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.

9 files changed, 104 insertions, 53 deletions

diff --git a/Eigen/src/Core/GenericPacketMath.h b/Eigen/src/Core/GenericPacketMath.h
index 8bdf16e16..8bcceaa7b 100644
--- a/Eigen/src/Core/GenericPacketMath.h
+++ b/Eigen/src/Core/GenericPacketMath.h
@@ -214,17 +214,13 @@ 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 a packet with constant coefficients \a a, e.g.: (a,a,a,a) */
+/** \internal \returns ones */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pset1(const typename unpacket_traits<Packet>::type& a) { return a; }
+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) {
-  typedef typename unpacket_traits<Packet>::type Scalar;
-  Packet ones  = pset1<Packet>(Scalar(1));
-  return pandnot(ones, a);
-}
+template <typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pnot(const Packet& a) { return pxor(pones(a), a);}
 
 /** \internal \returns \a a shifted by N bits to the right */
 template<int N> EIGEN_DEVICE_FUNC inline int
@@ -262,24 +258,19 @@ pselect(const Packet& mask, const Packet& a, const Packet& b) {
 
 /** \internal \returns a <= b as a bit mask */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pcmp_le(const Packet& a, const Packet& b); /* { return a<=b ? pnot(pxor(a,a)) : pxor(a,a); } */
+pcmp_le(const Packet& a, const Packet& b)  { return a<=b ? pones(a) : pzero(a); }
 
 /** \internal \returns a < b as a bit mask */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pcmp_lt(const Packet& a, const Packet& b); /* { return a<b  ? pnot(pxor(a,a)) : pxor(a,a); } */
+pcmp_lt(const Packet& a, const Packet& b)  { return a<b ? pones(a) : pzero(a); }
 
 /** \internal \returns a == b as a bit mask */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pcmp_eq(const Packet& a, const Packet& b)
-{
-  typedef typename unpacket_traits<Packet>::type Scalar;
-  Packet zeros  = pset1<Packet>(Scalar(0));
-  return a==b ? pnot(zeros) : zeros;
-}
+pcmp_eq(const Packet& a, const Packet& b) { return a==b ? pones(a) : pzero(a); }
 
 /** \internal \returns a < b or a==NaN or b==NaN as a bit mask */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pcmp_lt_or_nan(const Packet& a, const Packet& b); /* { return pnot(pcmp_le(b,a)); } */
+pcmp_lt_or_nan(const Packet& a, const Packet& b) { return pnot(pcmp_le(b,a)); } 
 
 /** \internal \returns a packet version of \a *from, from must be 16 bytes aligned */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
@@ -289,6 +280,10 @@ pload(const typename unpacket_traits<Packet>::type* from) { return *from; }
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 ploadu(const typename unpacket_traits<Packet>::type* from) { return *from; }
 
+/** \internal \returns a packet with constant coefficients \a a, e.g.: (a,a,a,a) */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pset1(const typename unpacket_traits<Packet>::type& a) { return a; }
+
 /** \internal \returns a packet with constant coefficients set from bits */
 template<typename Packet,typename BitsType> EIGEN_DEVICE_FUNC inline Packet
 pset1frombits(BitsType a);
diff --git a/Eigen/src/Core/arch/AVX/Complex.h b/Eigen/src/Core/arch/AVX/Complex.h
index 23687c624..9f1bb969e 100644
--- a/Eigen/src/Core/arch/AVX/Complex.h
+++ b/Eigen/src/Core/arch/AVX/Complex.h
@@ -72,10 +72,11 @@ template<> EIGEN_STRONG_INLINE Packet4cf pmul<Packet4cf>(const Packet4cf& a, con
 template <>
 EIGEN_STRONG_INLINE Packet4cf pcmp_eq(const Packet4cf& a, const Packet4cf& b) {
   __m256 eq = _mm256_cmp_ps(a.v, b.v, _CMP_EQ_OQ);
-  __m256 real_and_imag_equal = _mm256_and_ps(eq, _mm256_permute_ps(eq, 0xb1));
-  return Packet4cf(real_and_imag_equal);
+  return Packet4cf(_mm256_and_ps(eq, _mm256_permute_ps(eq, 0xb1)));
 }
 
+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)); }
@@ -286,10 +287,11 @@ template<> EIGEN_STRONG_INLINE Packet2cd pmul<Packet2cd>(const Packet2cd& a, con
 template <>
 EIGEN_STRONG_INLINE Packet2cd pcmp_eq(const Packet2cd& a, const Packet2cd& b) {
   __m256d eq = _mm256_cmp_pd(a.v, b.v, _CMP_EQ_OQ);
-  __m256d real_and_imag_equal = _mm256_and_pd(eq, _mm256_permute_pd(eq, 0x5));
-  return Packet2cd(real_and_imag_equal);
+  return Packet2cd(pand(eq, _mm256_permute_pd(eq, 0x5)));
 }
 
