GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-10 10:49:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

Revert "Fix Half NaN definition and test."

Browse Source 
This reverts commit c770746d709686ef2b8b652616d9232f9b028e78.
This commit is contained in:
Rasmus Munk Larsen 2020-11-24 12:57:28 -08:00
parent c770746d70
commit 79818216ed
8 changed files with 11 additions and 176 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
Eigen/src/Core/GenericPacketMath.h
View File

@ -539,20 +539,6 @@ inline void pbroadcast2(const typename unpacket_traits<Packet>::type *a,
template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet
plset(const typename unpacket_traits<Packet>::type& a) { return a; }
/** \internal \returns a packet with constant coefficients \a a, e.g.: (x, 0, x, 0),
where x is the value of all 1-bits. */
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
peven_mask(const Packet& /*a*/) {
typedef typename unpacket_traits<Packet>::type Scalar;
const size_t n = unpacket_traits<Packet>::size;
Scalar elements[n];
for(size_t i = 0; i < n; ++i) {
memset(elements+i, ((i & 1) == 0 ? 0xff : 0), sizeof(Scalar));
}
return ploadu<Packet>(elements);
}
/** \internal copy the packet \a from to \a *to, \a to must be 16 bytes aligned */
template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstore(Scalar* to, const Packet& from)
{ (*to) = from; }

36
Eigen/src/Core/arch/AVX/Complex.h
View File

@ -38,7 +38,6 @@ template<> struct packet_traits<std::complex<float> > : default_packet_traits
HasMul = 1,
HasDiv = 1,
HasNegate = 1,
HasSqrt = 1,
HasAbs = 0,
HasAbs2 = 0,
HasMin = 0,
@ -48,18 +47,7 @@ template<> struct packet_traits<std::complex<float> > : default_packet_traits
};
#endif
template<> struct unpacket_traits<Packet4cf> {
typedef std::complex<float> type;
typedef Packet2cf half;
typedef Packet8f real;
enum {
size=4,
alignment=Aligned32,
vectorizable=true,
masked_load_available=false,
masked_store_available=false
};
};
template<> struct unpacket_traits<Packet4cf> { typedef std::complex<float> type; enum {size=4, alignment=Aligned32, vectorizable=true, masked_load_available=false, masked_store_available=false}; typedef Packet2cf half; };
template<> EIGEN_STRONG_INLINE Packet4cf padd<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_add_ps(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet4cf psub<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_sub_ps(a.v,b.v)); }
@ -240,7 +228,6 @@ template<> struct packet_traits<std::complex<double> > : default_packet_traits
HasMul = 1,
HasDiv = 1,
HasNegate = 1,
HasSqrt = 1,
HasAbs = 0,
HasAbs2 = 0,
HasMin = 0,
@ -250,18 +237,7 @@ template<> struct packet_traits<std::complex<double> > : default_packet_traits
};
#endif
template<> struct unpacket_traits<Packet2cd> {
typedef std::complex<double> type;
typedef Packet1cd half;
typedef Packet4d real;
enum {
size=2,
alignment=Aligned32,
vectorizable=true,
masked_load_available=false,
masked_store_available=false
};
};
template<> struct unpacket_traits<Packet2cd> { typedef std::complex<double> type; enum {size=2, alignment=Aligned32, vectorizable=true, masked_load_available=false, masked_store_available=false}; typedef Packet1cd half; };
template<> EIGEN_STRONG_INLINE Packet2cd padd<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_add_pd(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cd psub<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_sub_pd(a.v,b.v)); }
@ -423,14 +399,6 @@ ptranspose(PacketBlock<Packet2cd,2>& kernel) {
kernel.packet[0].v = tmp;
}
template<> EIGEN_STRONG_INLINE Packet2cd psqrt<Packet2cd>(const Packet2cd& a) {
return psqrt_complex<Packet2cd>(a);
}
template<> EIGEN_STRONG_INLINE Packet4cf psqrt<Packet4cf>(const Packet4cf& a) {
return psqrt_complex<Packet4cf>(a);
}
} // end namespace internal
} // end namespace Eigen

