From 4a58f30aa083d2f5753465a36730ca7734b483be Mon Sep 17 00:00:00 2001
From: Chip Kerchner <chip.kerchner@ibm.com>
Date: Tue, 31 Jan 2023 19:40:24 +0000
Subject: [PATCH] Fix pre-POWER8_VECTOR bugs in pcmp_lt and pnegate and
 reactivate psqrt.

---
 Eigen/src/Core/arch/AltiVec/MathFunctions.h | 44 +++++++++++----------
 Eigen/src/Core/arch/AltiVec/PacketMath.h    | 21 +++++++++-
 2 files changed, 43 insertions(+), 22 deletions(-)

diff --git a/Eigen/src/Core/arch/AltiVec/MathFunctions.h b/Eigen/src/Core/arch/AltiVec/MathFunctions.h
index ae75787c2..45761e217 100644
--- a/Eigen/src/Core/arch/AltiVec/MathFunctions.h
+++ b/Eigen/src/Core/arch/AltiVec/MathFunctions.h
@@ -61,26 +61,6 @@ Packet4f patan<Packet4f>(const Packet4f& _x)
 }
 
 #ifdef EIGEN_VECTORIZE_VSX
-#if !EIGEN_COMP_CLANG
-template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet4f prsqrt<Packet4f>(const Packet4f& x)
-{
-  return  vec_rsqrt(x);
-}
-
-template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet2d prsqrt<Packet2d>(const Packet2d& x)
-{
-  return  vec_rsqrt(x);
-}
-
-template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet2d patan<Packet2d>(const Packet2d& _x)
-{
-  return patan_double(_x);
-}
-#endif
-
 template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet4f psqrt<Packet4f>(const Packet4f& x)
 {
@@ -93,6 +73,30 @@ Packet2d psqrt<Packet2d>(const Packet2d& x)
   return  vec_sqrt(x);
 }
 
+#if !EIGEN_COMP_CLANG
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet4f prsqrt<Packet4f>(const Packet4f& x)
+{
+  return pset1<Packet4f>(1.0f) / psqrt<Packet4f>(x);
+//  vec_rsqrt returns different results from the generic version
+//  return  vec_rsqrt(x);
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet2d prsqrt<Packet2d>(const Packet2d& x)
+{
+  return pset1<Packet2d>(1.0) / psqrt<Packet2d>(x);
+//  vec_rsqrt returns different results from the generic version
+//  return  vec_rsqrt(x);
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet2d patan<Packet2d>(const Packet2d& _x)
+{
+  return patan_double(_x);
+}
+#endif
+
 template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet2d pexp<Packet2d>(const Packet2d& _x)
 {
diff --git a/Eigen/src/Core/arch/AltiVec/PacketMath.h b/Eigen/src/Core/arch/AltiVec/PacketMath.h
index 96ca9d655..fa12892fe 100644
--- a/Eigen/src/Core/arch/AltiVec/PacketMath.h
+++ b/Eigen/src/Core/arch/AltiVec/PacketMath.h
@@ -175,6 +175,11 @@ struct packet_traits<float> : default_packet_traits {
     HasExp = 1,
 #ifdef EIGEN_VECTORIZE_VSX
     HasSqrt = 1,
+#if !EIGEN_COMP_CLANG
+    HasRsqrt = 1,
+#else
+    HasRsqrt = 0,
+#endif
     HasTanh = EIGEN_FAST_MATH,
     HasErf = EIGEN_FAST_MATH,
     HasRint = 1,
@@ -215,6 +220,11 @@ struct packet_traits<bfloat16> : default_packet_traits {
     HasExp = 1,
 #ifdef EIGEN_VECTORIZE_VSX
     HasSqrt = 1,
+#if !EIGEN_COMP_CLANG
+    HasRsqrt = 1,
+#else
+    HasRsqrt = 0,
+#endif
     HasRint = 1,
 #else
     HasSqrt = 0,
@@ -992,7 +1002,7 @@ template<> EIGEN_STRONG_INLINE Packet4f pnegate(const Packet4f& a)
 #ifdef __POWER8_VECTOR__
   return vec_neg(a);
 #else
-  return p4f_ZERO - a;
+  return vec_xor(a, p4f_MZERO);
 #endif
 }
 template<> EIGEN_STRONG_INLINE Packet4i pnegate(const Packet4i& a)
@@ -1086,7 +1096,10 @@ template<> EIGEN_STRONG_INLINE Packet16c pmax<Packet16c>(const Packet16c& a, con
 template<> EIGEN_STRONG_INLINE Packet16uc pmax<Packet16uc>(const Packet16uc& a, const Packet16uc& b) { return vec_max(a, b); }
 
 template<> EIGEN_STRONG_INLINE Packet4f pcmp_le(const Packet4f& a, const Packet4f& b) { return reinterpret_cast<Packet4f>(vec_cmple(a,b)); }
+// To fix bug with vec_cmplt on older versions
+#if defined(__POWER8_VECTOR__) || EIGEN_COMP_LLVM
 template<> EIGEN_STRONG_INLINE Packet4f pcmp_lt(const Packet4f& a, const Packet4f& b) { return reinterpret_cast<Packet4f>(vec_cmplt(a,b)); }
+#endif
 template<> EIGEN_STRONG_INLINE Packet4f pcmp_eq(const Packet4f& a, const Packet4f& b) { return reinterpret_cast<Packet4f>(vec_cmpeq(a,b)); }
 template<> EIGEN_STRONG_INLINE Packet4f pcmp_lt_or_nan(const Packet4f& a, const Packet4f& b) {
   Packet4f c = reinterpret_cast<Packet4f>(vec_cmpge(a,b));
@@ -2700,7 +2713,11 @@ template<> struct packet_traits<double> : default_packet_traits
     HasLog  = 0,
     HasExp  = 1,
     HasSqrt = 1,
+#if !EIGEN_COMP_CLANG
+    HasRsqrt = 1,
+#else
     HasRsqrt = 0,
+#endif
     HasRound = 1,
     HasFloor = 1,
     HasCeil = 1,
@@ -2806,7 +2823,7 @@ template<> EIGEN_STRONG_INLINE Packet2d pnegate(const Packet2d& a)
 #ifdef __POWER8_VECTOR__
   return vec_neg(a);
 #else
-  return p2d_ZERO - a;
+  return vec_xor(a, p2d_MZERO);
 #endif
 }