From 81c27325ae3b5b8cbc72762f74ecb7b82cd031f5 Mon Sep 17 00:00:00 2001
From: Gael Guennebaud <g.gael@free.fr>
Date: Sat, 8 Dec 2018 14:27:48 +0100
Subject: [PATCH] bug #1641: fix testing of pandnot and fix pandnot for complex
 on SSE/AVX/AVX512

---
 Eigen/src/Core/arch/AVX/Complex.h    |  4 ++--
 Eigen/src/Core/arch/AVX512/Complex.h |  4 ++--
 Eigen/src/Core/arch/SSE/Complex.h    |  4 ++--
 test/packetmath.cpp                  | 26 ++++++++++++++++++++------
 4 files changed, 26 insertions(+), 12 deletions(-)

diff --git a/Eigen/src/Core/arch/AVX/Complex.h b/Eigen/src/Core/arch/AVX/Complex.h
index 2bb40fc79..08d021b65 100644
--- a/Eigen/src/Core/arch/AVX/Complex.h
+++ b/Eigen/src/Core/arch/AVX/Complex.h
@@ -72,7 +72,7 @@ template<> EIGEN_STRONG_INLINE Packet4cf pmul<Packet4cf>(const Packet4cf& a, con
 template<> EIGEN_STRONG_INLINE Packet4cf pand   <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_and_ps(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet4cf por    <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_or_ps(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet4cf pxor   <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_xor_ps(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet4cf pandnot<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_andnot_ps(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet4cf pandnot<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_andnot_ps(b.v,a.v)); }
 
 template<> EIGEN_STRONG_INLINE Packet4cf pload <Packet4cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet4cf(pload<Packet8f>(&numext::real_ref(*from))); }
 template<> EIGEN_STRONG_INLINE Packet4cf ploadu<Packet4cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet4cf(ploadu<Packet8f>(&numext::real_ref(*from))); }
@@ -279,7 +279,7 @@ template<> EIGEN_STRONG_INLINE Packet2cd pmul<Packet2cd>(const Packet2cd& a, con
 template<> EIGEN_STRONG_INLINE Packet2cd pand   <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_and_pd(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet2cd por    <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_or_pd(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet2cd pxor   <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_xor_pd(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cd pandnot<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_andnot_pd(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cd pandnot<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_andnot_pd(b.v,a.v)); }
 
 template<> EIGEN_STRONG_INLINE Packet2cd pload <Packet2cd>(const std::complex<double>* from)
 { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cd(pload<Packet4d>((const double*)from)); }
diff --git a/Eigen/src/Core/arch/AVX512/Complex.h b/Eigen/src/Core/arch/AVX512/Complex.h
index 8750b07de..247f89860 100644
--- a/Eigen/src/Core/arch/AVX512/Complex.h
+++ b/Eigen/src/Core/arch/AVX512/Complex.h
@@ -78,7 +78,7 @@ template<> EIGEN_STRONG_INLINE Packet8cf pmul<Packet8cf>(const Packet8cf& a, con
 template<> EIGEN_STRONG_INLINE Packet8cf pand   <Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(_mm512_and_ps(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet8cf por    <Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(_mm512_or_ps(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet8cf pxor   <Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(_mm512_xor_ps(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet8cf pandnot<Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(_mm512_andnot_ps(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet8cf pandnot<Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(_mm512_andnot_ps(b.v,a.v)); }
 
 template<> EIGEN_STRONG_INLINE Packet8cf pload <Packet8cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet8cf(pload<Packet16f>(&numext::real_ref(*from))); }
 template<> EIGEN_STRONG_INLINE Packet8cf ploadu<Packet8cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet8cf(ploadu<Packet16f>(&numext::real_ref(*from))); }
@@ -265,7 +265,7 @@ template<> EIGEN_STRONG_INLINE Packet4cd pmul<Packet4cd>(const Packet4cd& a, con
 template<> EIGEN_STRONG_INLINE Packet4cd pand   <Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(_mm512_and_pd(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet4cd por    <Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(_mm512_or_pd(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet4cd pxor   <Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(_mm512_xor_pd(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet4cd pandnot<Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(_mm512_andnot_pd(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet4cd pandnot<Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(_mm512_andnot_pd(b.v,a.v)); }
 
