GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-04-22 01:29:35 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merge

Browse Source
This commit is contained in:
Rasmus Munk Larsen 2017-10-13 15:58:52 -07:00
commit 2dd63ed395
11 changed files with 967 additions and 442 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
CMakeLists.txt
View File

@ -271,12 +271,18 @@ if(NOT MSVC)
message(STATUS "Enabling NEON in tests/examples")
endif()
option(EIGEN_TEST_ZVECTOR "Enable/Disable S390X(zEC13) ZVECTOR in tests/examples" OFF)
if(EIGEN_TEST_ZVECTOR)
option(EIGEN_TEST_Z13 "Enable/Disable S390X(zEC13) ZVECTOR in tests/examples" OFF)
if(EIGEN_TEST_Z13)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=z13 -mzvector")
message(STATUS "Enabling S390X(zEC13) ZVECTOR in tests/examples")
endif()
option(EIGEN_TEST_Z14 "Enable/Disable S390X(zEC14) ZVECTOR in tests/examples" OFF)
if(EIGEN_TEST_Z14)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=z14 -mzvector")
message(STATUS "Enabling S390X(zEC13) ZVECTOR in tests/examples")
endif()
check_cxx_compiler_flag("-fopenmp" COMPILER_SUPPORT_OPENMP)
if(COMPILER_SUPPORT_OPENMP)
option(EIGEN_TEST_OPENMP "Enable/Disable OpenMP in tests/examples" OFF)

2
Eigen/QtAlignedMalloc
View File

@ -27,7 +27,7 @@ void qFree(void *ptr)
void *qRealloc(void *ptr, std::size_t size)
{
void* newPtr = Eigen::internal::aligned_malloc(size);
memcpy(newPtr, ptr, size);
std::memcpy(newPtr, ptr, size);
Eigen::internal::aligned_free(ptr);
return newPtr;
}

2
Eigen/src/Core/Map.h
View File

@ -114,7 +114,7 @@ template<typename PlainObjectType, int MapOptions, typename StrideType> class Ma
inline Index outerStride() const
{
return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer()
: internal::traits<Map>::OuterStrideAtCompileTime != Dynamic ? internal::traits<Map>::OuterStrideAtCompileTime
: internal::traits<Map>::OuterStrideAtCompileTime != Dynamic ? Index(internal::traits<Map>::OuterStrideAtCompileTime)
: IsVectorAtCompileTime ? (this->size() * innerStride())
: int(Flags)&RowMajorBit ? (this->cols() * innerStride())
: (this->rows() * innerStride());

