Add generic PacketMath implementation of the Error Function (erf).

14 files changed, 156 insertions, 70 deletions

diff --git a/Eigen/src/Core/arch/AVX/MathFunctions.h b/Eigen/src/Core/arch/AVX/MathFunctions.h
index c6d3cf6a0..089a682ba 100644
--- a/Eigen/src/Core/arch/AVX/MathFunctions.h
+++ b/Eigen/src/Core/arch/AVX/MathFunctions.h
@@ -62,6 +62,14 @@ ptanh<Packet8f>(const Packet8f& x) {
   return internal::generic_fast_tanh_float(x);
 }
 
+// Error function.
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
+perf<Packet8f>(const Packet8f& x) {
+  return internal::generic_fast_erf_float(x);
+}
+
+// Exponential function for doubles.
 template <>
 EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4d
 pexp<Packet4d>(const Packet4d& x) {
diff --git a/Eigen/src/Core/arch/AVX/PacketMath.h b/Eigen/src/Core/arch/AVX/PacketMath.h
index 0472e9850..ed8acfc61 100644
--- a/Eigen/src/Core/arch/AVX/PacketMath.h
+++ b/Eigen/src/Core/arch/AVX/PacketMath.h
@@ -62,21 +62,22 @@ template<> struct packet_traits<float>  : default_packet_traits
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
-    size=8,
+    size = 8,
     HasHalfPacket = 1,
 
-    HasDiv  = 1,
-    HasSin  = EIGEN_FAST_MATH,
-    HasCos  = EIGEN_FAST_MATH,
-    HasLog  = 1,
-    HasLog1p  = 1,
-    HasExpm1  = 1,
-    HasExp  = 1,
+    HasDiv = 1,
+    HasSin = EIGEN_FAST_MATH,
+    HasCos = EIGEN_FAST_MATH,
+    HasLog = 1,
+    HasLog1p = 1,
+    HasExpm1 = 1,
+    HasExp = 1,
     HasNdtri = 1,
-    HasBessel  = 1,
+    HasBessel = 1,
     HasSqrt = 1,
     HasRsqrt = 1,
-    HasTanh  = EIGEN_FAST_MATH,
+    HasTanh = EIGEN_FAST_MATH,
+    HasErf = EIGEN_FAST_MATH,
     HasBlend = 1,
     HasRound = 1,
     HasFloor = 1,
diff --git a/Eigen/src/Core/arch/AVX512/MathFunctions.h b/Eigen/src/Core/arch/AVX512/MathFunctions.h
index 9e37a720b..66581964c 100644
--- a/Eigen/src/Core/arch/AVX512/MathFunctions.h
+++ b/Eigen/src/Core/arch/AVX512/MathFunctions.h
@@ -405,6 +405,18 @@ Packet16f pexpm1<Packet16f>(const Packet16f& _x) {
 }
 #endif
 
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
+ptanh<Packet16f>(const Packet16f& _x) {
+  return internal::generic_fast_tanh_float(_x);
+}
+
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
+perf<Packet16f>(const Packet16f& _x) {
+  return internal::generic_fast_erf_float(_x);
+}
+
 }  // end namespace internal
 
 }  // end namespace Eigen
diff --git a/Eigen/src/Core/arch/AVX512/PacketMath.h b/Eigen/src/Core/arch/AVX512/PacketMath.h
index 589ccbb7a..19c03cfe2 100644
--- a/Eigen/src/Core/arch/AVX512/PacketMath.h
+++ b/Eigen/src/Core/arch/AVX512/PacketMath.h
@@ -98,11 +98,13 @@ template<> struct packet_traits<float>  : default_packet_traits
     HasLog1p  = 1,
     HasExpm1  = 1,
     HasNdtri = 1,
-#endif
     HasBessel  = 1,
+#endif
     HasExp = 1,
     HasSqrt = EIGEN_FAST_MATH,
     HasRsqrt = EIGEN_FAST_MATH,
+    HasTanh = EIGEN_FAST_MATH,
+    HasErf = EIGEN_FAST_MATH,
 #endif
     HasDiv = 1
   };
diff --git a/Eigen/src/Core/arch/AltiVec/MathFunctions.h b/Eigen/src/Core/arch/AltiVec/MathFunctions.h
index b6d1f47f2..00a0248e2 100644
--- a/Eigen/src/Core/arch/AltiVec/MathFunctions.h
+++ b/Eigen/src/Core/arch/AltiVec/MathFunctions.h
@@ -76,6 +76,20 @@ Packet2d pexp<Packet2d>(const Packet2d& _x)
 }
 #endif
 
