Cache support with ForwardDiff (#590)

* Cache support with ForwardDiff

* Fix rewrap

Author: gdalle

DifferentiationInterface/Project.toml

@@ -1,7 +1,7 @@
 name = "DifferentiationInterface"
 uuid = "a0c0ee7d-e4b9-4e03-894e-1c5f64a51d63"
 authors = ["Guillaume Dalle", "Adrian Hill"]
-version = "0.6.15"
+version = "0.6.16"
 
 [deps]
 ADTypes = "47edcb42-4c32-4615-8424-f2b9edc5f35b"

DifferentiationInterface/docs/src/explanation/advanced.md

@@ -19,7 +19,7 @@ Right now, there are two kinds of context: [`Constant`](@ref) and [`Cache`](@ref
     This feature is still experimental and will not be supported by all backends.
     At the moment:
     - `Constant` is supported by all backends except symbolic ones
-    - `Cache` is only supported by finite difference backends
+    - `Cache` is only supported by finite difference backends and [`AutoForwardDiff`](@ref), but it is not yet optimized
 
 Semantically, both of these calls compute the partial gradient of `f(x, c)` with respect to `x`, but they consider `c` differently:
 

DifferentiationInterface/docs/src/tutorials/advanced.md

@@ -49,6 +49,8 @@ prep_other_constant = prepare_gradient(f_multiarg, backend, x, Constant(-1))
 gradient(f_multiarg, prep_other_constant, backend, x, Constant(10))
 ```
 
+For additional arguments which act as mutated buffers, the [`Cache`](@ref) wrapper is the appropriate choice instead of [`Constant`](@ref).
+
 ## Sparsity
 
 Sparse AD is very useful when Jacobian or Hessian matrices have a lot of zeros.

DifferentiationInterface/ext/DifferentiationInterfaceForwardDiffExt/DifferentiationInterfaceForwardDiffExt.jl

@@ -5,6 +5,8 @@ using Base: Fix1, Fix2
 import DifferentiationInterface as DI
 using DifferentiationInterface:
     BatchSizeSettings,
+    Cache,
+    Constant,
     Context,
     DerivativePrep,
     DifferentiateWith,