errors

Project.toml

@@ -4,8 +4,10 @@ authors = ["Sungho Shin <sshin@anl.gov>"]
 version = "0.4.0"
 
 [deps]
+ForwardDiff = "f6369f11-7733-5829-9624-2563aa707210"
 NLPModels = "a4795742-8479-5a88-8948-cc11e1c8c1a6"
 SolverCore = "ff4d7338-4cf1-434d-91df-b86cb86fb843"
+SpecialFunctions = "276daf66-3868-5448-9aa4-cd146d93841b"
 
 [weakdeps]
 AMDGPU = "21141c5a-9bdb-4563-92ae-f87d6854732e"
@@ -34,6 +36,7 @@ oneAPI = "1"
 [extras]
 CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
 Downloads = "f43a241f-c20a-4ad4-852c-f6b1247861c6"
+ForwardDiff = "f6369f11-7733-5829-9624-2563aa707210"
 JuMP = "4076af6c-e467-56ae-b986-b466b2749572"
 KernelAbstractions = "63c18a36-062a-441e-b654-da1e3ab1ce7c"
 MadNLP = "2621e9c9-9eb4-46b1-8089-e8c72242dfb6"
@@ -42,8 +45,8 @@ NLPModelsIpopt = "f4238b75-b362-5c4c-b852-0801c9a21d71"
 NLPModelsJuMP = "792afdf1-32c1-5681-94e0-d7bf7a5df49e"
 Percival = "01435c0c-c90d-11e9-3788-63660f8fbccc"
 PowerModels = "c36e90e8-916a-50a6-bd94-075b64ef4655"
-Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
+Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [targets]
-test = ["Test", "NLPModels", "NLPModelsIpopt", "KernelAbstractions", "CUDA", "AMDGPU", "oneAPI", "MadNLP", "Percival", "PowerModels", "JuMP", "NLPModelsJuMP", "Downloads", "Random"]
+test = ["Test", "NLPModels", "NLPModelsIpopt", "KernelAbstractions", "CUDA", "AMDGPU", "oneAPI", "MadNLP", "Percival", "PowerModels", "JuMP", "NLPModelsJuMP", "Downloads", "Random", "ForwardDiff", "SpecialFunctions"]

src/gradient.jl

@@ -40,7 +40,7 @@ function gradient!(y, f, x, adj)
     return y
 end
 
-function gradient!(y, f, x, p = nothing, adj = one(eltype(y)))
+function gradient!(y, f, x, p, adj)
     graph = f(p, AdjointNodeSource(x))
     drpass(graph, y, adj)
     return y

src/graph.jl

@@ -13,7 +13,7 @@ abstract type AbstractSecondAdjointNode end
 A source of variable nodes
 
 """
-struct VarSource end
+struct VarSource <: AbstractNode end
 
 """
     Var{I}
@@ -33,7 +33,7 @@ end
 A source of parameterized data
 
 """
-struct ParSource end
+struct ParSource <: AbstractNode end
 
 """
     ParIndexed{I, J}

src/jacobian.jl

@@ -126,28 +126,29 @@ Performs sparse jacobian evalution
 """
 function sjacobian!(y1, y2, f, x, adj)
     @simd for i in eachindex(f.itr)
-        @inbounds jrpass(
-            f.f.f(f.itr[i], AdjointNodeSource(x)),
-            f.f.comp1,
-            offset0(f, i),
+        @inbounds sjacobian!(
             y1,
             y2,
+            f.f.f,
+            f.itr[i],
+            x,
+            f.f.comp1,
+            offset0(f, i),
             offset1(f, i),
-            0,
             adj,
         )
     end
 end
 
-function sjacobian!(y1, y2, f, p, x, comp, adj)
+function sjacobian!(y1, y2, f, p, x, comp, o0, o1, adj)
     graph = f(p, AdjointNodeSource(x))
     jrpass(
         graph,
         comp,
-        offset0(f, i),
+        o0,
         y1,
         y2,
-        offset1(f, i),
+        o1,
         0,
         adj,
     )

src/simdfunction.jl

@@ -41,7 +41,6 @@ function SIMDFunction(gen::Base.Generator, o0 = 0, o1 = 0, o2 = 0)
     p = Par(eltype(gen.iter))
     f = gen.f(p)
 
-
     d = f(Identity(), AdjointNodeSource(nothing))
     y1 = []
     ExaModels.grpass(d, nothing, y1, nothing, 0, NaN16)

test/ADTest/ADTest.jl

@@ -1,59 +1,82 @@
 module ADTest
 
 using ExaModels
-using Random, Test, ForwardDiff
-using KernelAbstractions, CUDA, AMDGPU, oneAPI
+using Random, Test, ForwardDiff, SpecialFunctions
 
 Random.seed!(0)
 
