From d87648a6bea315645b893c3815ca8c6bb00ec5d2 Mon Sep 17 00:00:00 2001
From: Jakub Lichman <jlichman@tachyum.com>
Date: Sun, 25 Apr 2021 20:58:56 +0000
Subject: Tests added and AVX512 bug fixed for pcmp_lt_or_nan

---
 Eigen/src/Core/arch/AVX512/PacketMath.h |  4 ++--
 test/packetmath.cpp                     | 32 ++++++++++++++++++++++++++++++++
 2 files changed, 34 insertions(+), 2 deletions(-)

diff --git a/Eigen/src/Core/arch/AVX512/PacketMath.h b/Eigen/src/Core/arch/AVX512/PacketMath.h
index 7d3362f48..59bbef0d1 100644
--- a/Eigen/src/Core/arch/AVX512/PacketMath.h
+++ b/Eigen/src/Core/arch/AVX512/PacketMath.h
@@ -487,7 +487,7 @@ template<> EIGEN_STRONG_INLINE Packet16f pcmp_lt(const Packet16f& a, const Packe
 }
 
 template<> EIGEN_STRONG_INLINE Packet16f pcmp_lt_or_nan(const Packet16f& a, const Packet16f& b) {
-  __mmask16 mask = _mm512_cmp_ps_mask(a, b, _CMP_NGT_UQ);
+  __mmask16 mask = _mm512_cmp_ps_mask(a, b, _CMP_NGE_UQ);
   return _mm512_castsi512_ps(
       _mm512_mask_set1_epi32(_mm512_set1_epi32(0), mask, 0xffffffffu));
 }
@@ -518,7 +518,7 @@ EIGEN_STRONG_INLINE Packet8d pcmp_lt(const Packet8d& a, const Packet8d& b) {
 }
 template <>
 EIGEN_STRONG_INLINE Packet8d pcmp_lt_or_nan(const Packet8d& a, const Packet8d& b) {
-  __mmask8 mask = _mm512_cmp_pd_mask(a, b, _CMP_NGT_UQ);
+  __mmask8 mask = _mm512_cmp_pd_mask(a, b, _CMP_NGE_UQ);
   return _mm512_castsi512_pd(
       _mm512_mask_set1_epi64(_mm512_set1_epi64(0), mask, 0xffffffffffffffffu));
 }
diff --git a/test/packetmath.cpp b/test/packetmath.cpp
index 79e91c819..0bb511d5a 100644
--- a/test/packetmath.cpp
+++ b/test/packetmath.cpp
@@ -279,6 +279,36 @@ void packetmath_boolean_mask_ops() {
   CHECK_CWISE2_IF(true, internal::pcmp_eq, internal::pcmp_eq);
 }
 
+template <typename Scalar, typename Packet>
+void packetmath_boolean_mask_ops_real() {
+  const int PacketSize = internal::unpacket_traits<Packet>::size;
+  const int size = 2 * PacketSize;
+  EIGEN_ALIGN_MAX Scalar data1[size];
+  EIGEN_ALIGN_MAX Scalar data2[size];
+  EIGEN_ALIGN_MAX Scalar ref[size];
+
+  for (int i = 0; i < PacketSize; ++i) {
+    data1[i] = internal::random<Scalar>();
+    data1[i + PacketSize] = internal::random<bool>() ? data1[i] : Scalar(0);
+  }
+
+  CHECK_CWISE2_IF(true, internal::pcmp_lt_or_nan, internal::pcmp_lt_or_nan);
+
+  //Test (-0) <=/< (0) for signed operations
+  for (int i = 0; i < PacketSize; ++i) {
+    data1[i] = Scalar(-0.0);
+    data1[i + PacketSize] = internal::random<bool>() ? data1[i] : Scalar(0);
+  }
+  CHECK_CWISE2_IF(true, internal::pcmp_lt_or_nan, internal::pcmp_lt_or_nan);
+
+  //Test NaN
+  for (int i = 0; i < PacketSize; ++i) {
+    data1[i] = NumTraits<Scalar>::quiet_NaN();
+    data1[i + PacketSize] = internal::random<bool>() ? data1[i] : Scalar(0);
+  }
+  CHECK_CWISE2_IF(true, internal::pcmp_lt_or_nan, internal::pcmp_lt_or_nan);
+}
+
 template <typename Scalar, typename Packet>
 void packetmath_boolean_mask_ops_notcomplex() {
   const int PacketSize = internal::unpacket_traits<Packet>::size;
@@ -609,6 +639,8 @@ void packetmath_real() {
   CHECK_CWISE1_EXACT_IF(PacketTraits::HasCeil, numext::ceil, internal::pceil);
   CHECK_CWISE1_EXACT_IF(PacketTraits::HasFloor, numext::floor, internal::pfloor);
   CHECK_CWISE1_EXACT_IF(PacketTraits::HasRint, numext::rint, internal::print);
+
+  packetmath_boolean_mask_ops_real<Scalar,Packet>();
   
   // Rounding edge cases.
   if (PacketTraits::HasRound || PacketTraits::HasCeil || PacketTraits::HasFloor || PacketTraits::HasRint) {
-- 
cgit v1.2.3