From 7c2b4b15490e58ffb9ebd35f94d0838dd98d99d5 Mon Sep 17 00:00:00 2001
From: Chris Elrod <elrodc@gmail.com>
Date: Tue, 5 Oct 2021 21:22:23 -0400
Subject: [PATCH] Use SIMDDualNumbers for forward diff special functions

---
 Project.toml                                 |  4 ++-
 src/simdfunctionals/vmap_grad_forwarddiff.jl | 30 +-------------------
 test/grouptests.jl                           |  3 +-
 3 files changed, 6 insertions(+), 31 deletions(-)

diff --git a/Project.toml b/Project.toml
index ce3ce3fcb..a622e7542 100644
--- a/Project.toml
+++ b/Project.toml
@@ -1,7 +1,7 @@
 name = "LoopVectorization"
 uuid = "bdcacae8-1622-11e9-2a5c-532679323890"
 authors = ["Chris Elrod <elrodc@gmail.com>"]
-version = "0.12.82"
+version = "0.12.83"
 
 [deps]
 ArrayInterface = "4fba245c-0d91-5ea0-9b3e-6abc04ee57a9"
@@ -15,6 +15,7 @@ LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 OffsetArrays = "6fe1bfb0-de20-5000-8ca7-80f57d26f881"
 PolyesterWeave = "1d0040c9-8b98-4ee7-8388-3f51789ca0ad"
 Requires = "ae029012-a4dd-5104-9daa-d747884805df"
+SIMDDualNumbers = "3cdde19b-5bb0-4aaf-8931-af3e248e098b"
 SLEEFPirates = "476501e8-09a2-5ece-8869-fb82de89a1fa"
 Static = "aedffcd0-7271-4cad-89d0-dc628f76c6d3"
 ThreadingUtilities = "8290d209-cae3-49c0-8002-c8c24d57dab5"
@@ -32,6 +33,7 @@ LayoutPointers = "0.1.2"
 OffsetArrays = "1.4.1"
 PolyesterWeave = "0.1"
 Requires = "1"
+SIMDDualNumbers = "0.1"
 SLEEFPirates = "0.6.23"
 Static = "0.3.3"
 ThreadingUtilities = "0.4.5"
diff --git a/src/simdfunctionals/vmap_grad_forwarddiff.jl b/src/simdfunctionals/vmap_grad_forwarddiff.jl
index 867141666..be064dfe9 100644
--- a/src/simdfunctionals/vmap_grad_forwarddiff.jl
+++ b/src/simdfunctionals/vmap_grad_forwarddiff.jl
@@ -1,34 +1,6 @@
 import .ForwardDiff
+using SIMDDualNumbers
 
-@generated function SLEEFPirates.tanh_fast(x::ForwardDiff.Dual{T,S,N}) where {T,S,N}
-  quote
-    $(Expr(:meta,:inline))
-    t = tanh_fast(x.value)
-    ∂t = vfnmadd_fast(t, t, one(S))
-    p = x.partials
-    ForwardDiff.Dual{T}(t, ForwardDiff.Partials(Base.Cartesian.@ntuple $N n -> mul_fast(∂t, p[n])))
-  end
-end
-@generated function SLEEFPirates.sigmoid_fast(x::ForwardDiff.Dual{T,S,N}) where {T,S,N}
-  quote
-    $(Expr(:meta,:inline))
-    s = sigmoid_fast(x.value)
-    ∂s = vfnmadd_fast(s,s,s)
-    p = x.partials
-    ForwardDiff.Dual{T}(s, ForwardDiff.Partials(Base.Cartesian.@ntuple $N n -> mul_fast(∂s, p[n])))
-  end
-end
-@generated function VectorizationBase.relu(x::ForwardDiff.Dual{T,S,N}) where {T,S,N}
-  quote
-    $(Expr(:meta,:inline))
-    v = x.value
-    z = zero(v)
-    cmp = v < z
-    r = ifelse(cmp, z, v)
-    p = x.partials
-    ForwardDiff.Dual{T}(r, ForwardDiff.Partials(Base.Cartesian.@ntuple $N n -> ifelse(cmp, z, p[n])))
-  end
-end
 @generated function init_dual(v::Tuple{Vararg{AbstractSIMD,A}}) where {A}
   res = Expr(:tuple)
   q = Expr(:block, Expr(:meta,:inline))
diff --git a/test/grouptests.jl b/test/grouptests.jl
index a1e1677eb..a7f16339c 100644
--- a/test/grouptests.jl
+++ b/test/grouptests.jl
@@ -12,7 +12,8 @@ const START_TIME = time()
 
   @time if LOOPVECTORIZATION_TEST == "all" || LOOPVECTORIZATION_TEST == "part2"
     using Aqua
-    @time Aqua.test_all(LoopVectorization, ambiguities = VERSION ≥ v"1.6")
+    @time Aqua.test_all(LoopVectorization, ambiguities = VERSION ≥ v"1.6", stale_deps = false)
+    @time Aqua.test_stale_deps(LoopVectorization, ignore = [:SIMDDualNumbers])
     # @test isempty(detect_unbound_args(LoopVectorization))
 
     @time include("printmethods.jl")