Merge pull request #153 from HodgeLab/mb/immutable

add ImmutableNonlinearProblem

Author: ChrisRackauckas

ext/SimpleNonlinearSolveChainRulesCoreExt.jl

@@ -2,13 +2,13 @@ module SimpleNonlinearSolveChainRulesCoreExt
 
 using ChainRulesCore: ChainRulesCore, NoTangent
 using DiffEqBase: DiffEqBase
-using SciMLBase: ChainRulesOriginator, NonlinearProblem, NonlinearLeastSquaresProblem
-using SimpleNonlinearSolve: SimpleNonlinearSolve
+using SciMLBase: ChainRulesOriginator, NonlinearLeastSquaresProblem
+using SimpleNonlinearSolve: SimpleNonlinearSolve, ImmutableNonlinearProblem
 
 # The expectation here is that no-one is using this directly inside a GPU kernel. We can
 # eventually lift this requirement using a custom adjoint
 function ChainRulesCore.rrule(::typeof(SimpleNonlinearSolve.__internal_solve_up),
-        prob::Union{NonlinearProblem, NonlinearLeastSquaresProblem},
+        prob::Union{ImmutableNonlinearProblem, NonlinearLeastSquaresProblem},
         sensealg, u0, u0_changed, p, p_changed, alg, args...; kwargs...)
     out, ∇internal = DiffEqBase._solve_adjoint(
         prob, sensealg, u0, p, ChainRulesOriginator(), alg, args...; kwargs...)

ext/SimpleNonlinearSolveReverseDiffExt.jl

@@ -3,11 +3,11 @@ module SimpleNonlinearSolveReverseDiffExt
 using ArrayInterface: ArrayInterface
 using DiffEqBase: DiffEqBase
 using ReverseDiff: ReverseDiff, TrackedArray, TrackedReal
-using SciMLBase: ReverseDiffOriginator, NonlinearProblem, NonlinearLeastSquaresProblem
-using SimpleNonlinearSolve: SimpleNonlinearSolve
+using SciMLBase: ReverseDiffOriginator, NonlinearLeastSquaresProblem
+using SimpleNonlinearSolve: SimpleNonlinearSolve, ImmutableNonlinearProblem
 import SimpleNonlinearSolve: __internal_solve_up
 
-for pType in (NonlinearProblem, NonlinearLeastSquaresProblem)
+for pType in (ImmutableNonlinearProblem, NonlinearLeastSquaresProblem)
     @eval begin
         function __internal_solve_up(prob::$(pType), sensealg, u0::TrackedArray, u0_changed,
                 p::TrackedArray, p_changed, alg, args...; kwargs...)

ext/SimpleNonlinearSolveTrackerExt.jl

@@ -1,11 +1,11 @@
 module SimpleNonlinearSolveTrackerExt
 
 using DiffEqBase: DiffEqBase
-using SciMLBase: TrackerOriginator, NonlinearProblem, NonlinearLeastSquaresProblem, remake
-using SimpleNonlinearSolve: SimpleNonlinearSolve
+using SciMLBase: TrackerOriginator, NonlinearLeastSquaresProblem, remake
+using SimpleNonlinearSolve: SimpleNonlinearSolve, ImmutableNonlinearProblem
 using Tracker: Tracker, TrackedArray
 
-for pType in (NonlinearProblem, NonlinearLeastSquaresProblem)
+for pType in (ImmutableNonlinearProblem, NonlinearLeastSquaresProblem)
     @eval begin
         function SimpleNonlinearSolve.__internal_solve_up(
                 prob::$(pType), sensealg, u0::TrackedArray,

src/SimpleNonlinearSolve.jl

@@ -19,10 +19,11 @@ using LinearAlgebra: LinearAlgebra, I, convert, copyto!, diagind, dot, issuccess
                      norm, transpose
 using MaybeInplace: @bb, setindex_trait, CanSetindex, CannotSetindex
 using Reexport: @reexport
-using SciMLBase: SciMLBase, AbstractNonlinearProblem, IntervalNonlinearProblem,
+using SciMLBase: @add_kwonly, SciMLBase, AbstractNonlinearProblem, IntervalNonlinearProblem,
+                 AbstractNonlinearFunction, StandardNonlinearProblem,
                  NonlinearFunction, NonlinearLeastSquaresProblem, NonlinearProblem,
                  ReturnCode, init, remake, solve, AbstractNonlinearAlgorithm,
-                 build_solution, isinplace, _unwrap_val
+                 build_solution, isinplace, _unwrap_val, warn_paramtype
 using Setfield: @set!
 using StaticArraysCore: StaticArray, SVector, SMatrix, SArray, MArray, Size
 