426
Eigen/src/Core/arch/ZVector/Complex.h
View File

@ -15,6 +15,10 @@ namespace Eigen {
namespace internal {
#if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ >= 12)
static Packet4ui p4ui_CONJ_XOR = { 0x00000000, 0x80000000, 0x00000000, 0x80000000 }; //vec_mergeh((Packet4ui)p4i_ZERO, (Packet4ui)p4f_MZERO);
#endif
static Packet2ul p2ul_CONJ_XOR1 = (Packet2ul) vec_sld((Packet4ui) p2d_ZERO_, (Packet4ui) p2l_ZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
static Packet2ul p2ul_CONJ_XOR2 = (Packet2ul) vec_sld((Packet4ui) p2l_ZERO, (Packet4ui) p2d_ZERO_, 8);//{ 0x8000000000000000, 0x0000000000000000 };
@ -29,10 +33,14 @@ struct Packet2cf
{
EIGEN_STRONG_INLINE Packet2cf() {}
EIGEN_STRONG_INLINE explicit Packet2cf(const Packet4f& a) : v(a) {}
#if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ < 12)
union {
Packet4f v;
Packet1cd cd[2];
};
#else
Packet4f v;
#endif
};
template<> struct packet_traits<std::complex<float> > : default_packet_traits
@ -89,63 +97,27 @@ template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type
/* Forward declaration */
EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel);
template<> EIGEN_STRONG_INLINE Packet2cf pload <Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(pload<Packet4f>((const float*)from)); }
/* complex<double> first */
template<> EIGEN_STRONG_INLINE Packet1cd pload <Packet1cd>(const std::complex<double>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet1cd(pload<Packet2d>((const double*)from)); }
template<> EIGEN_STRONG_INLINE Packet2cf ploadu<Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cf(ploadu<Packet4f>((const float*)from)); }
template<> EIGEN_STRONG_INLINE Packet1cd ploadu<Packet1cd>(const std::complex<double>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet1cd(ploadu<Packet2d>((const double*)from)); }
template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((float*)to, from.v); }
template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> * to, const Packet1cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, from.v); }
template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((float*)to, from.v); }
template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> * to, const Packet1cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, from.v); }
template<> EIGEN_STRONG_INLINE Packet1cd pset1<Packet1cd>(const std::complex<double>& from)
{ /* here we really have to use unaligned loads :( */ return ploadu<Packet1cd>(&from); }
template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>& from)
{
Packet2cf res;
res.cd[0] = Packet1cd(vec_ld2f((const float *)&from));
res.cd[1] = res.cd[0];
return res;
}
template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, Index stride)
{
std::complex<float> EIGEN_ALIGN16 af[2];
af[0] = from[0*stride];
af[1] = from[1*stride];
return pload<Packet2cf>(af);
}
template<> EIGEN_DEVICE_FUNC inline Packet1cd pgather<std::complex<double>, Packet1cd>(const std::complex<double>* from, Index stride EIGEN_UNUSED)
{
return pload<Packet1cd>(from);
}
template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, Index stride)
{
std::complex<float> EIGEN_ALIGN16 af[2];
pstore<std::complex<float> >((std::complex<float> *) af, from);
to[0*stride] = af[0];
to[1*stride] = af[1];
}
template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<double>, Packet1cd>(std::complex<double>* to, const Packet1cd& from, Index stride EIGEN_UNUSED)
{
pstore<std::complex<double> >(to, from);
}
template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(padd<Packet4f>(a.v, b.v)); }
template<> EIGEN_STRONG_INLINE Packet1cd padd<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(a.v + b.v); }
template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(psub<Packet4f>(a.v, b.v)); }
template<> EIGEN_STRONG_INLINE Packet1cd psub<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(a.v - b.v); }
template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) { return Packet1cd(pnegate(Packet2d(a.v))); }
template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a) { return Packet2cf(pnegate(Packet4f(a.v))); }
template<> EIGEN_STRONG_INLINE Packet1cd pconj(const Packet1cd& a) { return Packet1cd((Packet2d)vec_xor((Packet2d)a.v, (Packet2d)p2ul_CONJ_XOR2)); }
template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a)
{
Packet2cf res;
res.v.v4f[0] = pconj(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[0]))).v;
res.v.v4f[1] = pconj(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[1]))).v;
return res;
}
template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
{
Packet2d a_re, a_im, v1, v2;
@ -163,27 +135,12 @@ template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, con
return Packet1cd(v1 + v2);
}
template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
{
Packet2cf res;
res.v.v4f[0] = pmul(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[0])), Packet1cd(reinterpret_cast<Packet2d>(b.v.v4f[0]))).v;
