diff options
| -rw-r--r-- | src/graph/mutators/NodeFusionMutator.cpp | 6 |
1 files changed, 4 insertions, 2 deletions
diff --git a/src/graph/mutators/NodeFusionMutator.cpp b/src/graph/mutators/NodeFusionMutator.cpp index 9e963091e6..f7f3454fad 100644 --- a/src/graph/mutators/NodeFusionMutator.cpp +++ b/src/graph/mutators/NodeFusionMutator.cpp @@ -307,8 +307,10 @@ void NodeFusionMutator::mutate(Graph &g) 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); - // TODO (COMPMID-2524): Fuse batch normalization with convolution and depthwise convolution at graph level for NEON - if(target == Target::CL) + // Currently fuse batch normalization brings performance uplift only on OpenCL with FP32 data type + // TODO (COMPMID-2524): Fuse batch normalization with convolution and depthwise convolution at graph level for NEON - FP32 + // TODO (COMPMID-2581): Fuse batch normalization with convolution and depthwise convolution at graph level for OpenCL - FP16 + if(target == Target::CL && (g.nodes()[0].get()->output(0)->desc().data_type == DataType::F32)) { //Depthwise Convolution and Batch Normalization Fusion active only for CL detail::fuse_layer<ConvolutionLayerNode, BatchNormalizationLayerNode>(g, empty_prec, detail::fuse_convolution_with_batch_normalization); |