From c785464430bfc697debe3f8d49e49064aa08e0a3 Mon Sep 17 00:00:00 2001
From: Gael Guennebaud <g.gael@free.fr>
Date: Fri, 30 Nov 2018 16:21:33 +0100
Subject: [PATCH] Add packet sin and cos to Altivec/VSX and NEON

---
 Eigen/src/Core/arch/AltiVec/MathFunctions.h | 12 ++++++++++++
 Eigen/src/Core/arch/AltiVec/PacketMath.h    |  5 +++--
 Eigen/src/Core/arch/NEON/MathFunctions.h    | 20 ++++++++++++++++----
 Eigen/src/Core/arch/NEON/PacketMath.h       |  6 ++++--
 4 files changed, 35 insertions(+), 8 deletions(-)

diff --git a/Eigen/src/Core/arch/AltiVec/MathFunctions.h b/Eigen/src/Core/arch/AltiVec/MathFunctions.h
index acf665018..81097e668 100644
--- a/Eigen/src/Core/arch/AltiVec/MathFunctions.h
+++ b/Eigen/src/Core/arch/AltiVec/MathFunctions.h
@@ -30,6 +30,18 @@ Packet4f pexp<Packet4f>(const Packet4f& _x)
   return pexp_float(_x);
 }
 
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f psin<Packet4f>(const Packet4f& _x)
+{
+  return psin_float(_x);
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f pcos<Packet4f>(const Packet4f& _x)
+{
+  return pcos_float(_x);
+}
+
 #ifndef EIGEN_COMP_CLANG
 template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
 Packet4f prsqrt<Packet4f>(const Packet4f& x)
diff --git a/Eigen/src/Core/arch/AltiVec/PacketMath.h b/Eigen/src/Core/arch/AltiVec/PacketMath.h
index 446065fb7..d0ee93f4a 100755
--- a/Eigen/src/Core/arch/AltiVec/PacketMath.h
+++ b/Eigen/src/Core/arch/AltiVec/PacketMath.h
@@ -146,8 +146,8 @@ template<> struct packet_traits<float>  : default_packet_traits
     HasMin  = 1,
     HasMax  = 1,
     HasAbs  = 1,
-    HasSin  = 0,
-    HasCos  = 0,
+    HasSin  = EIGEN_FAST_MATH,
+    HasCos  = EIGEN_FAST_MATH,
     HasLog  = 1,
     HasExp  = 1,
 #ifdef __VSX__
@@ -437,6 +437,7 @@ template<> EIGEN_STRONG_INLINE Packet4f pcmp_lt_or_nan(const Packet4f& a, const
   Packet4f c = reinterpret_cast<Packet4f>(vec_cmpge(a,b));
   return vec_nor(c,c);
 }
+template<> EIGEN_STRONG_INLINE Packet4i pcmp_eq(const Packet4i& a, const Packet4i& b) { return reinterpret_cast<Packet4i>(vec_cmpeq(a,b)); }
 
 template<> EIGEN_STRONG_INLINE Packet4f pand<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_and(a, b); }
 template<> EIGEN_STRONG_INLINE Packet4i pand<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_and(a, b); }
diff --git a/Eigen/src/Core/arch/NEON/MathFunctions.h b/Eigen/src/Core/arch/NEON/MathFunctions.h
index addaacb9a..2e7d0e944 100644
--- a/Eigen/src/Core/arch/NEON/MathFunctions.h
+++ b/Eigen/src/Core/arch/NEON/MathFunctions.h
@@ -15,15 +15,27 @@ namespace Eigen {
 namespace internal {
 
 template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
-Packet4f pexp<Packet4f>(const Packet4f& _x)
+Packet4f pexp<Packet4f>(const Packet4f& x)
 {
-  return pexp_float(_x);
+  return pexp_float(x);
 }
 
 template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
-Packet4f plog<Packet4f>(const Packet4f& _x)
+Packet4f plog<Packet4f>(const Packet4f& x)
 {
-  return plog_float(_x);
+  return plog_float(x);
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f psin<Packet4f>(const Packet4f& x)
+{
+  return psin_float(x);
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f pcos<Packet4f>(const Packet4f& x)
+{
+  return pcos_float(x);
 }
 
 } // end namespace internal
diff --git a/Eigen/src/Core/arch/NEON/PacketMath.h b/Eigen/src/Core/arch/NEON/PacketMath.h
index 0a50153c7..ed3cec88a 100644
--- a/Eigen/src/Core/arch/NEON/PacketMath.h
+++ b/Eigen/src/Core/arch/NEON/PacketMath.h
@@ -111,8 +111,8 @@ template<> struct packet_traits<float>  : default_packet_traits
     HasDiv   = 1,
     HasFloor = 1,
     // FIXME check the Has*
-    HasSin  = 0,
-    HasCos  = 0,
+    HasSin  = EIGEN_FAST_MATH,
+    HasCos  = EIGEN_FAST_MATH,
     HasLog  = 1,
     HasExp  = 1,
     HasSqrt = 0
@@ -268,6 +268,8 @@ template<> EIGEN_STRONG_INLINE Packet4f pcmp_lt(const Packet4f& a, const Packet4
 template<> EIGEN_STRONG_INLINE Packet4f pcmp_eq(const Packet4f& a, const Packet4f& b) { return vreinterpretq_f32_u32(vceqq_f32(a,b)); }
 template<> EIGEN_STRONG_INLINE Packet4f pcmp_lt_or_nan(const Packet4f& a, const Packet4f& b) { return vreinterpretq_f32_u32(vmvnq_u32(vcgeq_f32(a,b))); }
 
+template<> EIGEN_STRONG_INLINE Packet4i pcmp_eq(const Packet4i& a, const Packet4i& b) { return vreinterpretq_s32_u32(vceqq_s32(a,b)); }
+
 template<> EIGEN_STRONG_INLINE Packet4f pfloor<Packet4f>(const Packet4f& a)
 {
   const Packet4f cst_1 = pset1<Packet4f>(1.0f);