Merge pull request #273 from j-fu/abstractsparse

Use AbstractSparseMatrixCSC for sparse factorizations

Author: ChrisRackauckas

lib/LinearSolvePardiso/Project.toml

@@ -8,12 +8,13 @@ LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 LinearSolve = "7ed4a6bd-45f5-4d41-b270-4a48e9bafcae"
 Pardiso = "46dd5b70-b6fb-5a00-ae2d-e8fea33afaf2"
 SciMLBase = "0bca4576-84f4-4d90-8ffe-ffa030f20462"
+SparseArrays = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"
 UnPack = "3a884ed6-31ef-47d7-9d2a-63182c4928ed"
 
 [compat]
-SciMLBase = "1.25"
 LinearSolve = "1.24"
-Pardiso =  "0.5"
+Pardiso = "0.5"
+SciMLBase = "1.25"
 UnPack = "1"
 julia = "1.6"
 

lib/LinearSolvePardiso/src/LinearSolvePardiso.jl

@@ -1,6 +1,9 @@
 module LinearSolvePardiso
 
 using Pardiso, LinearSolve, SciMLBase
+using SparseArrays
+using SparseArrays: nonzeros, rowvals, getcolptr
+
 using UnPack
 
 Base.@kwdef struct PardisoJL <: LinearSolve.SciMLLinearSolveAlgorithm
@@ -17,8 +20,16 @@ LinearSolve.needs_concrete_A(alg::PardisoJL) = true
 
 # TODO schur complement functionality
 
-function LinearSolve.init_cacheval(alg::PardisoJL, A, b, u, Pl, Pr, maxiters::Int, abstol,
-                                   reltol, verbose::Bool,
+function LinearSolve.init_cacheval(alg::PardisoJL,
+                                   A,
+                                   b,
+                                   u,
+                                   Pl,
+                                   Pr,
+                                   maxiters::Int,
+                                   abstol,
+                                   reltol,
+                                   verbose::Bool,
                                    assumptions::LinearSolve.OperatorAssumptions)
     @unpack nprocs, solver_type, matrix_type, iparm, dparm = alg
     A = convert(AbstractMatrix, A)
@@ -90,7 +101,10 @@ function LinearSolve.init_cacheval(alg::PardisoJL, A, b, u, Pl, Pr, maxiters::In
         Pardiso.set_iparm!(solver, 3, round(Int, abs(log10(reltol)), RoundDown) * 10 + 1)
     end
 
-    Pardiso.pardiso(solver, u, A, b)
+    Pardiso.pardiso(solver,
+                    u,
+                    SparseMatrixCSC(size(A)..., getcolptr(A), rowvals(A), nonzeros(A)),
+                    b)
 
     return solver
 end

lib/LinearSolvePardiso/test/runtests.jl

@@ -17,9 +17,7 @@ cache_kwargs = (; verbose = true, abstol = 1e-8, reltol = 1e-8, maxiter = 30)
 
 prob2 = LinearProblem(A2, b2)
 
-for alg in (PardisoJL(),
-            MKLPardisoFactorize(),
-            MKLPardisoIterate())
+for alg in (PardisoJL(), MKLPardisoFactorize(), MKLPardisoIterate())
     u = solve(prob1, alg; cache_kwargs...).u
     @test A1 * u ≈ b1
 

src/LinearSolve.jl

@@ -9,6 +9,7 @@ using Base: cache_dependencies, Bool
 using LinearAlgebra
 using IterativeSolvers: Identity
 using SparseArrays
+using SparseArrays: AbstractSparseMatrixCSC, nonzeros, rowvals, getcolptr
 using SciMLBase: AbstractLinearAlgorithm
 using SciMLOperators
 using SciMLOperators: AbstractSciMLOperator, IdentityOperator

src/default.jl

@@ -50,13 +50,13 @@ function defaultalg(A::Diagonal, b, ::OperatorAssumptions{Nothing})
     DiagonalFactorization()
 end
 
-function defaultalg(A::SparseMatrixCSC{Tv, Ti}, b,
+function defaultalg(A::AbstractSparseMatrixCSC{Tv, Ti}, b,
                     ::OperatorAssumptions{true}) where {Tv, Ti}
     SparspakFactorization()
 end
 
 @static if INCLUDE_SPARSE
-    function defaultalg(A::SparseMatrixCSC{<:Union{Float64, ComplexF64}, Ti}, b,
+    function defaultalg(A::AbstractSparseMatrixCSC{<:Union{Float64, ComplexF64}, Ti}, b,
                         ::OperatorAssumptions{true}) where {Ti}
         if length(b) <= 10_000
             KLUFactorization()

src/factorization.jl

@@ -52,8 +52,8 @@ end
 
 function do_factorization(alg::LUFactorization, A, b, u)
     A = convert(AbstractMatrix, A)
-    if A isa SparseMatrixCSC
-        return lu(A)
+    if A isa AbstractSparseMatrixCSC
+        return lu(SparseMatrixCSC(size(A)..., getcolptr(A), rowvals(A), nonzeros(A)))
     else
         fact = lu!(A, alg.pivot)
     end
@@ -277,7 +277,8 @@ function init_cacheval(alg::UMFPACKFactorization, A, b, u, Pl, Pr, maxiters::Int
                                              copy(nonzeros(A)), 0)
         finalizer(SuiteSparse.UMFPACK.umfpack_free_symbolic, res)
     else
-        return SuiteSparse.UMFPACK.UmfpackLU(A)
+        return SuiteSparse.UMFPACK.UmfpackLU(SparseMatrixCSC(size(A)..., getcolptr(A),
+                                                             rowvals(A), nonzeros(A)))
     end
 end
 