res.v.v4f[1] = pmul(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[1])), Packet1cd(reinterpret_cast<Packet2d>(b.v.v4f[1]))).v;
return res;
}
template<> EIGEN_STRONG_INLINE Packet1cd pand <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_and(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf pand <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pand<Packet4f>(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet1cd por <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_or(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf por <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(por<Packet4f>(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet1cd pxor <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_xor(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf pxor <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pxor<Packet4f>(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet1cd pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_and(a.v, vec_nor(b.v,b.v))); }
template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pandnot<Packet4f>(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet1cd pand <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_and(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet1cd por <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_or(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet1cd pxor <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_xor(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet1cd pandnot <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_and(a.v, vec_nor(b.v,b.v))); }
template<> EIGEN_STRONG_INLINE Packet1cd ploaddup<Packet1cd>(const std::complex<double>* from) { return pset1<Packet1cd>(*from); }
template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<float>* from) { return pset1<Packet2cf>(*from); }
template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> * addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
template<> EIGEN_STRONG_INLINE void prefetch<std::complex<double> >(const std::complex<double> * addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
template<> EIGEN_STRONG_INLINE std::complex<double> pfirst<Packet1cd>(const Packet1cd& a)
@ -193,61 +150,20 @@ template<> EIGEN_STRONG_INLINE std::complex<double> pfirst<Packet1cd>(const Pac
return res;
}
template<> EIGEN_STRONG_INLINE std::complex<float> pfirst<Packet2cf>(const Packet2cf& a)
{
std::complex<float> EIGEN_ALIGN16 res[2];
pstore<std::complex<float> >(res, a);
return res[0];
}
template<> EIGEN_STRONG_INLINE Packet1cd preverse(const Packet1cd& a) { return a; }
template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a)
{
Packet2cf res;
res.cd[0] = a.cd[1];
res.cd[1] = a.cd[0];
return res;
}
template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet1cd>(const Packet1cd& a)
{
return pfirst(a);
}
template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a)
{
std::complex<float> res;
Packet1cd b = padd<Packet1cd>(a.cd[0], a.cd[1]);
vec_st2f(b.v, (float*)&res);
return res;
}
template<> EIGEN_STRONG_INLINE Packet1cd preduxp<Packet1cd>(const Packet1cd* vecs)
{
return vecs[0];
}
template<> EIGEN_STRONG_INLINE Packet2cf preduxp<Packet2cf>(const Packet2cf* vecs)
{
PacketBlock<Packet2cf,2> transpose;
transpose.packet[0] = vecs[0];
transpose.packet[1] = vecs[1];
ptranspose(transpose);
return padd<Packet2cf>(transpose.packet[0], transpose.packet[1]);
}
template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet1cd>(const Packet1cd& a)
{
return pfirst(a);
}
template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const Packet2cf& a)
{
std::complex<float> res;
Packet1cd b = pmul<Packet1cd>(a.cd[0], a.cd[1]);
vec_st2f(b.v, (float*)&res);
return res;
}
template<int Offset>
struct palign_impl<Offset,Packet1cd>
{
@ -258,18 +174,6 @@ struct palign_impl<Offset,Packet1cd>
}
};
template<int Offset>
struct palign_impl<Offset,Packet2cf>
{
static EIGEN_STRONG_INLINE void run(Packet2cf& first, const Packet2cf& second)
{
if (Offset == 1) {
first.cd[0] = first.cd[1];
first.cd[1] = second.cd[0];
}
}
};
template<> struct conj_helper<Packet1cd, Packet1cd, false,true>
{
EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
@ -303,6 +207,156 @@ template<> struct conj_helper<Packet1cd, Packet1cd, true,true>
}
};
EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet1cd,Packet2d)
template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
{
// TODO optimize it for AltiVec
Packet1cd res = conj_helper<Packet1cd,Packet1cd,false,true>().pmul(a,b);
Packet2d s = vec_madd(b.v, b.v, p2d_ZERO_);
return Packet1cd(pdiv(res.v, s + vec_perm(s, s, p16uc_REVERSE64)));
}
EIGEN_STRONG_INLINE Packet1cd pcplxflip/*<Packet1cd>*/(const Packet1cd& x)
{
return Packet1cd(preverse(Packet2d(x.v)));
}
EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet1cd,2>& kernel)
