@smake/eigen 1.0.1 → 1.1.0

package/eigen/Eigen/src/Core/arch/CUDA/TypeCasting.h

@@ -1,212 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2016 Benoit Steiner <benoit.steiner.goog@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_TYPE_CASTING_CUDA_H
-#define EIGEN_TYPE_CASTING_CUDA_H
-
-namespace Eigen {
-
-namespace internal {
-
-template<>
-struct scalar_cast_op<float, Eigen::half> {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
-  typedef Eigen::half result_type;
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Eigen::half operator() (const float& a) const {
-    #if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
-      return __float2half(a);
-    #else
-      return Eigen::half(a);
-    #endif
-  }
-};
-
-template<>
-struct functor_traits<scalar_cast_op<float, Eigen::half> >
-{ enum { Cost = NumTraits<float>::AddCost, PacketAccess = false }; };
-
-
-template<>
-struct scalar_cast_op<int, Eigen::half> {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
-  typedef Eigen::half result_type;
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Eigen::half operator() (const int& a) const {
-    #if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
-      return __float2half(static_cast<float>(a));
-    #else
-      return Eigen::half(static_cast<float>(a));
-    #endif
-  }
-};
-
-template<>
-struct functor_traits<scalar_cast_op<int, Eigen::half> >
-{ enum { Cost = NumTraits<float>::AddCost, PacketAccess = false }; };
-
-
-template<>
-struct scalar_cast_op<Eigen::half, float> {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
-  typedef float result_type;
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float operator() (const Eigen::half& a) const {
-    #if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
-      return __half2float(a);
-    #else
-      return static_cast<float>(a);
-    #endif
-  }
-};
-
-template<>
-struct functor_traits<scalar_cast_op<Eigen::half, float> >
-{ enum { Cost = NumTraits<float>::AddCost, PacketAccess = false }; };
-
-
-
-#if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
-
-template <>
-struct type_casting_traits<Eigen::half, float> {
-  enum {
-    VectorizedCast = 1,
-    SrcCoeffRatio = 2,
-    TgtCoeffRatio = 1
-  };
-};
-
-template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pcast<half2, float4>(const half2& a, const half2& b) {
-  float2 r1 = __half22float2(a);
-  float2 r2 = __half22float2(b);
-  return make_float4(r1.x, r1.y, r2.x, r2.y);
-}
-
-template <>
-struct type_casting_traits<float, Eigen::half> {
-  enum {
-    VectorizedCast = 1,
-    SrcCoeffRatio = 1,
-    TgtCoeffRatio = 2
-  };
-};
-
-template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 pcast<float4, half2>(const float4& a) {
-  // Simply discard the second half of the input
-  return __floats2half2_rn(a.x, a.y);
-}
-
-#elif defined EIGEN_VECTORIZE_AVX512
-template <>
-struct type_casting_traits<half, float> {
-  enum {
-    VectorizedCast = 1,
-    SrcCoeffRatio = 1,
-    TgtCoeffRatio = 1
-  };
-};
-
-template<> EIGEN_STRONG_INLINE Packet16f pcast<Packet16h, Packet16f>(const Packet16h& a) {
-  return half2float(a);
-}
-
-template <>
-struct type_casting_traits<float, half> {
-  enum {
-    VectorizedCast = 1,
-    SrcCoeffRatio = 1,
-    TgtCoeffRatio = 1
-  };
-};
-
-template<> EIGEN_STRONG_INLINE Packet16h pcast<Packet16f, Packet16h>(const Packet16f& a) {
-  return float2half(a);
-}
-
-#elif defined EIGEN_VECTORIZE_AVX
-
-template <>
-struct type_casting_traits<Eigen::half, float> {
-  enum {
-    VectorizedCast = 1,
-    SrcCoeffRatio = 1,
-    TgtCoeffRatio = 1
-  };
-};
-
-template<> EIGEN_STRONG_INLINE Packet8f pcast<Packet8h, Packet8f>(const Packet8h& a) {
-  return half2float(a);
-}
-
-template <>
-struct type_casting_traits<float, Eigen::half> {
-  enum {
-    VectorizedCast = 1,
-    SrcCoeffRatio = 1,
-    TgtCoeffRatio = 1
-  };
-};
-
-template<> EIGEN_STRONG_INLINE Packet8h pcast<Packet8f, Packet8h>(const Packet8f& a) {
-  return float2half(a);
-}
-
-// Disable the following code since it's broken on too many platforms / compilers.