@@ -290,12 +291,15 @@ function SciMLBase.solve(cache::LinearCache, alg::UMFPACKFactorization; kwargs..
             if alg.check_pattern && !(SuiteSparse.decrement(SparseArrays.getcolptr(A)) ==
                  cache.cacheval.colptr &&
                  SuiteSparse.decrement(SparseArrays.getrowval(A)) == cache.cacheval.rowval)
-                fact = lu(A)
+                fact = lu(SparseMatrixCSC(size(A)..., getcolptr(A), rowvals(A),
+                                          nonzeros(A)))
             else
-                fact = lu!(cache.cacheval, A)
+                fact = lu!(cache.cacheval,
+                           SparseMatrixCSC(size(A)..., getcolptr(A), rowvals(A),
+                                           nonzeros(A)))
             end
         else
-            fact = lu(A)
+            fact = lu(SparseMatrixCSC(size(A)..., getcolptr(A), rowvals(A), nonzeros(A)))
         end
         cache = set_cacheval(cache, fact)
     end
@@ -312,7 +316,9 @@ end
 function init_cacheval(alg::KLUFactorization, A, b, u, Pl, Pr, maxiters::Int, abstol,
                        reltol,
                        verbose::Bool, assumptions::OperatorAssumptions)
-    return KLU.KLUFactorization(convert(AbstractMatrix, A)) # this takes care of the copy internally.
+    A = convert(AbstractMatrix, A)
+    return KLU.KLUFactorization(SparseMatrixCSC(size(A)..., getcolptr(A), rowvals(A),
+                                                nonzeros(A)))
 end
 
 function SciMLBase.solve(cache::LinearCache, alg::KLUFactorization; kwargs...)
@@ -323,21 +329,25 @@ function SciMLBase.solve(cache::LinearCache, alg::KLUFactorization; kwargs...)
             if alg.check_pattern && !(SuiteSparse.decrement(SparseArrays.getcolptr(A)) ==
                  cache.cacheval.colptr &&
                  SuiteSparse.decrement(SparseArrays.getrowval(A)) == cache.cacheval.rowval)
-                fact = KLU.klu(A)
+                fact = KLU.klu(SparseMatrixCSC(size(A)..., getcolptr(A), rowvals(A),
+                                               nonzeros(A)))
             else
                 # If we have a cacheval already, run umfpack_symbolic to ensure the symbolic factorization exists
                 # This won't recompute if it does.
                 KLU.klu_analyze!(cache.cacheval)
-                copyto!(cache.cacheval.nzval, A.nzval)
+                copyto!(cache.cacheval.nzval, nonzeros(A))
                 if cache.cacheval._numeric === C_NULL # We MUST have a numeric factorization for reuse, unlike UMFPACK.
                     KLU.klu_factor!(cache.cacheval)
                 end
-                fact = KLU.klu!(cache.cacheval, A)
+                fact = KLU.klu!(cache.cacheval,
+                                SparseMatrixCSC(size(A)..., getcolptr(A), rowvals(A),
+                                                nonzeros(A)))
             end
         else
             # New fact each time since the sparsity pattern can change
             # and thus it needs to reallocate
-            fact = KLU.klu(A)
+            fact = KLU.klu(SparseMatrixCSC(size(A)..., getcolptr(A), rowvals(A),
+                                           nonzeros(A)))
         end
         cache = set_cacheval(cache, fact)
     end
@@ -511,17 +521,20 @@ function init_cacheval(::SparspakFactorization, A, b, u, Pl, Pr, maxiters::Int,
                        reltol,
                        verbose::Bool, assumptions::OperatorAssumptions)
     A = convert(AbstractMatrix, A)
-    return sparspaklu(A, factorize = false)
+    return sparspaklu(SparseMatrixCSC(size(A)..., getcolptr(A), rowvals(A), nonzeros(A)),
+                      factorize = false)
 end
 
 function SciMLBase.solve(cache::LinearCache, alg::SparspakFactorization; kwargs...)
     A = cache.A
-    A = convert(AbstractMatrix, A)
     if cache.isfresh
         if cache.cacheval !== nothing && alg.reuse_symbolic
-            fact = sparspaklu!(cache.cacheval, A)
+            fact = sparspaklu!(cache.cacheval,
+                               SparseMatrixCSC(size(A)..., getcolptr(A), rowvals(A),
+                                               nonzeros(A)))
         else
-            fact = sparspaklu(A)
+            fact = sparspaklu(SparseMatrixCSC(size(A)..., getcolptr(A), rowvals(A),
+                                              nonzeros(A)))
         end
         cache = set_cacheval(cache, fact)
     end