{
Packet2d tmp = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_HI);
kernel.packet[1].v = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_LO);
kernel.packet[0].v = tmp;
}
/* complex<float> follows */
template<> EIGEN_STRONG_INLINE Packet2cf pload <Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(pload<Packet4f>((const float*)from)); }
template<> EIGEN_STRONG_INLINE Packet2cf ploadu<Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cf(ploadu<Packet4f>((const float*)from)); }
template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((float*)to, from.v); }
template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((float*)to, from.v); }
template<> EIGEN_STRONG_INLINE std::complex<float> pfirst<Packet2cf>(const Packet2cf& a)
{
std::complex<float> EIGEN_ALIGN16 res[2];
pstore<std::complex<float> >(res, a);
return res[0];
}
#if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ < 12)
template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>& from)
{
Packet2cf res;
res.cd[0] = Packet1cd(vec_ld2f((const float *)&from));
res.cd[1] = res.cd[0];
return res;
}
#else
template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>& from)
{
Packet2cf res;
if((std::ptrdiff_t(&from) % 16) == 0)
res.v = pload<Packet4f>((const float *)&from);
else
res.v = ploadu<Packet4f>((const float *)&from);
res.v = vec_perm(res.v, res.v, p16uc_PSET64_HI);
return res;
}
#endif
template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, Index stride)
{
std::complex<float> EIGEN_ALIGN16 af[2];
af[0] = from[0*stride];
af[1] = from[1*stride];
return pload<Packet2cf>(af);
}
template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, Index stride)
{
std::complex<float> EIGEN_ALIGN16 af[2];
pstore<std::complex<float> >((std::complex<float> *) af, from);
to[0*stride] = af[0];
to[1*stride] = af[1];
}
template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(padd<Packet4f>(a.v, b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(psub<Packet4f>(a.v, b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a) { return Packet2cf(pnegate(Packet4f(a.v))); }
template<> EIGEN_STRONG_INLINE Packet2cf pand <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pand<Packet4f>(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf por <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(por<Packet4f>(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf pxor <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pxor<Packet4f>(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pandnot<Packet4f>(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<float>* from) { return pset1<Packet2cf>(*from); }
template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> * addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
#if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ < 12)
template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a)
{
Packet2cf res;
res.v.v4f[0] = pconj(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[0]))).v;
res.v.v4f[1] = pconj(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[1]))).v;
return res;
}
template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
{
Packet2cf res;
res.v.v4f[0] = pmul(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[0])), Packet1cd(reinterpret_cast<Packet2d>(b.v.v4f[0]))).v;
res.v.v4f[1] = pmul(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[1])), Packet1cd(reinterpret_cast<Packet2d>(b.v.v4f[1]))).v;
return res;
}
template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a)
{
Packet2cf res;
res.cd[0] = a.cd[1];
res.cd[1] = a.cd[0];
return res;
}
template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a)
{
std::complex<float> res;
Packet1cd b = padd<Packet1cd>(a.cd[0], a.cd[1]);
vec_st2f(b.v, (float*)&res);
return res;
}
template<> EIGEN_STRONG_INLINE Packet2cf preduxp<Packet2cf>(const Packet2cf* vecs)
{
PacketBlock<Packet2cf,2> transpose;
transpose.packet[0] = vecs[0];
transpose.packet[1] = vecs[1];
ptranspose(transpose);
return padd<Packet2cf>(transpose.packet[0], transpose.packet[1]);
}
template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const Packet2cf& a)
{
std::complex<float> res;
Packet1cd b = pmul<Packet1cd>(a.cd[0], a.cd[1]);
vec_st2f(b.v, (float*)&res);
return res;
}
template<int Offset>
struct palign_impl<Offset,Packet2cf>
{
static EIGEN_STRONG_INLINE void run(Packet2cf& first, const Packet2cf& second)
{
if (Offset == 1) {
first.cd[0] = first.cd[1];
first.cd[1] = second.cd[0];
}
}
};
template<> struct conj_helper<Packet2cf, Packet2cf, false,true>
{
EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
@ -337,15 +391,6 @@ template<> struct conj_helper<Packet2cf, Packet2cf, true,true>
};
EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cf,Packet4f)
EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet1cd,Packet2d)
template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