+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/Complex.h b/Eigen/src/Core/arch/AVX512/Complex.h
index 2c613f870..154fedc25 100644
--- a/Eigen/src/Core/arch/AVX512/Complex.h
+++ b/Eigen/src/Core/arch/AVX512/Complex.h
@@ -83,10 +83,7 @@ template<> EIGEN_STRONG_INLINE Packet8cf pandnot<Packet8cf>(const Packet8cf& a,
 template <>
 EIGEN_STRONG_INLINE Packet8cf pcmp_eq(const Packet8cf& a, const Packet8cf& b) {
   __m512 eq = pcmp_eq<Packet16f>(a.v, b.v);
-  __m512 eq_swap_real_imag = _mm512_permute_ps(eq, 0xB1);
-  __m512i real_and_imag_equal = _mm512_and_si512(
-      _mm512_castps_si512(eq), _mm512_castps_si512(eq_swap_real_imag));
-  return Packet8cf(_mm512_castsi512_ps(real_and_imag_equal));
+  return Packet8cf(pand(eq, _mm512_permute_ps(eq, 0xB1)));
 }
 
 template<> EIGEN_STRONG_INLINE Packet8cf pload <Packet8cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet8cf(pload<Packet16f>(&numext::real_ref(*from))); }
@@ -279,10 +276,7 @@ template<> EIGEN_STRONG_INLINE Packet4cd pandnot<Packet4cd>(const Packet4cd& a,
 template <>
 EIGEN_STRONG_INLINE Packet4cd pcmp_eq(const Packet4cd& a, const Packet4cd& b) {
   __m512d eq = pcmp_eq<Packet8d>(a.v, b.v);
-  __m512d eq_swap_real_imag = _mm512_permute_pd(eq, 0x55);
-  __m512i real_and_imag_equal = _mm512_and_si512(
-      _mm512_castpd_si512(eq), _mm512_castpd_si512(eq_swap_real_imag));
-  return Packet4cd(_mm512_castsi512_pd(real_and_imag_equal));
+  return Packet4cd(pand(eq, _mm512_permute_pd(eq, 0x55)));
 }
 
 template<> EIGEN_STRONG_INLINE Packet4cd pload <Packet4cd>(const std::complex<double>* from)
diff --git a/Eigen/src/Core/arch/AVX512/PacketMath.h b/Eigen/src/Core/arch/AVX512/PacketMath.h
index 68adf5e57..9666c4e22 100644
--- a/Eigen/src/Core/arch/AVX512/PacketMath.h
+++ b/Eigen/src/Core/arch/AVX512/PacketMath.h
@@ -284,47 +284,56 @@ 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);
-}
-
-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);
+  __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 <>
 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..fa84097ac 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)); }
@@ -442,15 +447,13 @@ ptranspose(PacketBlock<Packet2cf,2>& kernel) {
 template<> EIGEN_STRONG_INLINE Packet2cf pcmp_eq(const Packet2cf& a, const Packet2cf& b)
 {
   __m128 eq = _mm_cmpeq_ps(a.v, b.v);
-  __m128 real_and_imag_equal = _mm_and_ps(eq, vec4f_swizzle1(eq, 1, 0, 3, 2));
-  return Packet2cf(real_and_imag_equal);
+  return Packet2cf(pand(eq, vec4f_swizzle1(eq, 1, 0, 3, 2)));
 }
 
 template<> EIGEN_STRONG_INLINE Packet1cd pcmp_eq(const Packet1cd& a, const Packet1cd& b)
 {
   __m128d eq = _mm_cmpeq_pd(a.v, b.v);
-  __m128d real_and_imag_equal = _mm_and_pd(eq, vec2d_swizzle1(eq, 1, 0));
-  return Packet1cd(real_and_imag_equal);
+  return Packet1cd(pand(eq, vec2d_swizzle1(eq, 1, 0)));
 }
 
 template<>  EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, const Packet2cf& thenPacket, const Packet2cf& elsePacket) {
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); }
diff --git a/test/packetmath.cpp b/test/packetmath.cpp
index a88b7bba9..460cfbdbe 100644
--- a/test/packetmath.cpp
+++ b/test/packetmath.cpp
@@ -239,6 +239,8 @@ template<typename Scalar,typename Packet> void packetmath()
   CHECK_CWISE2_IF(PacketTraits::HasDiv, REF_DIV, internal::pdiv);
 
   CHECK_CWISE1(internal::pnot, internal::pnot);
+  CHECK_CWISE1(internal::pzero, internal::pzero);
+  CHECK_CWISE1(internal::pones, internal::pones);
   CHECK_CWISE1(internal::negate, internal::pnegate);
   CHECK_CWISE1(numext::conj, internal::pconj);