@@ -35,7 +36,7 @@ abstract type AbstractBracketingAlgorithm <: AbstractSimpleNonlinearSolveAlgorit
 abstract type AbstractNewtonAlgorithm <: AbstractSimpleNonlinearSolveAlgorithm end
 
 @inline __is_extension_loaded(::Val) = false
-
+include("immutable_nonlinear_problem.jl")
 include("utils.jl")
 include("linesearch.jl")
 
@@ -70,6 +71,18 @@ end
 # By Pass the highlevel checks for NonlinearProblem for Simple Algorithms
 function SciMLBase.solve(prob::NonlinearProblem, alg::AbstractSimpleNonlinearSolveAlgorithm,
         args...; sensealg = nothing, u0 = nothing, p = nothing, kwargs...)
+    prob = convert(ImmutableNonlinearProblem, prob)
+    if sensealg === nothing && haskey(prob.kwargs, :sensealg)
+        sensealg = prob.kwargs[:sensealg]
+    end
+    new_u0 = u0 !== nothing ? u0 : prob.u0
+    new_p = p !== nothing ? p : prob.p
+    return __internal_solve_up(prob, sensealg, new_u0, u0 === nothing, new_p,
+        p === nothing, alg, args...; prob.kwargs..., kwargs...)
+end
+
+function SciMLBase.solve(prob::ImmutableNonlinearProblem, alg::AbstractSimpleNonlinearSolveAlgorithm,
+    args...; sensealg = nothing, u0 = nothing, p = nothing, kwargs...)
     if sensealg === nothing && haskey(prob.kwargs, :sensealg)
         sensealg = prob.kwargs[:sensealg]
     end
@@ -79,7 +92,7 @@ function SciMLBase.solve(prob::NonlinearProblem, alg::AbstractSimpleNonlinearSol
         p === nothing, alg, args...; prob.kwargs..., kwargs...)
 end
 
-function __internal_solve_up(_prob::NonlinearProblem, sensealg, u0, u0_changed,
+function __internal_solve_up(_prob::ImmutableNonlinearProblem, sensealg, u0, u0_changed,
         p, p_changed, alg, args...; kwargs...)
     prob = u0_changed || p_changed ? remake(_prob; u0, p) : _prob
     return SciMLBase.__solve(prob, alg, args...; kwargs...)

src/ad.jl

@@ -1,15 +1,26 @@
-for pType in (NonlinearProblem, NonlinearLeastSquaresProblem)
-    @eval function SciMLBase.solve(
-            prob::$(pType){<:Union{Number, <:AbstractArray}, iip,
-                <:Union{<:Dual{T, V, P}, <:AbstractArray{<:Dual{T, V, P}}}},
-            alg::AbstractSimpleNonlinearSolveAlgorithm,
-            args...;
-            kwargs...) where {T, V, P, iip}
-        sol, partials = __nlsolve_ad(prob, alg, args...; kwargs...)
-        dual_soln = __nlsolve_dual_soln(sol.u, partials, prob.p)
-        return SciMLBase.build_solution(
-            prob, alg, dual_soln, sol.resid; sol.retcode, sol.stats, sol.original)
-    end
+function SciMLBase.solve(
+        prob::NonlinearLeastSquaresProblem{<:Union{Number, <:AbstractArray}, iip,
+            <:Union{<:Dual{T, V, P}, <:AbstractArray{<:Dual{T, V, P}}}},
+        alg::AbstractSimpleNonlinearSolveAlgorithm,
+        args...;
+        kwargs...) where {T, V, P, iip}
+    sol, partials = __nlsolve_ad(prob, alg, args...; kwargs...)
+    dual_soln = __nlsolve_dual_soln(sol.u, partials, prob.p)
+    return SciMLBase.build_solution(
+        prob, alg, dual_soln, sol.resid; sol.retcode, sol.stats, sol.original)
+end
+
+function SciMLBase.solve(
+        prob::NonlinearProblem{<:Union{Number, <:AbstractArray}, iip,
+            <:Union{<:Dual{T, V, P}, <:AbstractArray{<:Dual{T, V, P}}}},
+        alg::AbstractSimpleNonlinearSolveAlgorithm,
+        args...;
+        kwargs...) where {T, V, P, iip}
+    prob = convert(ImmutableNonlinearProblem, prob)
+    sol, partials = __nlsolve_ad(prob, alg, args...; kwargs...)
+    dual_soln = __nlsolve_dual_soln(sol.u, partials, prob.p)
+    return SciMLBase.build_solution(
+        prob, alg, dual_soln, sol.resid; sol.retcode, sol.stats, sol.original)
 end
 