17
Eigen/src/Core/arch/AVX512/Complex.h
View File

@ -37,7 +37,6 @@ template<> struct packet_traits<std::complex<float> > : default_packet_traits
HasMul = 1,
HasDiv = 1,
HasNegate = 1,
HasSqrt = 1,
HasAbs = 0,
HasAbs2 = 0,
HasMin = 0,
@ -48,8 +47,6 @@ template<> struct packet_traits<std::complex<float> > : default_packet_traits
template<> struct unpacket_traits<Packet8cf> {
typedef std::complex<float> type;
typedef Packet4cf half;
typedef Packet16f real;
enum {
size = 8,
alignment=unpacket_traits<Packet16f>::alignment,
@ -57,6 +54,7 @@ template<> struct unpacket_traits<Packet8cf> {
masked_load_available=false,
masked_store_available=false
};
typedef Packet4cf half;
};
template<> EIGEN_STRONG_INLINE Packet8cf ptrue<Packet8cf>(const Packet8cf& a) { return Packet8cf(ptrue(Packet16f(a.v))); }
@ -225,7 +223,6 @@ template<> struct packet_traits<std::complex<double> > : default_packet_traits
HasMul = 1,
HasDiv = 1,
HasNegate = 1,
HasSqrt = 1,
HasAbs = 0,
HasAbs2 = 0,
HasMin = 0,
@ -236,8 +233,6 @@ template<> struct packet_traits<std::complex<double> > : default_packet_traits
template<> struct unpacket_traits<Packet4cd> {
typedef std::complex<double> type;
typedef Packet2cd half;
typedef Packet8d real;
enum {
size = 4,
alignment = unpacket_traits<Packet8d>::alignment,
@ -245,6 +240,7 @@ template<> struct unpacket_traits<Packet4cd> {
masked_load_available=false,
masked_store_available=false
};
typedef Packet2cd half;
};
template<> EIGEN_STRONG_INLINE Packet4cd padd<Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(_mm512_add_pd(a.v,b.v)); }
@ -441,15 +437,8 @@ ptranspose(PacketBlock<Packet4cd,4>& kernel) {
kernel.packet[0] = Packet4cd(_mm512_shuffle_f64x2(T0, T2, (shuffle_mask<0,2,0,2>::mask))); // [a0 b0 c0 d0]
}
template<> EIGEN_STRONG_INLINE Packet4cd psqrt<Packet4cd>(const Packet4cd& a) {
return psqrt_complex<Packet4cd>(a);
}
template<> EIGEN_STRONG_INLINE Packet8cf psqrt<Packet8cf>(const Packet8cf& a) {
return psqrt_complex<Packet8cf>(a);
}
} // end namespace internal
} // end namespace Eigen
#endif // EIGEN_COMPLEX_AVX512_H

56
Eigen/src/Core/arch/Default/GenericPacketMathFunctions.h
View File

@ -643,62 +643,6 @@ Packet pcos_float(const Packet& x)
return psincos_float<false>(x);
}
template<typename Packet>
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
EIGEN_UNUSED
Packet psqrt_complex(const Packet& a) {
typedef typename unpacket_traits<Packet>::type Scalar;
typedef typename Scalar::value_type RealScalar;
typedef typename unpacket_traits<Packet>::real RealPacket;
// Computes the principal sqrt of the complex numbers. For clarity, the comments
// below spell out the steps, assuming Packet contains 2 complex numbers, e.g.
// a = [a0_r, a0_i, a1_r, a1_i]
// In other words, the function computes b = [b0_r, b0_i, b1_r, b1_i] such that
// (b0_r + i*b0_i)^2 = a0_r + i*a0_i, and
// (b1_r + i*b1_i)^2 = a1_r + i*a1_i .
// Step 1. Compute l = [l0, l0, l1, l1], where
// l0 = sqrt(a0_r^2 + a0_i^2), l1 = sqrt(a1_r^2 + a1_i^2)
// To avoid over- and underflow, we use the stable formula for each hypotenuse
// l0 = (x0 == 0 ? x0 : x0 * sqrt(1 + (y0/x0)**2)),
// where x0 = max(|a0_r|, |a0_i|), y0 = min(|a0_r|, |a0_i|)
// and similarly for l1.
Packet a_flip = pcplxflip(a);
Packet zero_mask;
zero_mask.v = pcmp_eq(a.v, pzero(a.v));
RealPacket a_abs = pabs(a.v); // [|a0_i|, |a0_r|, |a1_i|, |a1_r|]
RealPacket a_abs_flip = pabs(a_flip.v); // [|a0_i|, |a0_r|, |a1_i|, |a1_r|]
RealPacket a_max = pmax(a_abs, a_abs_flip);
RealPacket a_min = pmin(a_abs, a_abs_flip);
RealPacket r = pdiv(a_min, a_max);
RealPacket one = pset1<RealPacket>(RealScalar(1));
RealPacket l = pmul(a_max, psqrt(padd(one, pmul(r, r)))); // [l0, l0, l1, l1]
// Set l to zero if both real and imaginary parts are zero.
l = pandnot(l, pand(zero_mask.v, pcplxflip(zero_mask).v));
// Step 2. Compute
// [ sqrt((l0 + a0_r)/2), sqrt((l0 - a0_r)/2),
// sqrt((l1 + a1_r)/2), sqrt((l1 - a1_r)/2) ]
Packet real_mask;
real_mask.v = peven_mask(real_mask.v);
Packet a_real = pand(a, real_mask);
l = padd(l, a_real.v);
l = psub(l, pcplxflip(a_real).v);
l = psqrt(pmul(l, pset1<RealPacket>(RealScalar(0.5))));
// If imag(a) is zero, we mask out the imaginary part, which should be zero.
l = pandnot(l, pandnot(zero_mask.v, real_mask.v));
//Step 3. Apply the sign of the imaginary parts of a to get the final result:
// b = [ sqrt((l0 + a0_r)/2), sign(a0_i)*sqrt((l0 - a0_r)/2),
// sqrt((l1 + a1_r)/2), sign(a1_i)*sqrt((l1 - a1_r)/2) ]
RealPacket imag_sign_mask = pset1<Packet>(Scalar(RealScalar(0.0), RealScalar(-0.0))).v;
RealPacket imag_signs = pand<RealPacket>(a.v, imag_sign_mask);
Packet result = Packet(pxor<RealPacket>(l, imag_signs));
return result;
}
/* polevl (modified for Eigen)
*
* Evaluate polynomial

