Add packet up "pones". Write pnot(a) as pxor(pones(a), a).

6 files changed, 73 insertions, 8 deletions

diff --git a/Eigen/src/Core/arch/AVX/Complex.h b/Eigen/src/Core/arch/AVX/Complex.h
index 23687c624..d880ef593 100644
--- a/Eigen/src/Core/arch/AVX/Complex.h
+++ b/Eigen/src/Core/arch/AVX/Complex.h
@@ -76,6 +76,8 @@ EIGEN_STRONG_INLINE Packet4cf pcmp_eq(const Packet4cf& a, const Packet4cf& b) {
   return Packet4cf(real_and_imag_equal);
 }
 
+template<> EIGEN_STRONG_INLINE Packet4cf pones<Packet4cf>(const Packet4cf& a) { return Packet4cf(pones(a.v)); }
+template<> EIGEN_STRONG_INLINE Packet4cf pnot<Packet4cf>(const Packet4cf& a) { return Packet4cf(pnot(a.v)); }
 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)); }
@@ -290,6 +292,8 @@ EIGEN_STRONG_INLINE Packet2cd pcmp_eq(const Packet2cd& a, const Packet2cd& b) {
   return Packet2cd(real_and_imag_equal);
 }
 
+template<> EIGEN_STRONG_INLINE Packet2cd pones<Packet2cd>(const Packet2cd& a) { return Packet2cd(pones(a.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cd pnot<Packet2cd>(const Packet2cd& a) { return Packet2cd(pnot(a.v)); }
 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)); }
diff --git a/Eigen/src/Core/arch/AVX/PacketMath.h b/Eigen/src/Core/arch/AVX/PacketMath.h
index a6af48f21..f6a514fbf 100644
--- a/Eigen/src/Core/arch/AVX/PacketMath.h
+++ b/Eigen/src/Core/arch/AVX/PacketMath.h
@@ -250,6 +250,25 @@ template<> EIGEN_STRONG_INLINE Packet4d pceil<Packet4d>(const Packet4d& a) { ret
 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); }
 
+
+#ifdef EIGEN_VECTORIZE_AVX2
+template<> EIGEN_STRONG_INLINE Packet8i pones<Packet8i>(const Packet8i& a) {
+  return _mm256_cmpeq_epi64(a,a);
+}
+#else
+template<> EIGEN_STRONG_INLINE Packet8i pones<Packet8i>(const Packet8i& /*a*/) {
+  const unsigned int o = 0xffffffffu;
+  return _mm256_set_epi32(o, o, o, o, o, o, o, o);
+}
+#endif
+template<> EIGEN_STRONG_INLINE Packet8f pones<Packet8f>(const Packet8f& a) {
+  return _mm256_castsi256_ps(pones<Packet8i>(_mm256_castps_si256(a)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4d pones<Packet4d>(const Packet4d& a) {
+  return _mm256_castsi256_pd(pones<Packet8i>(_mm256_castpd_si256(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); }
 template<> EIGEN_STRONG_INLINE Packet8i pand<Packet8i>(const Packet8i& a, const Packet8i& b) {
diff --git a/Eigen/src/Core/arch/AVX512/PacketMath.h b/Eigen/src/Core/arch/AVX512/PacketMath.h
index 68adf5e57..d258fd07b 100644
--- a/Eigen/src/Core/arch/AVX512/PacketMath.h
+++ b/Eigen/src/Core/arch/AVX512/PacketMath.h
@@ -295,12 +295,6 @@ template<> EIGEN_STRONG_INLINE Packet16f pcmp_lt(const Packet16f& a, const Packe
   return cat256(lo, hi);
 }
 
-template<> EIGEN_STRONG_INLINE Packet16f pcmp_eq(const Packet16f& a, const Packet16f& b) {
-  __m256 lo = pcmp_eq(extract256<0>(a), extract256<0>(b));
-  __m256 hi = pcmp_eq(extract256<1>(a), extract256<1>(b));
-  return cat256(lo, hi);
-}
-
 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));
@@ -317,14 +311,30 @@ template <>
 EIGEN_STRONG_INLINE Packet16f pcmp_eq(const Packet16f& a, const Packet16f& b) {
   __mmask16 mask = _mm512_cmp_ps_mask(a, b, _CMP_EQ_OQ);
   return _mm512_castsi512_ps(
-      _mm512_mask_set1_epi32(_mm512_set1_epi32(0), mask, 0xffffffff));
+      _mm512_mask_set1_epi32(_mm512_set1_epi32(0), mask, 0xffffffffu));
 }
 
 template <>
 EIGEN_STRONG_INLINE Packet8d pcmp_eq(const Packet8d& a, const Packet8d& b) {
   __mmask8 mask = _mm512_cmp_pd_mask(a, b, _CMP_EQ_OQ);
   return _mm512_castsi512_pd(
-      _mm512_mask_set1_epi64(_mm512_set1_epi64(0), mask, 0xffffffffffffffff));
+      _mm512_mask_set1_epi64(_mm512_set1_epi64(0), mask, 0xffffffffffffffffu));
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16i pones<Packet16i>(const Packet16i& /*a*/) {
+  const unsigned int o = 0xffffffffu;
+  return _mm512_set_epi32(o, o, o, o, o, o, o, o, o, o, o, o, o, o, o, o);
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pones<Packet16f>(const Packet16f& a) {
+  return _mm512_castsi512_ps(pones<Packet16i>(_mm512_castps_si512(a)));
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet8d pones<Packet8d>(const Packet8d& a) {
+  return _mm512_castsi512_pd(pones<Packet16i>(_mm512_castpd_si512(a)));
 }
 