{
// TODO optimize it for AltiVec
Packet1cd res = conj_helper<Packet1cd,Packet1cd,false,true>().pmul(a,b);
Packet2d s = vec_madd(b.v, b.v, p2d_ZERO_);
return Packet1cd(pdiv(res.v, s + vec_perm(s, s, p16uc_REVERSE64)));
}
template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
{
@ -356,11 +401,6 @@ template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, con
return res;
}
EIGEN_STRONG_INLINE Packet1cd pcplxflip/*<Packet1cd>*/(const Packet1cd& x)
{
return Packet1cd(preverse(Packet2d(x.v)));
}
EIGEN_STRONG_INLINE Packet2cf pcplxflip/*<Packet2cf>*/(const Packet2cf& x)
{
Packet2cf res;
@ -369,13 +409,6 @@ EIGEN_STRONG_INLINE Packet2cf pcplxflip/*<Packet2cf>*/(const Packet2cf& x)
return res;
}
EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet1cd,2>& kernel)
{
Packet2d tmp = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_HI);
kernel.packet[1].v = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_LO);
kernel.packet[0].v = tmp;
}
EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel)
{
Packet1cd tmp = kernel.packet[0].cd[1];
@ -389,6 +422,139 @@ template<> EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, con
result.v = pblend<Packet4f>(ifPacket4, thenPacket.v, elsePacket.v);
return result;
}
#else
template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a) { return Packet2cf(pxor<Packet4f>(a.v, reinterpret_cast<Packet4f>(p4ui_CONJ_XOR))); }
template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
{
Packet4f a_re, a_im, prod, prod_im;
// Permute and multiply the real parts of a and b
a_re = vec_perm(a.v, a.v, p16uc_PSET32_WODD);
// Get the imaginary parts of a
a_im = vec_perm(a.v, a.v, p16uc_PSET32_WEVEN);
// multiply a_im * b and get the conjugate result
prod_im = a_im * b.v;
prod_im = pxor<Packet4f>(prod_im, reinterpret_cast<Packet4f>(p4ui_CONJ_XOR));
// permute back to a proper order
prod_im = vec_perm(prod_im, prod_im, p16uc_COMPLEX32_REV);
// multiply a_re * b, add prod_im
prod = pmadd<Packet4f>(a_re, b.v, prod_im);
return Packet2cf(prod);
}
template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a)
{
Packet4f rev_a;
rev_a = vec_perm(a.v, a.v, p16uc_COMPLEX32_REV2);
return Packet2cf(rev_a);
}
template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a)
{
Packet4f b;
b = vec_sld(a.v, a.v, 8);
b = padd<Packet4f>(a.v, b);
return pfirst<Packet2cf>(Packet2cf(b));
}
template<> EIGEN_STRONG_INLINE Packet2cf preduxp<Packet2cf>(const Packet2cf* vecs)
{
Packet4f b1, b2;
b1 = vec_sld(vecs[0].v, vecs[1].v, 8);
b2 = vec_sld(vecs[1].v, vecs[0].v, 8);
b2 = vec_sld(b2, b2, 8);
b2 = padd<Packet4f>(b1, b2);
return Packet2cf(b2);
}
template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const Packet2cf& a)
{
Packet4f b;
Packet2cf prod;
b = vec_sld(a.v, a.v, 8);
prod = pmul<Packet2cf>(a, Packet2cf(b));
return pfirst<Packet2cf>(prod);
}
template<int Offset>
struct palign_impl<Offset,Packet2cf>
{
static EIGEN_STRONG_INLINE void run(Packet2cf& first, const Packet2cf& second)
{
if (Offset==1)
{
first.v = vec_sld(first.v, second.v, 8);
}
}
};
template<> struct conj_helper<Packet2cf, Packet2cf, false,true>
{
EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
{ return padd(pmul(x,y),c); }
EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
{
return internal::pmul(a, pconj(b));
}
};
template<> struct conj_helper<Packet2cf, Packet2cf, true,false>
{
EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
{ return padd(pmul(x,y),c); }
EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
{
return internal::pmul(pconj(a), b);
}
};
template<> struct conj_helper<Packet2cf, Packet2cf, true,true>
{
EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
{ return padd(pmul(x,y),c); }
EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
{
return pconj(internal::pmul(a, b));
}
};
EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cf,Packet4f)
template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
{
// TODO optimize it for AltiVec
Packet2cf res = conj_helper<Packet2cf,Packet2cf,false,true>().pmul(a, b);
Packet4f s = pmul<Packet4f>(b.v, b.v);
return Packet2cf(pdiv(res.v, padd<Packet4f>(s, vec_perm(s, s, p16uc_COMPLEX32_REV))));
}
template<> EIGEN_STRONG_INLINE Packet2cf pcplxflip<Packet2cf>(const Packet2cf& x)
{
return Packet2cf(vec_perm(x.v, x.v, p16uc_COMPLEX32_REV));
}
EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel)
{
Packet4f tmp = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_HI);
kernel.packet[1].v = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_LO);
kernel.packet[0].v = tmp;
}
template<> EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, const Packet2cf& thenPacket, const Packet2cf& elsePacket) {
Packet2cf result;
result.v = reinterpret_cast<Packet4f>(pblend<Packet2d>(ifPacket, reinterpret_cast<Packet2d>(thenPacket.v), reinterpret_cast<Packet2d>(elsePacket.v)));
return result;
}
#endif
} // end namespace internal