 template<> EIGEN_STRONG_INLINE Packet4cd pload <Packet4cd>(const std::complex<double>* from)
 { EIGEN_DEBUG_ALIGNED_LOAD return Packet4cd(pload<Packet8d>((const double*)from)); }
diff --git a/Eigen/src/Core/arch/SSE/Complex.h b/Eigen/src/Core/arch/SSE/Complex.h
index d075043ce..0f8960328 100644
--- a/Eigen/src/Core/arch/SSE/Complex.h
+++ b/Eigen/src/Core/arch/SSE/Complex.h
@@ -85,7 +85,7 @@ template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, con
 template<> EIGEN_STRONG_INLINE Packet2cf pand   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_and_ps(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet2cf por    <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_or_ps(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet2cf pxor   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_xor_ps(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_andnot_ps(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_andnot_ps(b.v,a.v)); }
 
 template<> EIGEN_STRONG_INLINE Packet2cf pload <Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(pload<Packet4f>(&numext::real_ref(*from))); }
 template<> EIGEN_STRONG_INLINE Packet2cf ploadu<Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cf(ploadu<Packet4f>(&numext::real_ref(*from))); }
@@ -308,7 +308,7 @@ template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, con
 template<> EIGEN_STRONG_INLINE Packet1cd pand   <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_and_pd(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet1cd por    <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_or_pd(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet1cd pxor   <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_xor_pd(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet1cd pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_andnot_pd(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_andnot_pd(b.v,a.v)); }
 
 // FIXME force unaligned load, this is a temporary fix
 template<> EIGEN_STRONG_INLINE Packet1cd pload <Packet1cd>(const std::complex<double>* from)
diff --git a/test/packetmath.cpp b/test/packetmath.cpp
index 60c9dbc36..916b37bef 100644
--- a/test/packetmath.cpp
+++ b/test/packetmath.cpp
@@ -64,6 +64,10 @@ struct bit_andnot{
   operator()(T a, T b) const { return a & (~b); }
 };
 EIGEN_TEST_MAKE_BITWISE(andnot, bit_andnot())
+template<typename T>
+bool biteq(T a, T b) {
+  return (bits(a) == bits(b)).all();
+}
 
 }
 }
@@ -92,7 +96,7 @@ template<typename Scalar> bool areApprox(const Scalar* a, const Scalar* b, int s
 {
   for (int i=0; i<size; ++i)
   {
-    if (a[i]!=b[i] && !internal::isApprox(a[i],b[i]))
+    if ((!internal::biteq(a[i],b[i])) && a[i]!=b[i] && !internal::isApprox(a[i],b[i]))
     {
       std::cout << "ref: [" << Map<const Matrix<Scalar,1,Dynamic> >(a,size) << "]" << " != vec: [" << Map<const Matrix<Scalar,1,Dynamic> >(b,size) << "]\n";
       return false;
@@ -344,11 +348,6 @@ template<typename Scalar,typename Packet> void packetmath()
     }
   }
 
-  CHECK_CWISE2_IF(true, internal::por, internal::por);
-  CHECK_CWISE2_IF(true, internal::pxor, internal::pxor);
-  CHECK_CWISE2_IF(true, internal::pand, internal::pand);
-  CHECK_CWISE2_IF(true, internal::pandnot, internal::pandnot);
-
   if (PacketTraits::HasBlend) {
     Packet thenPacket = internal::pload<Packet>(data1);
     Packet elsePacket = internal::pload<Packet>(data2);
@@ -384,6 +383,21 @@ template<typename Scalar,typename Packet> void packetmath()
     internal::pstore(data2, internal::pinsertlast(internal::pload<Packet>(data1),s));
     VERIFY(areApprox(ref, data2, PacketSize) && "internal::pinsertlast");
   }
+
+  {
+    for (int i=0; i<PacketSize; ++i)
+    {
+      data1[i] = internal::random<Scalar>();
+      unsigned char v = internal::random<bool>() ? 0xff : 0;
+      char* bytes = (char*)(data1+PacketSize+i);
+      for(int k=0; k<int(sizeof(Scalar)); ++k)
+        bytes[k] = v;
+    }
+    CHECK_CWISE2_IF(true, internal::por, internal::por);
+    CHECK_CWISE2_IF(true, internal::pxor, internal::pxor);
+    CHECK_CWISE2_IF(true, internal::pand, internal::pand);
+    CHECK_CWISE2_IF(true, internal::pandnot, internal::pandnot);
+  }
 }
 
 template<typename Scalar,typename Packet> void packetmath_real()