 for algType in (Bisection, Brent, Alefeld, Falsi, ITP, Ridder)
@@ -31,7 +42,7 @@ for algType in (Bisection, Brent, Alefeld, Falsi, ITP, Ridder)
 end
 
 function __nlsolve_ad(
-        prob::Union{IntervalNonlinearProblem, NonlinearProblem}, alg, args...; kwargs...)
+        prob::Union{IntervalNonlinearProblem, NonlinearProblem, ImmutableNonlinearProblem}, alg, args...; kwargs...)
     p = value(prob.p)
     if prob isa IntervalNonlinearProblem
         tspan = value.(prob.tspan)

src/immutable_nonlinear_problem.jl

@@ -0,0 +1,67 @@
+struct ImmutableNonlinearProblem{uType, isinplace, P, F, K, PT} <:
+    AbstractNonlinearProblem{uType, isinplace}
+ f::F
+ u0::uType
+ p::P
+ problem_type::PT
+ kwargs::K
+ @add_kwonly function ImmutableNonlinearProblem{iip}(f::AbstractNonlinearFunction{iip}, u0,
+         p = NullParameters(),
+         problem_type = StandardNonlinearProblem();
+         kwargs...) where {iip}
+     if haskey(kwargs, :p)
+         error("`p` specified as a keyword argument `p = $(kwargs[:p])` to `NonlinearProblem`. This is not supported.")
+     end
+     warn_paramtype(p)
+     new{typeof(u0), iip, typeof(p), typeof(f),
+         typeof(kwargs), typeof(problem_type)}(f,
+         u0,
+         p,
+         problem_type,
+         kwargs)
+ end
+
+ """
+ Define a steady state problem using the given function.
+ `isinplace` optionally sets whether the function is inplace or not.
+ This is determined automatically, but not inferred.
+ """
+ function ImmutableNonlinearProblem{iip}(f, u0, p = NullParameters(); kwargs...) where {iip}
+     ImmutableNonlinearProblem{iip}(NonlinearFunction{iip}(f), u0, p; kwargs...)
+ end
+end
+
+"""
+Define a nonlinear problem using an instance of
+[`AbstractNonlinearFunction`](@ref AbstractNonlinearFunction).
+"""
+function ImmutableNonlinearProblem(f::AbstractNonlinearFunction, u0, p = NullParameters(); kwargs...)
+ ImmutableNonlinearProblem{isinplace(f)}(f, u0, p; kwargs...)
+end
+
+function ImmutableNonlinearProblem(f, u0, p = NullParameters(); kwargs...)
+ ImmutableNonlinearProblem(NonlinearFunction(f), u0, p; kwargs...)
+end
+
+"""
+Define a ImmutableNonlinearProblem problem from SteadyStateProblem
+"""
+function ImmutableNonlinearProblem(prob::AbstractNonlinearProblem)
+ ImmutableNonlinearProblem{isinplace(prob)}(prob.f, prob.u0, prob.p)
+end
+
+
+function Base.convert(::Type{ImmutableNonlinearProblem}, prob::T) where {T <: NonlinearProblem}
+ ImmutableNonlinearProblem{isinplace(prob)}(prob.f,
+     prob.u0,
+     prob.p,
+     prob.problem_type;
+     prob.kwargs...)
+end
+
+function DiffEqBase.get_concrete_problem(prob::ImmutableNonlinearProblem, isadapt; kwargs...)
+    u0 = DiffEqBase.get_concrete_u0(prob, isadapt, nothing, kwargs)
+    u0 = DiffEqBase.promote_u0(u0, prob.p, nothing)
+    p = DiffEqBase.get_concrete_p(prob, kwargs)
+    DiffEqBase.remake(prob; u0 = u0, p = p)
+end

src/nlsolve/broyden.jl

@@ -22,7 +22,7 @@ end
 
 __get_linesearch(::SimpleBroyden{LS}) where {LS} = Val(LS)
 
-function SciMLBase.__solve(prob::NonlinearProblem, alg::SimpleBroyden, args...;
+function SciMLBase.__solve(prob::ImmutableNonlinearProblem, alg::SimpleBroyden, args...;
         abstol = nothing, reltol = nothing, maxiters = 1000,
         alias_u0 = false, termination_condition = nothing, kwargs...)
     x = __maybe_unaliased(prob.u0, alias_u0)

src/nlsolve/dfsane.jl

@@ -54,7 +54,7 @@ function SimpleDFSane(; σ_min::Real = 1e-10, σ_max::Real = 1e10, σ_1::Real =
         σ_min, σ_max, σ_1, γ, τ_min, τ_max, nexp, η_strategy)
 end
 
