Changes to reduce allocation

Author: dmbates

src/linalg.jl

@@ -126,9 +126,9 @@ function A_rdiv_Bc!(A::Matrix{T}, B::LowerTriangular{T,UniformBlockDiagonal{T}})
     m, n, k = size(B.data.data)
     @argcheck size(A, 2) == size(B, 1) && m == n DimensionMismatch
     offset = 0
+    one2m = 1:m
     for f in B.data.facevec
-        ## FIXME call BLAS.trsm directly
-        A_rdiv_Bc!(view(A, :, (1:m) + offset), LowerTriangular(f))
+        BLAS.trsm!('R', 'L', 'T', 'N', one(T), f, view(A, :, one2m + offset))
         offset += m
     end
     A

src/linalg/logdet.jl

@@ -8,12 +8,23 @@ Return `log(det(tril(A)))` evaluated in place.
 LD(d::Diagonal{T}) where {T<:Number} = sum(log, d.diag)
 
 function LD(d::UniformBlockDiagonal{T}) where T
-    m, n, k = size(d.data)
-    dind = diagind(m, n)
-    sum(log, f[i] for f in d.facevec, i in dind)
+    dat = d.data
+    m, n, k = size(dat)
+    m == n || throw(ArgumentError("Blocks of d must be square"))
+    s = log(one(T))
+    @inbounds for j in 1:k, i in 1:m
+        s += log(dat[i,i,j])
+    end
+    s
 end
 
-LD(d::DenseMatrix{T}) where {T} = sum(log, d[i] for i in diagind(d))
+function LD(d::DenseMatrix{T}) where T
+    s = log(one(T))
+    for i in 1:Base.LinAlg.checksquare(d)
+        s += log(d[i, i])
+    end
+    s
+end
 
 """
     logdet(m::LinearMixedModel)

src/linalg/scaleInflate.jl

@@ -36,13 +36,16 @@ function scaleInflate!(Ljj::LowerTriangular{T,UniformBlockDiagonal{T}},
                        Ajj::UniformBlockDiagonal{T},
                        Λj::VectorFactorReTerm{T}) where {T}
     @argcheck size(Ljj) == size(Ajj) DimensionMismatch
+    Ljjdat = Ljj.data
+    Ljjdd = Ljjdat.data
+    copy!(Ljjdd, Ajj.data)
+    k, m, n = size(Ljjdd)
     λ = LowerTriangular(Λj.Λ)
-    k = vsize(Λj)
-    for (Lf, Af) in zip(Ljj.data.facevec, Ajj.facevec)
-        Ac_mul_B!(λ, A_mul_B!(copy!(Lf, Af), λ))
-        for j in 1:k
-            Lf[j, j] += one(T)
-        end
+    for Lf in Ljjdat.facevec
+        Ac_mul_B!(λ, A_mul_B!(Lf, λ))
+    end
+    for j in 1:n, i in 1:k
+        Ljjdd[i, i, j] += one(T)
     end
     Ljj
 end