105
Eigen/src/Core/arch/ZVector/MathFunctions.h
View File

@ -20,6 +20,50 @@ namespace Eigen {
namespace internal {
#if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ >= 12)
static _EIGEN_DECLARE_CONST_Packet4f(1 , 1.0f);
static _EIGEN_DECLARE_CONST_Packet4f(half, 0.5f);
static _EIGEN_DECLARE_CONST_Packet4i(0x7f, 0x7f);
static _EIGEN_DECLARE_CONST_Packet4i(23, 23);
static _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(inv_mant_mask, ~0x7f800000);
/* the smallest non denormalized float number */
static _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(min_norm_pos, 0x00800000);
static _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(minus_inf, 0xff800000); // -1.f/0.f
static _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(minus_nan, 0xffffffff);
/* natural logarithm computed for 4 simultaneous float
return NaN for x <= 0
*/
static _EIGEN_DECLARE_CONST_Packet4f(cephes_SQRTHF, 0.707106781186547524f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p0, 7.0376836292E-2f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p1, - 1.1514610310E-1f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p2, 1.1676998740E-1f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p3, - 1.2420140846E-1f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p4, + 1.4249322787E-1f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p5, - 1.6668057665E-1f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p6, + 2.0000714765E-1f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p7, - 2.4999993993E-1f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p8, + 3.3333331174E-1f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_q1, -2.12194440e-4f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_q2, 0.693359375f);
static _EIGEN_DECLARE_CONST_Packet4f(exp_hi, 88.3762626647950f);
static _EIGEN_DECLARE_CONST_Packet4f(exp_lo, -88.3762626647949f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_LOG2EF, 1.44269504088896341f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_C1, 0.693359375f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_C2, -2.12194440e-4f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p0, 1.9875691500E-4f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p1, 1.3981999507E-3f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p2, 8.3334519073E-3f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p3, 4.1665795894E-2f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p4, 1.6666665459E-1f);
static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p5, 5.0000001201E-1f);
#endif
static _EIGEN_DECLARE_CONST_Packet2d(1 , 1.0);
static _EIGEN_DECLARE_CONST_Packet2d(2 , 2.0);
static _EIGEN_DECLARE_CONST_Packet2d(half, 0.5);
@ -93,40 +137,91 @@ Packet2d pexp<Packet2d>(const Packet2d& _x)
}
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
Packet4f pexp<Packet4f>(const Packet4f& x)
Packet4f pexp<Packet4f>(const Packet4f& _x)
{
#if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ >= 12)
/*
Packet4f x = _x;
Packet4f tmp, fx;
Packet4i emm0;
// clamp x
x = pmax(pmin(x, p4f_exp_hi), p4f_exp_lo);
// express exp(x) as exp(g + n*log(2))
fx = pmadd(x, p4f_cephes_LOG2EF, p4f_half);
fx = pfloor(fx);
tmp = pmul(fx, p4f_cephes_exp_C1);
Packet4f z = pmul(fx, p4f_cephes_exp_C2);
x = psub(x, tmp);
x = psub(x, z);
z = pmul(x,x);
Packet4f y = p4f_cephes_exp_p0;
y = pmadd(y, x, p4f_cephes_exp_p1);
y = pmadd(y, x, p4f_cephes_exp_p2);
y = pmadd(y, x, p4f_cephes_exp_p3);
y = pmadd(y, x, p4f_cephes_exp_p4);
y = pmadd(y, x, p4f_cephes_exp_p5);
y = pmadd(y, z, x);
y = padd(y, p4f_1);
// build 2^n
emm0 = vec_cts(fx, 0);
emm0 = emm0 + p4i_0x7f;
emm0 = emm0 << reinterpret_cast<Packet4i>(p4i_23);
// Altivec's max & min operators just drop silent NaNs. Check NaNs in
// inputs and return them unmodified.
Packet4ui isnumber_mask = reinterpret_cast<Packet4ui>(vec_cmpeq(_x, _x));
return vec_sel(_x, pmax(pmul(y, reinterpret_cast<Packet4f>(emm0)), _x),
isnumber_mask);*/
return _x;
#else
Packet4f res;
res.v4f[0] = pexp<Packet2d>(x.v4f[0]);
res.v4f[1] = pexp<Packet2d>(x.v4f[1]);
res.v4f[0] = pexp<Packet2d>(_x.v4f[0]);
res.v4f[1] = pexp<Packet2d>(_x.v4f[1]);
return res;
#endif
}
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
Packet2d psqrt<Packet2d>(const Packet2d& x)
{
return __builtin_s390_vfsqdb(x);
return vec_sqrt(x);
}
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
Packet4f psqrt<Packet4f>(const Packet4f& x)
{
Packet4f res;
#if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ >= 12)
res = vec_sqrt(x);
#else
res.v4f[0] = psqrt<Packet2d>(x.v4f[0]);
res.v4f[1] = psqrt<Packet2d>(x.v4f[1]);
#endif
return res;
}
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
Packet2d prsqrt<Packet2d>(const Packet2d& x) {
// Unfortunately we can't use the much faster mm_rqsrt_pd since it only provides an approximation.
return pset1<Packet2d>(1.0) / psqrt<Packet2d>(x);
}
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
Packet4f prsqrt<Packet4f>(const Packet4f& x) {
Packet4f res;
#if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ >= 12)
res = pset1<Packet4f>(1.0) / psqrt<Packet4f>(x);
#else
res.v4f[0] = prsqrt<Packet2d>(x.v4f[0]);
res.v4f[1] = prsqrt<Packet2d>(x.v4f[1]);
#endif
return res;
}

