Add special functions to Eigen: lgamma, erf, erfc.
Includes CUDA support and unit tests.
diff --git a/Eigen/src/Core/arch/CUDA/MathFunctions.h b/Eigen/src/Core/arch/CUDA/MathFunctions.h
index 3bea88b..ecd5c44 100644
--- a/Eigen/src/Core/arch/CUDA/MathFunctions.h
+++ b/Eigen/src/Core/arch/CUDA/MathFunctions.h
@@ -66,6 +66,43 @@
return make_double2(rsqrt(a.x), rsqrt(a.y));
}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+float4 plgamma<float4>(const float4& a)
+{
+ return make_float4(lgammaf(a.x), lgammaf(a.y), lgammaf(a.z), lgammaf(a.w));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+double2 plgamma<double2>(const double2& a)
+{
+ return make_double2(lgamma(a.x), lgamma(a.y));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+float4 perf<float4>(const float4& a)
+{
+ return make_float4(erf(a.x), erf(a.y), erf(a.z), erf(a.w));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+double2 perf<double2>(const double2& a)
+{
+ return make_double2(erf(a.x), erf(a.y));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+float4 perfc<float4>(const float4& a)
+{
+ return make_float4(erfc(a.x), erfc(a.y), erfc(a.z), erfc(a.w));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+double2 perfc<double2>(const double2& a)
+{
+ return make_double2(erfc(a.x), erfc(a.y));
+}
+
+
#endif
} // end namespace internal
diff --git a/Eigen/src/Core/arch/CUDA/PacketMath.h b/Eigen/src/Core/arch/CUDA/PacketMath.h
index 0d2c2fe..cb1b547 100644
--- a/Eigen/src/Core/arch/CUDA/PacketMath.h
+++ b/Eigen/src/Core/arch/CUDA/PacketMath.h
@@ -39,6 +39,9 @@
HasExp = 1,
HasSqrt = 1,
HasRsqrt = 1,
+ HasLGamma = 1,
+ HasErf = 1,
+ HasErfc = 1,
HasBlend = 0,
};
@@ -59,6 +62,9 @@
HasExp = 1,
HasSqrt = 1,
HasRsqrt = 1,
+ HasLGamma = 1,
+ HasErf = 1,
+ HasErfc = 1,
HasBlend = 0,
};
