feat: support contexts with symbolic backends

DifferentiationInterface/Project.toml

@@ -1,7 +1,7 @@
 name = "DifferentiationInterface"
 uuid = "a0c0ee7d-e4b9-4e03-894e-1c5f64a51d63"
 authors = ["Guillaume Dalle", "Adrian Hill"]
-version = "0.6.33"
+version = "0.6.34"
 
 [deps]
 ADTypes = "47edcb42-4c32-4615-8424-f2b9edc5f35b"
@@ -55,7 +55,7 @@ Diffractor = "=0.2.6"
 Enzyme = "0.13.17"
 EnzymeCore = "0.8.8"
 ExplicitImports = "1.10.1"
-FastDifferentiation = "0.4.1"
+FastDifferentiation = "0.4.3"
 FiniteDiff = "2.23.1"
 FiniteDifferences = "0.12.31"
 ForwardDiff = "0.10.36"

DifferentiationInterface/docs/src/explanation/backends.md

@@ -56,23 +56,23 @@ In practice, many AD backends have custom implementations for high-level operato
 
 Moreover, each context type is supported by a specific subset of backends:
 
-|                            | [`Constant`](@ref) |
-| -------------------------- | ------------------ |
-| `AutoChainRules`           | ✅                  |
-| `AutoDiffractor`           | ❌                  |
-| `AutoEnzyme` (forward)     | ✅                  |
-| `AutoEnzyme` (reverse)     | ✅                  |
-| `AutoFastDifferentiation`  | ❌                  |
-| `AutoFiniteDiff`           | ✅                  |
-| `AutoFiniteDifferences`    | ✅                  |
-| `AutoForwardDiff`          | ✅                  |
-| `AutoGTPSA`                | ✅                  |
-| `AutoMooncake`             | ✅                  |
-| `AutoPolyesterForwardDiff` | ✅                  |
-| `AutoReverseDiff`          | ✅                  |
-| `AutoSymbolics`            | ❌                  |
-| `AutoTracker`              | ✅                  |
-| `AutoZygote`               | ✅                  |
+|                            | [`Constant`](@ref) | [`Cache`](@ref) |
+| -------------------------- | ------------------ | --------------- |
+| `AutoChainRules`           | ✅                  | ❌               |
+| `AutoDiffractor`           | ❌                  | ❌               |
+| `AutoEnzyme` (forward)     | ✅                  | ✅               |
+| `AutoEnzyme` (reverse)     | ✅                  | ✅               |
+| `AutoFastDifferentiation`  | ✅                  | ✅               |
+| `AutoFiniteDiff`           | ✅                  | ✅               |
+| `AutoFiniteDifferences`    | ✅                  | ✅               |
+| `AutoForwardDiff`          | ✅                  | ✅               |
+| `AutoGTPSA`                | ✅                  | ❌               |
+| `AutoMooncake`             | ✅                  | ❌               |
+| `AutoPolyesterForwardDiff` | ✅                  | ✅               |
+| `AutoReverseDiff`          | ✅                  | ❌               |
+| `AutoSymbolics`            | ✅                  | ❌               |
+| `AutoTracker`              | ✅                  | ❌               |
+| `AutoZygote`               | ✅                  | ❌               |
 
 ## Second order
 

DifferentiationInterface/ext/DifferentiationInterfaceFastDifferentiationExt/DifferentiationInterfaceFastDifferentiationExt.jl

@@ -18,14 +18,23 @@
 
 DI.check_available(::AutoFastDifferentiation) = true
 
-monovec(x::Number) = [x]
-
-myvec(x::Number) = monovec(x)
+myvec(x::Number) = [x]
 myvec(x::AbstractArray) = vec(x)
 
+variablize(::Number, name::Symbol) = only(make_variables(name))
+variablize(x::AbstractArray, name::Symbol) = make_variables(name, size(x)...)
+
+function variablize(contexts::NTuple{C,DI.Context}) where {C}
+    map(enumerate(contexts)) do (k, c)
+        variablize(DI.unwrap(c), Symbol("context$k"))
+    end
+end
+
 dense_ad(backend::AutoFastDifferentiation) = backend
 dense_ad(backend::AutoSparse{<:AutoFastDifferentiation}) = ADTypes.dense_ad(backend)
 
+myvec_unwrap(x) = myvec(DI.unwrap(x))
+
 include("onearg.jl")
 include("twoarg.jl")