840
Eigen/src/Core/arch/ZVector/PacketMath.h
View File

File diff suppressed because it is too large Load Diff

2
Eigen/src/Core/util/Macros.h
View File

@ -412,7 +412,7 @@
// Does the compiler support type_trais?
#ifndef EIGEN_HAS_TYPE_TRAITS
#if EIGEN_MAX_CPP_VER>=11 && (EIGEN_HAS_CXX11 || __has_feature(is_enum) || EIGEN_COMP_MSVC >= 1700)
#if EIGEN_MAX_CPP_VER>=11 && (EIGEN_HAS_CXX11 || EIGEN_COMP_MSVC >= 1700)
#define EIGEN_HAS_TYPE_TRAITS 1
#define EIGEN_INCLUDE_TYPE_TRAITS
#else

12
Eigen/src/Core/util/Memory.h
View File

@ -493,7 +493,7 @@ template<typename T> struct smart_copy_helper<T,true> {
IntPtr size = IntPtr(end)-IntPtr(start);
if(size==0) return;
eigen_internal_assert(start!=0 && end!=0 && target!=0);
memcpy(target, start, size);
std::memcpy(target, start, size);
}
};
@ -696,7 +696,15 @@ template<typename T> void swap(scoped_array<T> &a,scoped_array<T> &b)
/** \class aligned_allocator
* \ingroup Core_Module
*
* \brief STL compatible allocator to use with with 16 byte aligned types
* \brief STL compatible allocator to use with types requiring a non standrad alignment.
*
* The memory is aligned as for dynamically aligned matrix/array types such as MatrixXd.
* By default, it will thus provide at least 16 bytes alignment and more in following cases:
* - 32 bytes alignment if AVX is enabled.
* - 64 bytes alignment if AVX512 is enabled.
*
* This can be controled using the \c EIGEN_MAX_ALIGN_BYTES macro as documented
* \link TopicPreprocessorDirectivesPerformance there \endlink.
*
* Example:
* \code

6
Eigen/src/Core/util/XprHelper.h
View File

@ -36,14 +36,16 @@ inline IndexDest convert_index(const IndexSrc& idx) {
// true if T can be considered as an integral index (i.e., and integral type or enum)
template<typename T> struct is_valid_index_type
: std::integral_constant<bool,
{
enum { value =
#if EIGEN_HAS_TYPE_TRAITS
internal::is_integral<T>::value || std::is_enum<T>::value
#else
// without C++11, we use is_convertible to Index instead of is_integral in order to treat enums as Index.
internal::is_convertible<T,Index>::value
#endif
> {};
};
};
// promote_scalar_arg is an helper used in operation between an expression and a scalar, like:
// expression * scalar

2
Eigen/src/SparseCore/AmbiVector.h
View File

@ -94,7 +94,7 @@ class AmbiVector
Index allocSize = m_allocatedElements * sizeof(ListEl);
allocSize = (allocSize + sizeof(Scalar) - 1)/sizeof(Scalar);
Scalar* newBuffer = new Scalar[allocSize];
memcpy(newBuffer, m_buffer, copyElements * sizeof(ListEl));
std::memcpy(newBuffer, m_buffer, copyElements * sizeof(ListEl));
delete[] m_buffer;
m_buffer = newBuffer;
}

2
Eigen/src/plugins/IndexedViewMethods.h
View File

@ -248,6 +248,8 @@ operator()(const IndicesT (&indices)[IndicesN]) EIGEN_INDEXED_VIEW_METHOD_CONST
*
* For 1D vectors and arrays, you better use the operator()(const Indices&) overload, which behave the same way but taking a single parameter.
*
* See also this <a href="https://stackoverflow.com/questions/46110917/eigen-replicate-items-along-one-dimension-without-useless-allocations">question</a> and its answer for an example of how to duplicate coefficients.
*
* \sa operator()(const Indices&), class Block, class IndexedView, DenseBase::block(Index,Index,Index,Index)
*/
template<typename RowIndices, typename ColIndices>