-function SciMLBase.__solve(prob::NonlinearProblem, alg::SimpleDFSane{M}, args...;
+function SciMLBase.__solve(prob::ImmutableNonlinearProblem, alg::SimpleDFSane{M}, args...;
         abstol = nothing, reltol = nothing, maxiters = 1000, alias_u0 = false,
         termination_condition = nothing, kwargs...) where {M}
     x = __maybe_unaliased(prob.u0, alias_u0)

src/nlsolve/halley.jl

@@ -24,7 +24,7 @@ A low-overhead implementation of Halley's Method.
     autodiff = nothing
 end
 
-function SciMLBase.__solve(prob::NonlinearProblem, alg::SimpleHalley, args...;
+function SciMLBase.__solve(prob::ImmutableNonlinearProblem, alg::SimpleHalley, args...;
         abstol = nothing, reltol = nothing, maxiters = 1000,
         alias_u0 = false, termination_condition = nothing, kwargs...)
     x = __maybe_unaliased(prob.u0, alias_u0)

src/nlsolve/klement.jl

@@ -6,7 +6,7 @@ method is non-allocating on scalar and static array problems.
 """
 struct SimpleKlement <: AbstractSimpleNonlinearSolveAlgorithm end
 
-function SciMLBase.__solve(prob::NonlinearProblem, alg::SimpleKlement, args...;
+function SciMLBase.__solve(prob::ImmutableNonlinearProblem, alg::SimpleKlement, args...;
         abstol = nothing, reltol = nothing, maxiters = 1000,
         alias_u0 = false, termination_condition = nothing, kwargs...)
     x = __maybe_unaliased(prob.u0, alias_u0)

src/nlsolve/lbroyden.jl

@@ -29,7 +29,7 @@ function SimpleLimitedMemoryBroyden(;
     return SimpleLimitedMemoryBroyden{_unwrap_val(threshold), _unwrap_val(linesearch)}(alpha)
 end
 
-function SciMLBase.__solve(prob::NonlinearProblem, alg::SimpleLimitedMemoryBroyden,
+function SciMLBase.__solve(prob::ImmutableNonlinearProblem, alg::SimpleLimitedMemoryBroyden,
         args...; termination_condition = nothing, kwargs...)
     if prob.u0 isa SArray
         if termination_condition === nothing ||
@@ -44,7 +44,7 @@ function SciMLBase.__solve(prob::NonlinearProblem, alg::SimpleLimitedMemoryBroyd
     return __generic_solve(prob, alg, args...; termination_condition, kwargs...)
 end
 
-@views function __generic_solve(prob::NonlinearProblem, alg::SimpleLimitedMemoryBroyden,
+@views function __generic_solve(prob::ImmutableNonlinearProblem, alg::SimpleLimitedMemoryBroyden,
         args...; abstol = nothing, reltol = nothing, maxiters = 1000,
         alias_u0 = false, termination_condition = nothing, kwargs...)
     x = __maybe_unaliased(prob.u0, alias_u0)
@@ -114,7 +114,7 @@ end
 # Non-allocating StaticArrays version of SimpleLimitedMemoryBroyden is actually quite
 # finicky, so we'll implement it separately from the generic version
 # Ignore termination_condition. Don't pass things into internal functions
-function __static_solve(prob::NonlinearProblem{<:SArray}, alg::SimpleLimitedMemoryBroyden,
+function __static_solve(prob::ImmutableNonlinearProblem{<:SArray}, alg::SimpleLimitedMemoryBroyden,
         args...; abstol = nothing, maxiters = 1000, kwargs...)
     x = prob.u0
     fx = _get_fx(prob, x)

src/nlsolve/raphson.jl

@@ -23,7 +23,7 @@ end
 
 const SimpleGaussNewton = SimpleNewtonRaphson
 
-function SciMLBase.__solve(prob::Union{NonlinearProblem, NonlinearLeastSquaresProblem},
+function SciMLBase.__solve(prob::Union{ImmutableNonlinearProblem, NonlinearLeastSquaresProblem},
         alg::SimpleNewtonRaphson, args...; abstol = nothing, reltol = nothing,
         maxiters = 1000, termination_condition = nothing, alias_u0 = false, kwargs...)
     x = __maybe_unaliased(prob.u0, alias_u0)

src/nlsolve/trustRegion.jl

@@ -55,7 +55,7 @@ scalar and static array problems.
     nlsolve_update_rule = Val(false)
 end
 
