Merge pull request #76 from avik-pal/ap/inplace_raphson

Add support for inplace BatchedSimpleNewtonRaphson

Author: ChrisRackauckas

Project.toml

@@ -1,7 +1,7 @@
 name = "SimpleNonlinearSolve"
 uuid = "727e6d20-b764-4bd8-a329-72de5adea6c7"
 authors = ["SciML"]
-version = "0.1.18"
+version = "0.1.19"
 
 [deps]
 ArrayInterface = "4fba245c-0d91-5ea0-9b3e-6abc04ee57a9"

src/batched/raphson.jl

@@ -20,7 +20,8 @@ end
 function SciMLBase.__solve(prob::NonlinearProblem, alg::BatchedSimpleNewtonRaphson;
     abstol = nothing, reltol = nothing, maxiters = 1000, kwargs...)
     iip = SciMLBase.isinplace(prob)
-    @assert !iip "BatchedSimpleNewtonRaphson currently only supports out-of-place nonlinear problems."
+    iip &&
+        @assert alg_autodiff(alg) "Inplace BatchedSimpleNewtonRaphson currently only supports autodiff."
     u, f, reconstruct = _construct_batched_problem_structure(prob)
 
     tc = alg.termination_condition
@@ -35,12 +36,26 @@ function SciMLBase.__solve(prob::NonlinearProblem, alg::BatchedSimpleNewtonRaphs
     rtol = _get_tolerance(reltol, tc.reltol, T)
     termination_condition = tc(storage)
 
+    if iip
+        𝓙 = similar(xₙ, length(xₙ), length(xₙ))
+        fₙ = similar(xₙ)
+        jac_cfg = ForwardDiff.JacobianConfig(f, fₙ, xₙ)
+    end
+
     for i in 1:maxiters
-        if alg_autodiff(alg)
-            fₙ, 𝓙 = value_derivative(f, xₙ)
+        if iip
+            value_derivative!(𝓙, fₙ, f, xₙ, jac_cfg)
         else
-            fₙ = f(xₙ)
-            𝓙 = FiniteDiff.finite_difference_jacobian(f, xₙ, diff_type(alg), eltype(xₙ), fₙ)
+            if alg_autodiff(alg)
+                fₙ, 𝓙 = value_derivative(f, xₙ)
+            else
+                fₙ = f(xₙ)
+                𝓙 = FiniteDiff.finite_difference_jacobian(f,
+                    xₙ,
+                    diff_type(alg),
+                    eltype(xₙ),
+                    fₙ)
+            end
         end
 
         iszero(fₙ) && return DiffEqBase.build_solution(prob,
@@ -66,7 +81,7 @@ function SciMLBase.__solve(prob::NonlinearProblem, alg::BatchedSimpleNewtonRaphs
 
     if mode ∈ DiffEqBase.SAFE_BEST_TERMINATION_MODES
         xₙ = storage.u
-        fₙ = f(xₙ)
+        @maybeinplace iip fₙ=f(xₙ)
     end
 
     return DiffEqBase.build_solution(prob,

src/utils.jl

@@ -30,6 +30,20 @@ function value_derivative(f::F, x::R) where {F, R}
 end
 value_derivative(f::F, x::AbstractArray) where {F} = f(x), ForwardDiff.jacobian(f, x)
 
+"""
+    value_derivative!(J, y, f!, x, cfg = JacobianConfig(f!, y, x))
+
+Inplace version of [`SimpleNonlinearSolve.value_derivative`](@ref).
+"""
+function value_derivative!(J::AbstractMatrix,
+    y::AbstractArray,
+    f!::F,
+    x::AbstractArray,
+    cfg::ForwardDiff.JacobianConfig = ForwardDiff.JacobianConfig(f!, y, x)) where {F}
+    ForwardDiff.jacobian!(J, f!, y, x, cfg)
+    return y, J
+end
+
 value(x) = x
 value(x::Dual) = ForwardDiff.value(x)
 value(x::AbstractArray{<:Dual}) = map(ForwardDiff.value, x)

test/inplace.jl

@@ -2,29 +2,28 @@ using SimpleNonlinearSolve,
     StaticArrays, BenchmarkTools, DiffEqBase, LinearAlgebra, Test,
     NNlib
 
-# Supported Solvers: BatchedBroyden, BatchedSimpleDFSane
+# Supported Solvers: BatchedBroyden, BatchedSimpleDFSane, BatchedSimpleNewtonRaphson
 function f!(du::AbstractArray{<:Number, N},
     u::AbstractArray{<:Number, N},
     p::AbstractVector) where {N}
     u_ = reshape(u, :, size(u, N))
-    du .= reshape(sum(abs2, u_; dims = 1) .- reshape(p, 1, :),
-        ntuple(_ -> 1, N - 1)...,
-        size(u, N))
+    du .= reshape(sum(abs2, u_; dims = 1) .- u_ .- reshape(p, 1, :), size(u))
     return du
 end
 
 function f!(du::AbstractMatrix, u::AbstractMatrix, p::AbstractVector)
-    du .= sum(abs2, u; dims = 1) .- reshape(p, 1, :)
+    du .= sum(abs2, u; dims = 1) .- u .- reshape(p, 1, :)
     return du
 end
 
 function f!(du::AbstractVector, u::AbstractVector, p::AbstractVector)
-    du .= sum(abs2, u) .- p
+    du .= sum(abs2, u) .- u .- p
     return du
 end
 
-@testset "Solver: $(nameof(typeof(solver)))" for solver in (Broyden(batched = true),
-    SimpleDFSane(batched = true))
+@testset "Solver: $(nameof(typeof(solver)))" for solver in (Broyden(; batched = true),
+    SimpleDFSane(; batched = true),
+    SimpleNewtonRaphson(; batched = true))
     @testset "T: $T" for T in (Float32, Float64)
         p = rand(T, 5)
         @testset "size(u0): $sz" for sz in ((2, 5), (1, 5), (2, 3, 5))