+// Hyperbolic Tangent function.
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4f
+ptanh<Packet4f>(const Packet4f& x) {
+  return internal::generic_fast_tanh_float(x);
+}
+
+// Error function.
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4f
+perf<Packet4f>(const Packet4f& x) {
+  return internal::generic_fast_erf_float(x);
+}
+
 }  // end namespace internal
 
 }  // end namespace Eigen
diff --git a/Eigen/src/Core/arch/AltiVec/PacketMath.h b/Eigen/src/Core/arch/AltiVec/PacketMath.h
index 7ee290a29..2bba3a798 100755
--- a/Eigen/src/Core/arch/AltiVec/PacketMath.h
+++ b/Eigen/src/Core/arch/AltiVec/PacketMath.h
@@ -120,27 +120,27 @@ static Packet16uc p16uc_COMPLEX32_REV2 = vec_sld(p16uc_PSET64_HI, p16uc_PSET64_L
   #define EIGEN_PPC_PREFETCH(ADDR) asm( "   dcbt [%[addr]]\n" :: [addr] "r" (ADDR) : "cc" );
 #endif
 
-template<> struct packet_traits<float>  : default_packet_traits
-{
+template <>
+struct packet_traits<float> : default_packet_traits {
   typedef Packet4f type;
   typedef Packet4f half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
-    size=4,
+    size = 4,
     HasHalfPacket = 1,
 
-    HasAdd  = 1,
-    HasSub  = 1,
-    HasMul  = 1,
-    HasDiv  = 1,
-    HasMin  = 1,
-    HasMax  = 1,
-    HasAbs  = 1,
-    HasSin  = EIGEN_FAST_MATH,
-    HasCos  = EIGEN_FAST_MATH,
-    HasLog  = 1,
-    HasExp  = 1,
+    HasAdd = 1,
+    HasSub = 1,
+    HasMul = 1,
+    HasDiv = 1,
+    HasMin = 1,
+    HasMax = 1,
+    HasAbs = 1,
+    HasSin = EIGEN_FAST_MATH,
+    HasCos = EIGEN_FAST_MATH,
+    HasLog = 1,
+    HasExp = 1,
 #ifdef __VSX__
     HasSqrt = 1,
 #if !EIGEN_COMP_CLANG
@@ -151,6 +151,8 @@ template<> struct packet_traits<float>  : default_packet_traits
 #else
     HasSqrt = 0,
     HasRsqrt = 0,
+    HasTanh = EIGEN_FAST_MATH,
+    HasErf = EIGEN_FAST_MATH,
 #endif
     HasRound = 1,
     HasFloor = 1,
@@ -159,8 +161,8 @@ template<> struct packet_traits<float>  : default_packet_traits
     HasBlend = 1
   };
 };
-template<> struct packet_traits<int>    : default_packet_traits
-{
+template <>
+struct packet_traits<int> : default_packet_traits {
   typedef Packet4i type;
   typedef Packet4i half;
   enum {
@@ -177,7 +179,6 @@ template<> struct packet_traits<int>    : default_packet_traits
   };
 };
 
-
 template<> struct unpacket_traits<Packet4f>
 {
   typedef float     type;
diff --git a/Eigen/src/Core/arch/MSA/MathFunctions.h b/Eigen/src/Core/arch/MSA/MathFunctions.h
index f5181b90e..23e4b8664 100644
--- a/Eigen/src/Core/arch/MSA/MathFunctions.h
+++ b/Eigen/src/Core/arch/MSA/MathFunctions.h
@@ -321,6 +321,13 @@ pcos<Packet4f>(const Packet4f& x) {
   return psincos_inner_msa_float</* sine */ false>(x);
 }
 