-//#elif defined(EIGEN_VECTORIZE_SSE) && (!EIGEN_ARCH_x86_64) && (!EIGEN_COMP_MSVC)
-#elif 0
-
-template <>
-struct type_casting_traits<Eigen::half, float> {
-  enum {
-    VectorizedCast = 1,
-    SrcCoeffRatio = 1,
-    TgtCoeffRatio = 1
-  };
-};
-
-template<> EIGEN_STRONG_INLINE Packet4f pcast<Packet4h, Packet4f>(const Packet4h& a) {
-  __int64_t a64 = _mm_cvtm64_si64(a.x);
-  Eigen::half h = raw_uint16_to_half(static_cast<unsigned short>(a64));
-  float f1 = static_cast<float>(h);
-  h = raw_uint16_to_half(static_cast<unsigned short>(a64 >> 16));
-  float f2 = static_cast<float>(h);
-  h = raw_uint16_to_half(static_cast<unsigned short>(a64 >> 32));
-  float f3 = static_cast<float>(h);
-  h = raw_uint16_to_half(static_cast<unsigned short>(a64 >> 48));
-  float f4 = static_cast<float>(h);
-  return _mm_set_ps(f4, f3, f2, f1);
-}
-
-template <>
-struct type_casting_traits<float, Eigen::half> {
-  enum {
-    VectorizedCast = 1,
-    SrcCoeffRatio = 1,
-    TgtCoeffRatio = 1
-  };
-};
-
-template<> EIGEN_STRONG_INLINE Packet4h pcast<Packet4f, Packet4h>(const Packet4f& a) {
-  EIGEN_ALIGN16 float aux[4];
-  pstore(aux, a);
-  Eigen::half h0(aux[0]);
-  Eigen::half h1(aux[1]);
-  Eigen::half h2(aux[2]);
-  Eigen::half h3(aux[3]);
-
-  Packet4h result;
-  result.x = _mm_set_pi16(h3.x, h2.x, h1.x, h0.x);
-  return result;
-}
-
-#endif
-
-} // end namespace internal
-
-} // end namespace Eigen
-
-#endif // EIGEN_TYPE_CASTING_CUDA_H

package/eigen/Eigen/src/Geometry/arch/Geometry_SSE.h

@@ -1,161 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2009 Rohit Garg <rpg.314@gmail.com>
-// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_GEOMETRY_SSE_H
-#define EIGEN_GEOMETRY_SSE_H
-
-namespace Eigen { 
-
-namespace internal {
-
-template<class Derived, class OtherDerived>
-struct quat_product<Architecture::SSE, Derived, OtherDerived, float>
-{
-  enum {
-    AAlignment = traits<Derived>::Alignment,
-    BAlignment = traits<OtherDerived>::Alignment,
-    ResAlignment = traits<Quaternion<float> >::Alignment
-  };
-  static inline Quaternion<float> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
-  {
-    Quaternion<float> res;
-    const __m128 mask = _mm_setr_ps(0.f,0.f,0.f,-0.f);
-    __m128 a = _a.coeffs().template packet<AAlignment>(0);
-    __m128 b = _b.coeffs().template packet<BAlignment>(0);
-    __m128 s1 = _mm_mul_ps(vec4f_swizzle1(a,1,2,0,2),vec4f_swizzle1(b,2,0,1,2));
-    __m128 s2 = _mm_mul_ps(vec4f_swizzle1(a,3,3,3,1),vec4f_swizzle1(b,0,1,2,1));
-    pstoret<float,Packet4f,ResAlignment>(
-              &res.x(),
-              _mm_add_ps(_mm_sub_ps(_mm_mul_ps(a,vec4f_swizzle1(b,3,3,3,3)),
-                                    _mm_mul_ps(vec4f_swizzle1(a,2,0,1,0),
-                                               vec4f_swizzle1(b,1,2,0,0))),
-                         _mm_xor_ps(mask,_mm_add_ps(s1,s2))));
-    
-    return res;
-  }
-};
-
-template<class Derived>
-struct quat_conj<Architecture::SSE, Derived, float>
-{
-  enum {
-    ResAlignment = traits<Quaternion<float> >::Alignment
-  };
-  static inline Quaternion<float> run(const QuaternionBase<Derived>& q)
-  {
-    Quaternion<float> res;
-    const __m128 mask = _mm_setr_ps(-0.f,-0.f,-0.f,0.f);
-    pstoret<float,Packet4f,ResAlignment>(&res.x(), _mm_xor_ps(mask, q.coeffs().template packet<traits<Derived>::Alignment>(0)));
