Merged in aavenel/eigen (pull request PR-142)

Add round, ceil and floor for SSE4.1/AVX (Bug #70)

2 files changed, 42 insertions, 4 deletions

diff --git a/Eigen/src/Core/arch/AVX/PacketMath.h b/Eigen/src/Core/arch/AVX/PacketMath.h
index dfdb71abd..b313fb09a 100644
--- a/Eigen/src/Core/arch/AVX/PacketMath.h
+++ b/Eigen/src/Core/arch/AVX/PacketMath.h
@@ -66,7 +66,10 @@ template<> struct packet_traits<float>  : default_packet_traits
     HasExp  = 1,
     HasSqrt = 1,
     HasRsqrt = 1,
-    HasBlend = 1
+    HasBlend = 1,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1
   };
 };
 template<> struct packet_traits<double> : default_packet_traits
@@ -83,7 +86,10 @@ template<> struct packet_traits<double> : default_packet_traits
     HasExp  = 1,
     HasSqrt = 1,
     HasRsqrt = 1,
-    HasBlend = 1
+    HasBlend = 1,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1
   };
 };
 
@@ -176,6 +182,15 @@ 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 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 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 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/Eigen/src/Core/arch/SSE/PacketMath.h b/Eigen/src/Core/arch/SSE/PacketMath.h
index 3c30b2cf0..3fcb1c138 100755
--- a/Eigen/src/Core/arch/SSE/PacketMath.h
+++ b/Eigen/src/Core/arch/SSE/PacketMath.h
@@ -109,7 +109,13 @@ template<> struct packet_traits<float>  : default_packet_traits
     HasExp  = 1,
     HasSqrt = 1,
     HasRsqrt = 1,
-    HasBlend = 1
+    HasBlend = 1,
+
+#ifdef EIGEN_VECTORIZE_SSE4_1
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1
+#endif
   };
 };
 template<> struct packet_traits<double> : default_packet_traits
@@ -126,7 +132,13 @@ template<> struct packet_traits<double> : default_packet_traits
     HasExp  = 1,
     HasSqrt = 1,
     HasRsqrt = 1,
-    HasBlend = 1
+    HasBlend = 1,
+
+#ifdef EIGEN_VECTORIZE_SSE4_1
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1
+#endif
   };
 };
 #endif
@@ -256,6 +268,17 @@ template<> EIGEN_STRONG_INLINE Packet4i pmax<Packet4i>(const Packet4i& a, const
 #endif
 }
 
+#ifdef EIGEN_VECTORIZE_SSE4_1
+template<> EIGEN_STRONG_INLINE Packet4f pround<Packet4f>(const Packet4f& a) { return _mm_round_ps(a, 0); }
+template<> EIGEN_STRONG_INLINE Packet2d pround<Packet2d>(const Packet2d& a) { return _mm_round_pd(a, 0); }
+
+template<> EIGEN_STRONG_INLINE Packet4f pceil<Packet4f>(const Packet4f& a) { return _mm_ceil_ps(a); }
+template<> EIGEN_STRONG_INLINE Packet2d pceil<Packet2d>(const Packet2d& a) { return _mm_ceil_pd(a); }
+
+template<> EIGEN_STRONG_INLINE Packet4f pfloor<Packet4f>(const Packet4f& a) { return _mm_floor_ps(a); }
+template<> EIGEN_STRONG_INLINE Packet2d pfloor<Packet2d>(const Packet2d& a) { return _mm_floor_pd(a); }
+#endif
+
 template<> EIGEN_STRONG_INLINE Packet4f pand<Packet4f>(const Packet4f& a, const Packet4f& b) { return _mm_and_ps(a,b); }
 template<> EIGEN_STRONG_INLINE Packet2d pand<Packet2d>(const Packet2d& a, const Packet2d& b) { return _mm_and_pd(a,b); }
 template<> EIGEN_STRONG_INLINE Packet4i pand<Packet4i>(const Packet4i& a, const Packet4i& b) { return _mm_and_si128(a,b); }