-function SciMLBase.__solve(prob::NonlinearProblem, alg::SimpleTrustRegion, args...;
+function SciMLBase.__solve(prob::ImmutableNonlinearProblem, alg::SimpleTrustRegion, args...;
         abstol = nothing, reltol = nothing, maxiters = 1000,
         alias_u0 = false, termination_condition = nothing, kwargs...)
     x = __maybe_unaliased(prob.u0, alias_u0)

src/utils.jl

@@ -123,7 +123,7 @@ end
         error("Inplace NonlinearLeastSquaresProblem requires a `resid_prototype`")
     return _get_fx(prob.f, x, prob.p)
 end
-@inline _get_fx(prob::NonlinearProblem, x) = _get_fx(prob.f, x, prob.p)
+@inline _get_fx(prob::ImmutableNonlinearProblem, x) = _get_fx(prob.f, x, prob.p)
 @inline function _get_fx(f::NonlinearFunction, x, p)
     if isinplace(f)
         if f.resid_prototype !== nothing
@@ -145,7 +145,7 @@ end
 # different. NonlinearSolve is more for robust / cached solvers while SimpleNonlinearSolve
 # is meant for low overhead solvers, users can opt into the other termination modes but the
 # default is to use the least overhead version.
-function init_termination_cache(prob::NonlinearProblem, abstol, reltol, du, u, ::Nothing)
+function init_termination_cache(prob::ImmutableNonlinearProblem, abstol, reltol, du, u, ::Nothing)
     return init_termination_cache(
         prob, abstol, reltol, du, u, AbsNormTerminationMode(Base.Fix1(maximum, abs)))
 end
@@ -155,14 +155,14 @@ function init_termination_cache(
         prob, abstol, reltol, du, u, AbsNormTerminationMode(Base.Fix2(norm, 2)))
 end
 
-function init_termination_cache(prob::Union{NonlinearProblem, NonlinearLeastSquaresProblem},
+function init_termination_cache(prob::Union{ImmutableNonlinearProblem, NonlinearLeastSquaresProblem},
         abstol, reltol, du, u, tc::AbstractNonlinearTerminationMode)
     T = promote_type(eltype(du), eltype(u))
     abstol = __get_tolerance(u, abstol, T)
     reltol = __get_tolerance(u, reltol, T)
     tc_ = if hasfield(typeof(tc), :internalnorm) && tc.internalnorm === nothing
         internalnorm = ifelse(
-            prob isa NonlinearProblem, Base.Fix1(maximum, abs), Base.Fix2(norm, 2))
+            prob isa ImmutableNonlinearProblem, Base.Fix1(maximum, abs), Base.Fix2(norm, 2))
         DiffEqBase.set_termination_mode_internalnorm(tc, internalnorm)
     else
         tc

test/core/adjoint_tests.jl

@@ -16,6 +16,6 @@
     ∂p_forwarddiff = ForwardDiff.gradient(solve_nlprob, p)
     ∂p_tracker = Tracker.data(only(Tracker.gradient(solve_nlprob, p)))
     ∂p_reversediff = ReverseDiff.gradient(solve_nlprob, p)
-
+    @test ∂p_zygote ≈ ∂p_tracker ≈ ∂p_reversediff
     @test ∂p_zygote ≈ ∂p_forwarddiff ≈ ∂p_tracker ≈ ∂p_reversediff
 end

test/gpu/cuda_tests.jl

@@ -51,6 +51,7 @@ end
     end
 
     prob = NonlinearProblem{false}(f, @SVector[1.0f0, 1.0f0])
+    prob = convert(SimpleNonlinearSolve.ImmutableNonlinearProblem, prob)
 
     @testset "$(nameof(typeof(alg)))" for alg in (
         SimpleNewtonRaphson(), SimpleDFSane(), SimpleTrustRegion(),