From 90323f1751ca170490b5a7590410ef5d05d62ac5 Mon Sep 17 00:00:00 2001
From: Gael Guennebaud <g.gael@free.fr>
Date: Wed, 4 Nov 2015 22:15:57 +0100
Subject: [PATCH] Fix AVX round/ceil/floor, and fix respective unit test

---
 Eigen/src/Core/arch/AVX/PacketMath.h | 12 ++++++------
 test/packetmath.cpp                  | 12 ++++++------
 2 files changed, 12 insertions(+), 12 deletions(-)

diff --git a/Eigen/src/Core/arch/AVX/PacketMath.h b/Eigen/src/Core/arch/AVX/PacketMath.h
index b313fb09a..717ae67c5 100644
--- a/Eigen/src/Core/arch/AVX/PacketMath.h
+++ b/Eigen/src/Core/arch/AVX/PacketMath.h
@@ -182,14 +182,14 @@ template<> EIGEN_STRONG_INLINE Packet4d pmin<Packet4d>(const Packet4d& a, const
 template<> EIGEN_STRONG_INLINE Packet8f pmax<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_max_ps(a,b); }
 template<> EIGEN_STRONG_INLINE Packet4d pmax<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_max_pd(a,b); }
 
-template<> EIGEN_STRONG_INLINE Packet4f pround<Packet8f>(const Packet8f& a) { return _mm256_round_ps(a, 0); }
-template<> EIGEN_STRONG_INLINE Packet2d pround<Packet4d>(const Packet4d& a) { return _mm256_round_pd(a, 0); }
+template<> EIGEN_STRONG_INLINE Packet8f pround<Packet8f>(const Packet8f& a) { return _mm256_round_ps(a, _MM_FROUND_CUR_DIRECTION); }
+template<> EIGEN_STRONG_INLINE Packet4d pround<Packet4d>(const Packet4d& a) { return _mm256_round_pd(a, _MM_FROUND_CUR_DIRECTION); }
 
-template<> EIGEN_STRONG_INLINE Packet4f pceil<Packet8f>(const Packet8f& a) { return _mm256_ceil_ps(a); }
-template<> EIGEN_STRONG_INLINE Packet2d pceil<Packet4d>(const Packet4d& a) { return _mm256_ceil_pd(a); }
+template<> EIGEN_STRONG_INLINE Packet8f pceil<Packet8f>(const Packet8f& a) { return _mm256_ceil_ps(a); }
+template<> EIGEN_STRONG_INLINE Packet4d pceil<Packet4d>(const Packet4d& a) { return _mm256_ceil_pd(a); }
 
-template<> EIGEN_STRONG_INLINE Packet4f pfloor<Packet8f>(const Packet8f& a) { return _mm256_floor_ps(a); }
-template<> EIGEN_STRONG_INLINE Packet2d pfloor<Packet4d>(const Packet4d& a) { return _mm256_floor_pd(a); }
+template<> EIGEN_STRONG_INLINE Packet8f pfloor<Packet8f>(const Packet8f& a) { return _mm256_floor_ps(a); }
+template<> EIGEN_STRONG_INLINE Packet4d pfloor<Packet4d>(const Packet4d& a) { return _mm256_floor_pd(a); }
 
 template<> EIGEN_STRONG_INLINE Packet8f pand<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_and_ps(a,b); }
 template<> EIGEN_STRONG_INLINE Packet4d pand<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_and_pd(a,b); }
diff --git a/test/packetmath.cpp b/test/packetmath.cpp
index dea648002..b6616ac5e 100644
--- a/test/packetmath.cpp
+++ b/test/packetmath.cpp
@@ -29,7 +29,7 @@ template<typename Scalar> bool areApproxAbs(const Scalar* a, const Scalar* b, in
   {
     if (!isApproxAbs(a[i],b[i],refvalue))
     {
-      std::cout << "[" << Map<const Matrix<Scalar,1,Dynamic> >(a,size) << "]" << " != " << Map<const Matrix<Scalar,1,Dynamic> >(b,size) << "\n";
+      std::cout << "ref: [" << Map<const Matrix<Scalar,1,Dynamic> >(a,size) << "]" << " != vec: [" << Map<const Matrix<Scalar,1,Dynamic> >(b,size) << "]\n";
       return false;
     }
   }
@@ -42,7 +42,7 @@ template<typename Scalar> bool areApprox(const Scalar* a, const Scalar* b, int s
   {
     if (a[i]!=b[i] && !internal::isApprox(a[i],b[i]))
     {
-      std::cout << "[" << Map<const Matrix<Scalar,1,Dynamic> >(a,size) << "]" << " != " << Map<const Matrix<Scalar,1,Dynamic> >(b,size) << "\n";
+      std::cout << "ref: [" << Map<const Matrix<Scalar,1,Dynamic> >(a,size) << "]" << " != vec: [" << Map<const Matrix<Scalar,1,Dynamic> >(b,size) << "]\n";
       return false;
     }
   }
@@ -296,10 +296,6 @@ template<typename Scalar> void packetmath_real()
   EIGEN_ALIGN_MAX Scalar data2[PacketTraits::size*4];
   EIGEN_ALIGN_MAX Scalar ref[PacketTraits::size*4];
 
-  CHECK_CWISE1_IF(PacketTraits::HasRound, std::round, internal::pround);
-  CHECK_CWISE1_IF(PacketTraits::HasCeil, std::ceil, internal::pceil);
-  CHECK_CWISE1_IF(PacketTraits::HasFloor, std::floor, internal::pfloor);
-
   for (int i=0; i<size; ++i)
   {
     data1[i] = internal::random<Scalar>(-1,1) * std::pow(Scalar(10), internal::random<Scalar>(-3,3));
@@ -308,6 +304,10 @@ template<typename Scalar> void packetmath_real()
   CHECK_CWISE1_IF(PacketTraits::HasSin, std::sin, internal::psin);
   CHECK_CWISE1_IF(PacketTraits::HasCos, std::cos, internal::pcos);
   CHECK_CWISE1_IF(PacketTraits::HasTan, std::tan, internal::ptan);
+
+  CHECK_CWISE1_IF(PacketTraits::HasRound, numext::round, internal::pround);
+  CHECK_CWISE1_IF(PacketTraits::HasCeil, numext::ceil, internal::pceil);
+  CHECK_CWISE1_IF(PacketTraits::HasFloor, numext::floor, internal::pfloor);
   
   for (int i=0; i<size; ++i)
   {