Merge pull request #18 from tlycken/gradient-evaluation

Gradient evaluation

src/Interpolations.jl

@@ -3,7 +3,12 @@ module Interpolations
 using Base.Cartesian
 using Compat
 
-import Base: size, eltype, getindex, ndims
+import Base:
+    eltype,
+    gradient,
+    getindex,
+    ndims,
+    size
 
 export 
     Interpolation,
@@ -152,9 +157,31 @@ for IT in (
                 ret
             end
         ))
+
+        eval(ngenerate(
+            :N,
+            :(Array{promote_type(T,typeof(x)...),1}),
+            :(gradient!{T,N}(g::Array{T,1}, itp::Interpolation{T,N,$IT,$EB}, x::NTuple{N,Real}...)),
+            N->quote
+                $(extrap_transform_x(gr,eb,N))
+                $(define_indices(it,N))
+                @nexprs $N dim->begin
+                    @nexprs $N d->begin
+                        (d==dim
+                            ? $(gradient_coefficients(it,N,:d))
+                            : $(coefficients(it,N,:d)))
+                    end
+
+                    @inbounds g[dim] = $(index_gen(degree(it),N))
+                end
+                g
+            end
+        ))
     end
 end
 
+gradient{T}(itp::Interpolation{T}, x...) = gradient!(Array(T,ndims(itp)), itp, x...)
+
 # This creates prefilter specializations for all interpolation types that need them
 for IT in (
         Quadratic{Flat,OnCell},

src/constant.jl

@@ -6,8 +6,18 @@ function define_indices(::Constant, N)
     :(@nexprs $N d->(ix_d = clamp(round(Int,x_d), 1, size(itp,d))))
 end
 
-function coefficients(::Constant, N)
-    :(@nexprs $N d->(c_d = one(typeof(x_d))))
+function coefficients(c::Constant, N)
+    :(@nexprs $N d->($(coefficients(c, N, :d))))
+end
+
+function coefficients(::Constant, N, d)
+    sym, symx = symbol(string("c_",d)), symbol(string("x_",d))
+    :($sym = one(typeof($symx)))
+end
+
+function gradient_coefficients(::Constant, N, d)
+    sym, symx = symbol(string("c_",d)), symbol(string("x_",d))
+    :($sym = zero(typeof($symx)))
 end
 
 function index_gen(degree::ConstantDegree, N::Integer, offsets...)

src/linear.jl

@@ -12,12 +12,23 @@ function define_indices(::Linear, N)
     end
 end
 
-function coefficients(::Linear, N)
+function coefficients(l::Linear, N)
+    :(@nexprs $N d->($(coefficients(l, N, :d))))
+end
+
+function coefficients(::Linear, N, d)
+    sym, symp, symfx = symbol(string("c_",d)), symbol(string("cp_",d)), symbol(string("fx_",d))
     quote
-        @nexprs $N d->begin
-            c_d = one(typeof(fx_d)) - fx_d
-            cp_d = fx_d
-        end
+        $sym = one(typeof($symfx)) - $symfx
+        $symp = $symfx
+    end
+end
+
+function gradient_coefficients(::Linear,N,d)
+    sym, symp, symfx = symbol(string("c_",d)), symbol(string("cp_",d)), symbol(string("fx_",d))
+    quote
+        $sym = -one(typeof($symfx))
+        $symp = one(typeof($symfx))
     end
 end
 

src/quadratic.jl

@@ -28,13 +28,27 @@ function define_indices(q::Quadratic{Periodic}, N)
     end
 end
 
-function coefficients(::Quadratic, N)
+function coefficients(q::Quadratic, N)
+    :(@nexprs $N d->($(coefficients(q, N, :d))))
+end
+
+function coefficients(q::Quadratic, N, d)
+    symm, sym, symp =  symbol(string("cm_",d)), symbol(string("c_",d)), symbol(string("cp_",d))
+    symfx = symbol(string("fx_",d))
     quote
-        @nexprs $N d->begin
-            cm_d = .5 * (fx_d-.5)^2
-            c_d = .75 - fx_d^2
-            cp_d = .5 * (fx_d+.5)^2
-        end
+        $symm = .5 * ($symfx - .5)^2
+        $sym  = .75 - $symfx^2
+        $symp = .5 * ($symfx + .5)^2
+    end
+end
+
+function gradient_coefficients(q::Quadratic, N, d)
+    symm, sym, symp =  symbol(string("cm_",d)), symbol(string("c_",d)), symbol(string("cp_",d))
+    symfx = symbol(string("fx_",d))
+    quote
+        $symm = $symfx-.5
+        $sym = -2*$symfx
+        $symp = $symfx+.5
     end
 end
 

test/gradient.jl

@@ -0,0 +1,30 @@
+module GradientTests
+println("Testing gradient evaluation")
+using Base.Test, Interpolations
+
+nx = 10
+f1(x) = sin((x-3)*2pi/(nx-1) - 1)
+g1(x) = 2pi/(nx-1) * cos((x-3)*2pi/(nx-1) - 1)
+
+# Gradient of Constant should always be 0
+itp1 = Interpolation(Float64[f1(x) for x in 1:nx-1],
+            Constant(OnGrid()), ExtrapPeriodic())
+for x in 1:nx
+    @test gradient(itp1, x)[1] == 0
+end
+
+# Since Linear is OnGrid in the domain, check the gradients between grid points
+itp1 = Interpolation(Float64[f1(x) for x in 1:nx-1],
+            Linear(OnGrid()), ExtrapPeriodic())
+for x in 2.5:nx-1.5
+    @test_approx_eq_eps g1(x) gradient(itp1, x)[1] abs(.1*g1(x))
+end
+
+# Since Quadratic is OnCell in the domain, check gradients at grid points
+itp1 = Interpolation(Float64[f1(x) for x in 1:nx-1], 
+            Quadratic(Periodic(),OnGrid()), ExtrapPeriodic())
+for x in 2:nx-1
+    @test_approx_eq_eps g1(x) gradient(itp1, x)[1] abs(.05*g1(x))
+end
+
+end

test/runtests.jl

@@ -12,6 +12,9 @@ include("quadratic.jl")
 # indices inbounds in A.
 include("on-grid.jl")
 
+# test gradient evaluation
+include("gradient.jl")
+
 # Tests copied from Grid.jl's old test suite
 #include("grid.jl")