Merge pull request #550 from j-fu/jf/PardisoAbstractSparse

Fix Pardiso extension for the case of an AbstractSparseMatrixCSC

Author: ChrisRackauckas

ext/LinearSolvePardisoExt.jl

@@ -134,12 +134,11 @@ function SciMLBase.solve!(cache::LinearSolve.LinearCache, alg::PardisoJL; kwargs
     if cache.isfresh
         phase = alg.cache_analysis ? Pardiso.NUM_FACT : Pardiso.ANALYSIS_NUM_FACT
         Pardiso.set_phase!(cache.cacheval, phase)
-        Pardiso.pardiso(cache.cacheval, A, eltype(A)[])
+        Pardiso.pardiso(cache.cacheval, SparseMatrixCSC(size(A)..., getcolptr(A), rowvals(A), nonzeros(A)), eltype(A)[])
         cache.isfresh = false
     end
     Pardiso.set_phase!(cache.cacheval, Pardiso.SOLVE_ITERATIVE_REFINE)
-    Pardiso.pardiso(cache.cacheval, u, A, b)
-
+    Pardiso.pardiso(cache.cacheval, u, SparseMatrixCSC(size(A)..., getcolptr(A), rowvals(A), nonzeros(A)), b)
     return SciMLBase.build_linear_solution(alg, cache.u, nothing, cache)
 end
 

src/extension_algs.jl

@@ -217,7 +217,7 @@ All values default to `nothing` and the solver internally determines the values
 given the input types, and these keyword arguments are only for overriding the
 default handling process. This should not be required by most users.
 """
-struct PardisoJL{T1, T2} <: LinearSolve.SciMLLinearSolveAlgorithm
+struct PardisoJL{T1, T2}  <: AbstractSparseFactorization
     nprocs::Union{Int, Nothing}
     solver_type::T1
     matrix_type::T2

test/pardiso/pardiso.jl

@@ -177,3 +177,40 @@ for solver in solvers
         @test Pardiso.get_iparm(solver, i) == iparm[i][2]
     end
 end
+
+@testset "AbstractSparseMatrixCSC" begin
+    struct MySparseMatrixCSC2{Tv, Ti} <: SparseArrays.AbstractSparseMatrixCSC{Tv, Ti}
+        csc::SparseMatrixCSC{Tv, Ti}
+    end
+
+    Base.size(m::MySparseMatrixCSC2) = size(m.csc)
+    SparseArrays.getcolptr(m::MySparseMatrixCSC2) = SparseArrays.getcolptr(m.csc)
+    SparseArrays.rowvals(m::MySparseMatrixCSC2) = SparseArrays.rowvals(m.csc)
+    SparseArrays.nonzeros(m::MySparseMatrixCSC2) = SparseArrays.nonzeros(m.csc)
+
+    for alg in algs
+        N = 100
+        u0 = ones(N)
+        A0 = spdiagm(1 => -ones(N - 1), 0 => fill(10.0, N), -1 => -ones(N - 1))
+        b0 = A0 * u0
+        B0 = MySparseMatrixCSC2(A0)
+        A1 = spdiagm(1 => -ones(N - 1), 0 => fill(100.0, N), -1 => -ones(N - 1))
+        b1=A1*u0
+        B1= MySparseMatrixCSC2(A1)
+
+        
+        pr = LinearProblem(B0, b0)
+        # test default algorithn
+        u=solve(pr,alg)
+        @test norm(u - u0, Inf) < 1.0e-13
+    
+        # test factorization with reinit!
+        pr = LinearProblem(B0, b0)
+        cache=init(pr,alg)
+        u=solve!(cache)
+        @test norm(u - u0, Inf) < 1.0e-13
+        reinit!(cache; A=B1, b=b1)
+        u=solve!(cache)
+        @test norm(u - u0, Inf) < 1.0e-13
+    end
+end