Add accurate arithmetic for summation, fix order of input particles

src/operation/apply.jl

@@ -160,7 +160,6 @@ function node_fusion!(
     # clear snapshot
     get_snapshot_diff(graph)
 
-
     # save children and parents
     n1_children = children(n1)
     n3_parents = parents(n3)
@@ -181,26 +180,18 @@ function node_fusion!(
         ComputeTaskNode(FusedComputeTask{typeof(n1.task), typeof(n3.task)}())
     insert_node!(graph, new_node)
 
-    # use a set for combined children of n1 and n3 to not get duplicates
-    n1and3_children = Set{Node}()
-
-    # remove edges from n1 children to n1
     for child in n1_children
         remove_edge!(graph, child, n1)
-        push!(n1and3_children, child)
-    end
-
-    # remove edges from n3 children to n3
-    for child in n3_children
-        remove_edge!(graph, child, n3)
-        push!(n1and3_children, child)
-    end
-
-    for child in n1and3_children
         insert_edge!(graph, child, new_node)
     end
 
-    # "repoint" parents of n3 from new node
+    for child in n3_children
+        remove_edge!(graph, child, n3)
+        if !(child in n1_children)
+            insert_edge!(graph, child, new_node)
+        end
+    end
+
     for parent in n3_parents
         remove_edge!(graph, n3, parent)
         insert_edge!(graph, new_node, parent)

src/operation/clean.jl

@@ -71,14 +71,8 @@ function find_reductions!(graph::DAG, node::Node)
     partners_ = partners(node)
     delete!(partners_, node)
     for partner in partners_
-        if partner ∉ graph.nodes
-            error("Partner is not part of the graph")
-        end
-
+        @assert partner in graph.nodes
         if can_reduce(node, partner)
-            if Set(node.children) != Set(partner.children)
-                error("Not equal children")
-            end
             if reductionVector === nothing
                 # only when there's at least one reduction partner, insert the vector
                 reductionVector = Vector{Node}()