+// Error function.
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4f
+perf<Packet4f>(const Packet4f& x) {
+  return internal::generic_fast_erf_float(x);
+}
+
 template <>
 EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet2d
 pexp<Packet2d>(const Packet2d& _x) {
diff --git a/Eigen/src/Core/arch/MSA/PacketMath.h b/Eigen/src/Core/arch/MSA/PacketMath.h
index 3c922d4de..94ee0e302 100644
--- a/Eigen/src/Core/arch/MSA/PacketMath.h
+++ b/Eigen/src/Core/arch/MSA/PacketMath.h
@@ -88,6 +88,7 @@ struct packet_traits<float> : default_packet_traits {
     HasSin = EIGEN_FAST_MATH,
     HasCos = EIGEN_FAST_MATH,
     HasTanh = EIGEN_FAST_MATH,
+    HasErf = EIGEN_FAST_MATH,
     HasLog = 1,
     HasExp = 1,
     HasSqrt = 1,
diff --git a/Eigen/src/Core/arch/NEON/MathFunctions.h b/Eigen/src/Core/arch/NEON/MathFunctions.h
index fdee9f9b4..13e3f4989 100644
--- a/Eigen/src/Core/arch/NEON/MathFunctions.h
+++ b/Eigen/src/Core/arch/NEON/MathFunctions.h
@@ -36,6 +36,20 @@ Packet4f pcos<Packet4f>(const Packet4f& x)
   return pcos_float(x);
 }
 
+// Hyperbolic Tangent function.
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4f
+ptanh<Packet4f>(const Packet4f& x) {
+  return internal::generic_fast_tanh_float(x);
+}
+
+// Error function.
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4f
+perf<Packet4f>(const Packet4f& x) {
+  return internal::generic_fast_erf_float(x);
+}
+
 } // end namespace internal
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/arch/NEON/PacketMath.h b/Eigen/src/Core/arch/NEON/PacketMath.h
index 84153ece1..83447547c 100644
--- a/Eigen/src/Core/arch/NEON/PacketMath.h
+++ b/Eigen/src/Core/arch/NEON/PacketMath.h
@@ -98,28 +98,30 @@ typedef uint32x4_t  Packet4ui;
   #define EIGEN_ARM_PREFETCH(ADDR)
 #endif
 
-template<> struct packet_traits<float>  : default_packet_traits
-{
+template <>
+struct packet_traits<float> : default_packet_traits {
   typedef Packet4f type;
-  typedef Packet4f half; // Packet2f intrinsics not implemented yet
+  typedef Packet4f half;  // Packet2f intrinsics not implemented yet
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
     size = 4,
-    HasHalfPacket=0, // Packet2f intrinsics not implemented yet
-   
-    HasDiv   = 1,
+    HasHalfPacket = 0,  // Packet2f intrinsics not implemented yet
+
+    HasDiv = 1,
     HasFloor = 1,
     // FIXME check the Has*
-    HasSin  = EIGEN_FAST_MATH,
-    HasCos  = EIGEN_FAST_MATH,
-    HasLog  = 1,
-    HasExp  = 1,
-    HasSqrt = 0
+    HasSin = EIGEN_FAST_MATH,
+    HasCos = EIGEN_FAST_MATH,
+    HasLog = 1,
+    HasExp = 1,
+    HasSqrt = 0,
+    HasTanh = EIGEN_FAST_MATH,
+    HasErf = EIGEN_FAST_MATH
   };
 };
-template<> struct packet_traits<int32_t>    : default_packet_traits
-{
+template <>
+struct packet_traits<int32_t> : default_packet_traits {
   typedef Packet4i type;
   typedef Packet4i half; // Packet2i intrinsics not implemented yet
   enum {
diff --git a/Eigen/src/Core/arch/SSE/MathFunctions.h b/Eigen/src/Core/arch/SSE/MathFunctions.h
index b21bb93bf..a991b1a39 100644
--- a/Eigen/src/Core/arch/SSE/MathFunctions.h
+++ b/Eigen/src/Core/arch/SSE/MathFunctions.h
@@ -147,6 +147,13 @@ ptanh<Packet4f>(const Packet4f& x) {
   return internal::generic_fast_tanh_float(x);
 }
 
+// Error function.
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4f
+perf<Packet4f>(const Packet4f& a) {
+  return internal::generic_fast_erf_float(a);
+}
+
 } // end namespace internal
 
 namespace numext {
diff --git a/Eigen/src/Core/arch/SSE/PacketMath.h b/Eigen/src/Core/arch/SSE/PacketMath.h
index b48d70afd..4a772d078 100755
--- a/Eigen/src/Core/arch/SSE/PacketMath.h
+++ b/Eigen/src/Core/arch/SSE/PacketMath.h
@@ -96,28 +96,29 @@ template<> struct is_arithmetic<__m128d> { enum { value = true }; };
 // Use the packet_traits defined in AVX/PacketMath.h instead if we're going
 // to leverage AVX instructions.
 #ifndef EIGEN_VECTORIZE_AVX
-template<> struct packet_traits<float>  : default_packet_traits
-{
+template <>
+struct packet_traits<float> : default_packet_traits {
   typedef Packet4f type;
   typedef Packet4f half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
-    size=4,
+    size = 4,
     HasHalfPacket = 0,
 
