuse_node_with_activation<DepthwiseConvolutionLayerNode>, supported_fused_activations_conv);
+    detail::fuse_layer<FullyConnectedLayerNode, ActivationLayerNode>(g, empty_prec, detail::fuse_node_with_activation<FullyConnectedLayerNode>, supported_fused_activations_conv);
+    detail::fuse_layer<EltwiseLayerNode, ActivationLayerNode>(g, cl_target_prec, detail::fuse_node_with_activation<EltwiseLayerNode>, supported_fused_activations_eltwise);
     detail::fuse_layer<ConvolutionLayerNode, BatchNormalizationLayerNode>(g, empty_prec, detail::fuse_convolution_with_batch_normalization);
     detail::fuse_layer<DepthwiseConvolutionLayerNode, BatchNormalizationLayerNode>(g, empty_prec, detail::fuse_depthwise_convolution_with_batch_normalization);
 }
-- 
cgit v1.2.1