-const BACKENDS = Any[nothing, CPU()]
-
-if CUDA.has_cuda()
-    push!(BACKENDS, CUDABackend())
-    @info "testing CUDA"
-end
-
-if AMDGPU.has_rocm_gpu()
-    push!(BACKENDS, ROCBackend())
-    @info "testing AMDGPU"
-end
-
-try
-    oneAPI.oneL0.zeInit(0)
-    push!(BACKENDS, oneAPIBackend())
-    push!(EXCLUDE2, ("percival", oneAPIBackend()))
-    @info "testing oneAPI"
-catch e
-end
-
-
 const FUNCTIONS = [
     ("function-test-1-1",x->beta(erf(x[1]/x[2]/3.0)+3.0*x[2],erf(x[9])^2)), 
     ("function-test-1-2",x->0*x[1]), 
-    ("function-test-1-3",x->beta(cos(log(inv(inv(x[1])))),erfc(tanh(0*x[1])))), 
+    ("function-test-1-3",x->beta(cos(log(abs2(inv(inv(x[1])))+1.)),erfc(tanh(0*x[1])))), 
     ("function-test-1-4",x->(0*x[1]^x[3]^1.0+x[1])/x[9]/x[10]), 
-    ("function-test-1-5",x->(x[1]+1.)^x[2]*log(x[3])/tanh(x[2])), 
+    ("function-test-1-5",x->exp(x[1]+1.)^x[2]*log(abs2(x[3])+3)/tanh(x[2])), 
     ("function-test-1-6",x->beta(2*logbeta(x[1],x[5]),beta(x[2],x[3]))), 
     ("function-test-1-7",x->besselj0(exp(erf(-x[1])))), 
-    ("function-test-1-8",x->erfc((x[1]^2/x[2])^x[9]/x[10])), 
+    ("function-test-1-8",x->erfc(abs2(x[1]^2/x[2])^x[9]/x[10])), 
     ("function-test-1-9",x->erfc(x[1])^erf(2.5x[2])), 
     ("function-test-1-10",x->sin(1/x[1])), 
-    ("function-test-1-13",x->exp(x[2])/cos(x[1])^2+sin(x[1]^2)), 
-    ("function-test-1-14",x->airyai(exp(x[1]+x[2]*2.0^8))), 
-    ("function-test-1-16",x->sin(x[9]inv(x[1])-x[8]inv(x[2]))), 
-    ("function-test-1-19",x->x[1]/log(x[2]^2+9.)), 
-    ("function-test-1-21",x->beta(beta(tan(beta(x[1],1)+2.0),cos(sin(x[2]))),x[3])), 
-    ("function-test-1-24",x->beta(cos(beta(beta(x[1]^9,x[2]),x[2]*x[3])),sin(x[2]*x[3]/2.0)/1.0)),
+    ("function-test-1-11",x->exp(x[2])/cos(x[1])^2+sin(x[1]^2)), 
+    ("function-test-1-12",x->sin(x[9]inv(x[1])-x[8]inv(x[2]))), 
+    ("function-test-1-13",x->x[1]/log(x[2]^2+9.)), 
+    ("function-test-1-14",x->beta(beta(tan(beta(x[1],1)+2.0),cos(sin(x[2]))),x[3])), 
+    ("function-test-1-15",x->beta(cos(beta(beta(x[1]^9,x[2]),x[2]*x[3])),sin(x[2]*x[3]/2.0)/1.0)),
 ]
 
 function gradient(f, x)
-    y = similar(y)
+    T = eltype(x)
+    y = fill!(similar(x), zero(T))
+    ExaModels.gradient!(y, (p,x)->f(x), x, nothing, one(T))
+    return y
 end
 
 function sgradient(f, x)
+    T = eltype(x)
+
+    ff = f(ExaModels.VarSource())
+    d = ff(ExaModels.Identity(), ExaModels.AdjointNodeSource(nothing))
+    y1 = []
+    ExaModels.grpass(d, nothing, y1, nothing, 0, NaN16)
+
+    a1 = unique(y1)
+    comp = ExaModels.Compressor(Tuple(findfirst(isequal(i), a1) for i in y1))
+
+    n = length(a1)
+    buffer = similar(x, n)
+    buffer_I = similar(x, Tuple{Int,Int}, n)
+
+    ExaModels.sgradient!(buffer_I, ff, nothing, nothing, comp, 0, NaN32)
+    ExaModels.sgradient!(buffer, ff, nothing, x, comp, 0, one(T))
+
+    y = zeros(length(x))
+    y[collect(i for (i,j) in buffer_I)] = buffer
+
+    return y
 end
 
 function sjacobian(f, x)
+    T = eltype(x)
+
+    ff = f(ExaModels.VarSource())
+    d = ff(ExaModels.Identity(), ExaModels.AdjointNodeSource(nothing))
+    y1 = []
+    ExaModels.grpass(d, nothing, y1, nothing, 0, NaN16)
+
+    a1 = unique(y1)
+    comp = ExaModels.Compressor(Tuple(findfirst(isequal(i), a1) for i in y1))
+
+    n = length(a1)
+    buffer = similar(x, n)
+    buffer_I = similar(x, Int, n)
+    buffer_J = similar(x, Int, n)
+
+    ExaModels.sjacobian!(buffer_I, buffer_J, ff, nothing, nothing, comp, 0, 0, NaN32)
+    ExaModels.sjacobian!(buffer_I, buffer_J, ff, nothing, x, comp, 0, 0, one(T))
+
+    y = zeros(length(x))
+    y[collect(i for (i,j) in buffer_J)] = buffer
+
+    return y
 end
 
 function shessian(f, x)
@@ -64,7 +87,10 @@ function runtests()
         for (name, f) in FUNCTIONS
             x0 = randn(10)
             @testset "$name" begin
-                @test gradient(f, x0) ≈ ForwardDiff.gradient(f, x0) atol=1e-6
+                g = ForwardDiff.gradient(f, x0)
+                @test gradient(f, x0) ≈ g atol=1e-6
+                @test sgradient(f, x0) ≈ g atol=1e-6
+                @test sjacobian(f, x0) ≈ g atol=1e-6
             end
         end
     end

test/runtests.jl

@@ -7,5 +7,5 @@ include("ADTest/ADTest.jl")
 
 @testset "ExaModels test" begin
     ADTest.runtests()
-    NLPTest.runtests()
+    # NLPTest.runtests()
 end