-    HasDiv  = 1,
-    HasSin  = EIGEN_FAST_MATH,
-    HasCos  = EIGEN_FAST_MATH,
-    HasLog  = 1,
-    HasLog1p  = 1,
-    HasExpm1  = 1,
+    HasDiv = 1,
+    HasSin = EIGEN_FAST_MATH,
+    HasCos = EIGEN_FAST_MATH,
+    HasLog = 1,
+    HasLog1p = 1,
+    HasExpm1 = 1,
     HasNdtri = 1,
-    HasExp  = 1,
-    HasBessel  = 1,
+    HasExp = 1,
+    HasBessel = 1,
     HasSqrt = 1,
     HasRsqrt = 1,
-    HasTanh  = EIGEN_FAST_MATH,
+    HasTanh = EIGEN_FAST_MATH,
+    HasErf = EIGEN_FAST_MATH,
     HasBlend = 1,
     HasFloor = 1
 
@@ -128,8 +129,8 @@ template<> struct packet_traits<float>  : default_packet_traits
 #endif
   };
 };
-template<> struct packet_traits<double> : default_packet_traits
-{
+template <>
+struct packet_traits<double> : default_packet_traits {
   typedef Packet2d type;
   typedef Packet2d half;
   enum {
diff --git a/Eigen/src/Core/arch/ZVector/MathFunctions.h b/Eigen/src/Core/arch/ZVector/MathFunctions.h
index ff33a975f..d4b33b56f 100644
--- a/Eigen/src/Core/arch/ZVector/MathFunctions.h
+++ b/Eigen/src/Core/arch/ZVector/MathFunctions.h
@@ -225,6 +225,20 @@ Packet4f prsqrt<Packet4f>(const Packet4f& x) {
   return res;
 }
 
+// Hyperbolic Tangent function.
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4f
+ptanh<Packet4f>(const Packet4f& x) {
+  return internal::generic_fast_tanh_float(x);
+}
+
+// Error function.
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4f
+perf<Packet4f>(const Packet4f& x) {
+  return internal::generic_fast_erf_float(x);
+}
+
 }  // end namespace internal
 
 }  // end namespace Eigen
diff --git a/Eigen/src/Core/arch/ZVector/PacketMath.h b/Eigen/src/Core/arch/ZVector/PacketMath.h
index b1c788168..5036106cd 100755
--- a/Eigen/src/Core/arch/ZVector/PacketMath.h
+++ b/Eigen/src/Core/arch/ZVector/PacketMath.h
@@ -173,33 +173,35 @@ template<> struct packet_traits<int>    : default_packet_traits
   };
 };
 
-template<> struct packet_traits<float> : default_packet_traits
-{
+template <>
+struct packet_traits<float> : default_packet_traits {
   typedef Packet4f type;
   typedef Packet4f half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
-    size=4,
+    size = 4,
     HasHalfPacket = 0,
 
-    HasAdd  = 1,
-    HasSub  = 1,
-    HasMul  = 1,
-    HasDiv  = 1,
-    HasMin  = 1,
-    HasMax  = 1,
-    HasAbs  = 1,
-    HasSin  = 0,
-    HasCos  = 0,
-    HasLog  = 0,
+    HasAdd = 1,
+    HasSub = 1,
+    HasMul = 1,
+    HasDiv = 1,
+    HasMin = 1,
+    HasMax = 1,
+    HasAbs = 1,
+    HasSin = 0,
+    HasCos = 0,
+    HasLog = 0,
 #if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ >= 12)
-    HasExp  = 0,
+    HasExp = 0,
 #else
-    HasExp  = 1,
+    HasExp = 1,
 #endif
     HasSqrt = 1,
     HasRsqrt = 1,
+    HasTanh = 1,
+    HasErf = 1,
     HasRound = 1,
     HasFloor = 1,
     HasCeil = 1,