 template <>
diff --git a/Eigen/src/Core/arch/GPU/PacketMathHalf.h b/Eigen/src/Core/arch/GPU/PacketMathHalf.h
index 3e35f96cc..c4dfedcf8 100644
--- a/Eigen/src/Core/arch/GPU/PacketMathHalf.h
+++ b/Eigen/src/Core/arch/GPU/PacketMathHalf.h
@@ -143,6 +143,10 @@ template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 pabs<half2>(const half2&
   return result;
 }
 
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 pones<half2>(const half2& a) {
+  half2 result;
+  *(reinterpret_cast<unsigned*>(&(result))) = 0xffffffffu;
+}  
 
 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void
 ptranspose(PacketBlock<half2,2>& kernel) {
@@ -640,6 +644,14 @@ EIGEN_STRONG_INLINE Packet16h float2half(const Packet16f& a) {
 #endif
 }
 
+template<> EIGEN_STRONG_INLINE Packet16h pnot(const Packet16h& a) {
+  Packet16h r; r.x = _mm256_xor_si256(a.x, pcmp_eq(a.x, a.x)); return r;
+}
+
+template<> EIGEN_STRONG_INLINE Packet16h pones(const Packet16h& a) {
+  Packet16h r; r.x = Packet8i(pones(a.x)); return r;
+}
+
 template<> EIGEN_STRONG_INLINE Packet16h por(const Packet16h& a,const Packet16h& b) {
   // in some cases Packet8i is a wrapper around __m256i, so we need to 
   // cast to Packet8i to call the correct overload.
@@ -1085,6 +1097,10 @@ EIGEN_STRONG_INLINE Packet8h float2half(const Packet8f& a) {
 #endif
 }
 
+template<> EIGEN_STRONG_INLINE Packet8h pones(const Packet8h& a) {
+  Packet8h r; r.x = _mm_cmpeq_epi32(a.x, a.x); return r;
+}
+
 template<> EIGEN_STRONG_INLINE Packet8h por(const Packet8h& a,const Packet8h& b) {
   // in some cases Packet4i is a wrapper around __m128i, so we either need to 
   // cast to Packet4i to directly call the intrinsics as below:
diff --git a/Eigen/src/Core/arch/SSE/Complex.h b/Eigen/src/Core/arch/SSE/Complex.h
index a7304193b..8372cedfb 100644
--- a/Eigen/src/Core/arch/SSE/Complex.h
+++ b/Eigen/src/Core/arch/SSE/Complex.h
@@ -82,6 +82,9 @@ template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, con
   #endif
 }
 
+template<> EIGEN_STRONG_INLINE Packet2cf pones   <Packet2cf>(const Packet2cf& a) { return Packet2cf(pones(a.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pnot   <Packet2cf>(const Packet2cf& a) { return Packet2cf(pnot(a.v)); }
+
 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)); }
@@ -305,6 +308,8 @@ template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, con
   #endif
 }
 
+template<> EIGEN_STRONG_INLINE Packet1cd pones  <Packet1cd>(const Packet1cd& a) { return Packet1cd(pones(a.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pnot   <Packet1cd>(const Packet1cd& a) { return Packet1cd(pnot(a.v)); }
 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)); }
diff --git a/Eigen/src/Core/arch/SSE/PacketMath.h b/Eigen/src/Core/arch/SSE/PacketMath.h
index b8a5497a9..6dd2f8a46 100755
--- a/Eigen/src/Core/arch/SSE/PacketMath.h
+++ b/Eigen/src/Core/arch/SSE/PacketMath.h
@@ -378,6 +378,17 @@ template<> EIGEN_STRONG_INLINE Packet4i pcmp_eq(const Packet4i& a, const Packet4
 template<> EIGEN_STRONG_INLINE Packet2d pcmp_eq(const Packet2d& a, const Packet2d& b) { return _mm_cmpeq_pd(a,b); }
 template<> EIGEN_STRONG_INLINE Packet4f pcmp_lt_or_nan(const Packet4f& a, const Packet4f& b) { return _mm_cmpnge_ps(a,b); }
 
+template<> EIGEN_STRONG_INLINE Packet4i pones<Packet4i>(const Packet4i& a) { return _mm_cmpeq_epi32(a, a); }
+template<> EIGEN_STRONG_INLINE Packet4f
+pones<Packet4f>(const Packet4f& a) {
+  Packet4i b = _mm_castps_si128(a);
+  return _mm_castsi128_ps(_mm_cmpeq_epi32(b, b));
+}
+template<> EIGEN_STRONG_INLINE Packet2d
+pones<Packet2d>(const Packet2d& a) {
+  Packet4i b = _mm_castpd_si128(a);
+  return _mm_castsi128_pd(_mm_cmpeq_epi32(b, b));
+}
 
 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); }