Reduce mem usage a little

Author: ranjanan

src/classical.jl

@@ -30,21 +30,21 @@ function ruge_stuben(A::SparseMatrixCSC;
 end
 
 function extend_heirarchy!(levels::Vector{Level}, strength, CF, A)
-    S = strength_of_connection(strength, A)
-    splitting = split_nodes(CF, S)
-    P, R = direct_interpolation(A, S, splitting)
+    S, T = strength_of_connection(strength, A)
+    splitting = split_nodes(CF, S, T)
+    P, R = direct_interpolation(A, T, splitting)
     push!(levels, Level(A, P, R))
     A = R * A * P
 end
 
-function direct_interpolation{T,V}(A::T, S::T, splitting::Vector{V})
+function direct_interpolation(A, T, splitting)
 
-    fill!(S.nzval, 1.)
-    S = A .* S
-    Pp = rs_direct_interpolation_pass1(S, A, splitting)
+    fill!(T.nzval, 1.)
+    T = A .* T
+    Pp = rs_direct_interpolation_pass1(T, A, splitting)
     Pp = Pp .+ 1
 
-    Px, Pj, Pp = rs_direct_interpolation_pass2(A, S, splitting, Pp)
+    Px, Pj, Pp = rs_direct_interpolation_pass2(A, T, splitting, Pp)
 
     # Px .= abs.(Px)
     Pj = Pj .+ 1
@@ -56,10 +56,9 @@ function direct_interpolation{T,V}(A::T, S::T, splitting::Vector{V})
 end
 
 
-function rs_direct_interpolation_pass1(S, A, splitting)
+function rs_direct_interpolation_pass1(T, A, splitting)
 
      Bp = zeros(Int, size(A.colptr))
-     T = S'
      #=Sp = S.colptr
      Sj = S.rowval
      n_nodes = size(A, 1)
@@ -97,12 +96,11 @@ function rs_direct_interpolation_pass1(S, A, splitting)
 
 
  function rs_direct_interpolation_pass2{Tv, Ti}(A::SparseMatrixCSC{Tv,Ti},
-                                                S::SparseMatrixCSC{Tv,Ti},
+                                                T::SparseMatrixCSC{Tv, Ti},
                                                 splitting::Vector{Ti},
                                                 Bp::Vector{Ti})
 
 
-    T = S'
     Bx = zeros(Float64, Bp[end] - 1)
     Bj = zeros(Ti, Bp[end] - 1)
 

src/splitting.jl

@@ -4,15 +4,15 @@ const U_NODE = 2
 
 struct RS
 end
-
-split_nodes(::RS, S::SparseMatrixCSC) = RS_CF_splitting(S - spdiagm(diag(S)))
-function RS_CF_splitting(S::SparseMatrixCSC)
+split_nodes(x, S) = split_nodes(x, S, S')
+split_nodes(::RS, S::SparseMatrixCSC, T::SparseMatrixCSC) = RS_CF_splitting(S - spdiagm(diag(S)), T - spdiagm(diag(T)))
+function RS_CF_splitting(S::SparseMatrixCSC, T::SparseMatrixCSC)
 
 	m,n = size(S)
 
 	n_nodes = n
 	lambda = zeros(Int, n)
-	T = S'
+
 	Tp = T.colptr
 	Tj = T.rowval
 	Sp = S.colptr

src/strength.jl

@@ -39,7 +39,8 @@ function strength_of_connection{T}(c::Classical{T}, A::SparseMatrixCSC)
 
     scale_cols_by_largest_entry!(S)
 
-    S'
+
+    S', S
 end
 
 function find_max_off_diag(neighbors, col)

test/runtests.jl

@@ -12,13 +12,13 @@ ref_split = readdlm("ref_split_test.txt")
 
 # classical strength of connection
 A = poisson(5)
-S = strength_of_connection(Classical(0.2), A)
+S, T = strength_of_connection(Classical(0.2), A)
 @test full(S) == [ 1.0  0.5  0.0  0.0  0.0
                    0.5  1.0  0.5  0.0  0.0
                    0.0  0.5  1.0  0.5  0.0
                    0.0  0.0  0.5  1.0  0.5
                    0.0  0.0  0.0  0.5  1.0 ]
-S = strength_of_connection(Classical(0.25), graph)
+S, T = strength_of_connection(Classical(0.25), graph)
 diff = S - ref_S
 @test maximum(diff) < 1e-10
 
@@ -34,12 +34,12 @@ S = sprand(10,10,0.1); S = S + S'
 @test split_nodes(RS(), S) ==  [0, 1, 1, 0, 0, 0, 0, 0, 1, 1]
 
 a = load("thing.jld")["G"]
-S = AMG.strength_of_connection(AMG.Classical(0.25), a)
+S, T = AMG.strength_of_connection(AMG.Classical(0.25), a)
 @test split_nodes(RS(), S) == [0, 0, 1, 0, 1, 1, 0, 1, 0, 0, 1, 0, 1, 1, 0, 0, 1, 0,
 0, 0, 1, 0, 1, 0, 1, 0, 0, 1, 0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 0, 0, 0, 1, 1, 0,
 1, 0]
 
-@test split_nodes(RS(), ref_S) == Int.(vec(ref_split))
+@test split_nodes(RS(), ref_S, ref_S') == Int.(vec(ref_split))
 
 end