-    return res;
-  }
-};
-
-
-template<typename VectorLhs,typename VectorRhs>
-struct cross3_impl<Architecture::SSE,VectorLhs,VectorRhs,float,true>
-{
-  enum {
-    ResAlignment = traits<typename plain_matrix_type<VectorLhs>::type>::Alignment
-  };
-  static inline typename plain_matrix_type<VectorLhs>::type
-  run(const VectorLhs& lhs, const VectorRhs& rhs)
-  {
-    __m128 a = lhs.template packet<traits<VectorLhs>::Alignment>(0);
-    __m128 b = rhs.template packet<traits<VectorRhs>::Alignment>(0);
-    __m128 mul1=_mm_mul_ps(vec4f_swizzle1(a,1,2,0,3),vec4f_swizzle1(b,2,0,1,3));
-    __m128 mul2=_mm_mul_ps(vec4f_swizzle1(a,2,0,1,3),vec4f_swizzle1(b,1,2,0,3));
-    typename plain_matrix_type<VectorLhs>::type res;
-    pstoret<float,Packet4f,ResAlignment>(&res.x(),_mm_sub_ps(mul1,mul2));
-    return res;
-  }
-};
-
-
-
-
-template<class Derived, class OtherDerived>
-struct quat_product<Architecture::SSE, Derived, OtherDerived, double>
-{
-  enum {
-    BAlignment = traits<OtherDerived>::Alignment,
-    ResAlignment = traits<Quaternion<double> >::Alignment
-  };
-
-  static inline Quaternion<double> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
-  {
-  const Packet2d mask = _mm_castsi128_pd(_mm_set_epi32(0x0,0x0,0x80000000,0x0));
-
-  Quaternion<double> res;
-
-  const double* a = _a.coeffs().data();
-  Packet2d b_xy = _b.coeffs().template packet<BAlignment>(0);
-  Packet2d b_zw = _b.coeffs().template packet<BAlignment>(2);
-  Packet2d a_xx = pset1<Packet2d>(a[0]);
-  Packet2d a_yy = pset1<Packet2d>(a[1]);
-  Packet2d a_zz = pset1<Packet2d>(a[2]);
-  Packet2d a_ww = pset1<Packet2d>(a[3]);
-
-  // two temporaries:
-  Packet2d t1, t2;
-
-  /*
-   * t1 = ww*xy + yy*zw
-   * t2 = zz*xy - xx*zw
-   * res.xy = t1 +/- swap(t2)
-   */
-  t1 = padd(pmul(a_ww, b_xy), pmul(a_yy, b_zw));
-  t2 = psub(pmul(a_zz, b_xy), pmul(a_xx, b_zw));
-#ifdef EIGEN_VECTORIZE_SSE3
-  EIGEN_UNUSED_VARIABLE(mask)
-  pstoret<double,Packet2d,ResAlignment>(&res.x(), _mm_addsub_pd(t1, preverse(t2)));
-#else
-  pstoret<double,Packet2d,ResAlignment>(&res.x(), padd(t1, pxor(mask,preverse(t2))));
-#endif
-  
-  /*
-   * t1 = ww*zw - yy*xy
-   * t2 = zz*zw + xx*xy
-   * res.zw = t1 -/+ swap(t2) = swap( swap(t1) +/- t2)
-   */
-  t1 = psub(pmul(a_ww, b_zw), pmul(a_yy, b_xy));
-  t2 = padd(pmul(a_zz, b_zw), pmul(a_xx, b_xy));
-#ifdef EIGEN_VECTORIZE_SSE3
-  EIGEN_UNUSED_VARIABLE(mask)
-  pstoret<double,Packet2d,ResAlignment>(&res.z(), preverse(_mm_addsub_pd(preverse(t1), t2)));
-#else
-  pstoret<double,Packet2d,ResAlignment>(&res.z(), psub(t1, pxor(mask,preverse(t2))));
-#endif
-
-  return res;
-}
-};
-
-template<class Derived>
-struct quat_conj<Architecture::SSE, Derived, double>
-{
-  enum {
-    ResAlignment = traits<Quaternion<double> >::Alignment
-  };
-  static inline Quaternion<double> run(const QuaternionBase<Derived>& q)
-  {
-    Quaternion<double> res;
-    const __m128d mask0 = _mm_setr_pd(-0.,-0.);
-    const __m128d mask2 = _mm_setr_pd(-0.,0.);
-    pstoret<double,Packet2d,ResAlignment>(&res.x(), _mm_xor_pd(mask0, q.coeffs().template packet<traits<Derived>::Alignment>(0)));
-    pstoret<double,Packet2d,ResAlignment>(&res.z(), _mm_xor_pd(mask2, q.coeffs().template packet<traits<Derived>::Alignment>(2)));
-    return res;
-  }
-};
-
-} // end namespace internal
-
-} // end namespace Eigen
-
-#endif // EIGEN_GEOMETRY_SSE_H