From 8f178429b9d0517a5a63da6dd73adbd84a15b375 Mon Sep 17 00:00:00 2001
From: Rasmus Munk Larsen <rmlarsen@google.com>
Date: Wed, 9 Jan 2019 16:34:23 -0800
Subject: [PATCH] Collapsed revision * Collapsed revision * Add packet up
 "pones". Write pnot(a) as pxor(pones(a), a). * Collapsed revision * Simplify
 a bit. * Undo useless diffs. * Fix typo.

---
 Eigen/src/Core/GenericPacketMath.h      |  8 ++++----
 Eigen/src/Core/arch/AVX512/PacketMath.h | 23 +++++++++++------------
 2 files changed, 15 insertions(+), 16 deletions(-)

diff --git a/Eigen/src/Core/GenericPacketMath.h b/Eigen/src/Core/GenericPacketMath.h
index 048fc5157..8bcceaa7b 100644
--- a/Eigen/src/Core/GenericPacketMath.h
+++ b/Eigen/src/Core/GenericPacketMath.h
@@ -214,6 +214,10 @@ pxor(const Packet& a, const Packet& b) { return a ^ b; }
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pandnot(const Packet& a, const Packet& b) { return a & (~b); }
 
+/** \internal \returns ones */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pones(const Packet& /*a*/) { Packet b; memset(&b, 0xff, sizeof(b)); return b;}
+
 /** \internal \returns the bitwise not of \a a */
 template <typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pnot(const Packet& a) { return pxor(pones(a), a);}
@@ -246,10 +250,6 @@ pldexp(const Packet &a, const Packet &exponent) { return std::ldexp(a,exponent);
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pzero(const Packet& a) { return pxor(a,a); }
 
-/** \internal \returns ones */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pones(const Packet& /*a*/) { Packet b; memset(&b, 0xff, sizeof(b)); return b;}
-
 /** \internal \returns bits of \a or \b according to the input bit mask \a mask */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pselect(const Packet& mask, const Packet& a, const Packet& b) {
diff --git a/Eigen/src/Core/arch/AVX512/PacketMath.h b/Eigen/src/Core/arch/AVX512/PacketMath.h
index d258fd07b..9666c4e22 100644
--- a/Eigen/src/Core/arch/AVX512/PacketMath.h
+++ b/Eigen/src/Core/arch/AVX512/PacketMath.h
@@ -284,27 +284,26 @@ EIGEN_STRONG_INLINE Packet16f cat256(Packet8f a, Packet8f b) {
 #endif
 
 template<> EIGEN_STRONG_INLINE Packet16f pcmp_le(const Packet16f& a, const Packet16f& b) {
-  __m256 lo = pcmp_le(extract256<0>(a), extract256<0>(b));
-  __m256 hi = pcmp_le(extract256<1>(a), extract256<1>(b));
-  return cat256(lo, hi);
+  __mmask16 mask = _mm512_cmp_ps_mask(a, b, _CMP_LE_OQ);
+  return _mm512_castsi512_ps(
+      _mm512_mask_set1_epi32(_mm512_set1_epi32(0), mask, 0xffffffffu));
 }
 
 template<> EIGEN_STRONG_INLINE Packet16f pcmp_lt(const Packet16f& a, const Packet16f& b) {
-  __m256 lo = pcmp_lt(extract256<0>(a), extract256<0>(b));
-  __m256 hi = pcmp_lt(extract256<1>(a), extract256<1>(b));
-  return cat256(lo, hi);
+  __mmask16 mask = _mm512_cmp_ps_mask(a, b, _CMP_LT_OQ);
+  return _mm512_castsi512_ps(
+      _mm512_mask_set1_epi32(_mm512_set1_epi32(0), mask, 0xffffffffu));
 }
 
 template<> EIGEN_STRONG_INLINE Packet16f pcmp_lt_or_nan(const Packet16f& a, const Packet16f& b) {
-  __m256 lo = pcmp_lt_or_nan(extract256<0>(a), extract256<0>(b));
-  __m256 hi = pcmp_lt_or_nan(extract256<1>(a), extract256<1>(b));
-  return cat256(lo, hi);
+  __mmask16 mask = _mm512_cmp_ps_mask(a, b, _CMP_NGT_UQ);
+  return _mm512_castsi512_ps(
+      _mm512_mask_set1_epi32(_mm512_set1_epi32(0), mask, 0xffffffffu));
 }
 
 template<> EIGEN_STRONG_INLINE Packet16i pcmp_eq(const Packet16i& a, const Packet16i& b) {
-  __m256i lo = _mm256_cmpeq_epi32(_mm512_extracti64x4_epi64(a, 0), _mm512_extracti64x4_epi64(b, 0));
-  __m256i hi = _mm256_cmpeq_epi32(_mm512_extracti64x4_epi64(a, 1), _mm512_extracti64x4_epi64(b, 1));
-  return _mm512_inserti64x4(_mm512_castsi256_si512(lo), hi, 1);
+  __mmask16 mask = _mm512_cmp_epi32_mask(a, b, _CMP_EQ_OQ);
+  return _mm512_mask_set1_epi32(_mm512_set1_epi32(0), mask, 0xffffffffu);
 }
 
 template <>