7
Eigen/src/Core/arch/Default/GenericPacketMathFunctionsFwd.h
View File

@ -70,15 +70,8 @@ EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
EIGEN_UNUSED
Packet pcos_float(const Packet& x);
/** \internal \returns sqrt(x) for complex types */
template<typename Packet>
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
EIGEN_UNUSED
Packet psqrt_complex(const Packet& x);
template <typename Packet, int N> struct ppolevl;
} // end namespace internal
} // end namespace Eigen

38
Eigen/src/Core/arch/SSE/Complex.h
View File

@ -40,7 +40,6 @@ template<> struct packet_traits<std::complex<float> > : default_packet_traits
HasMul = 1,
HasDiv = 1,
HasNegate = 1,
HasSqrt = 1,
HasAbs = 0,
HasAbs2 = 0,
HasMin = 0,
@ -51,18 +50,7 @@ template<> struct packet_traits<std::complex<float> > : default_packet_traits
};
#endif
template<> struct unpacket_traits<Packet2cf> {
typedef std::complex<float> type;
typedef Packet2cf half;
typedef Packet4f real;
enum {
size=2,
alignment=Aligned16,
vectorizable=true,
masked_load_available=false,
masked_store_available=false
};
};
template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2, alignment=Aligned16, vectorizable=true, masked_load_available=false, masked_store_available=false}; typedef Packet2cf half; };
template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_add_ps(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_sub_ps(a.v,b.v)); }
@ -95,6 +83,7 @@ template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, con
}
template<> EIGEN_STRONG_INLINE Packet2cf ptrue <Packet2cf>(const Packet2cf& a) { return Packet2cf(ptrue(Packet4f(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)); }
@ -266,7 +255,6 @@ template<> struct packet_traits<std::complex<double> > : default_packet_traits
HasMul = 1,
HasDiv = 1,
HasNegate = 1,
HasSqrt = 1,
HasAbs = 0,
HasAbs2 = 0,
HasMin = 0,
@ -276,18 +264,7 @@ template<> struct packet_traits<std::complex<double> > : default_packet_traits
};
#endif
template<> struct unpacket_traits<Packet1cd> {
typedef std::complex<double> type;
typedef Packet1cd half;
typedef Packet2d real;
enum {
size=1,
alignment=Aligned16,
vectorizable=true,
masked_load_available=false,
masked_store_available=false
};
};
template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type; enum {size=1, alignment=Aligned16, vectorizable=true, masked_load_available=false, masked_store_available=false}; typedef Packet1cd half; };
template<> EIGEN_STRONG_INLINE Packet1cd padd<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_add_pd(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet1cd psub<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_sub_pd(a.v,b.v)); }
@ -449,15 +426,8 @@ template<> EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, co
return Packet2cf(_mm_castpd_ps(result));
}
template<> EIGEN_STRONG_INLINE Packet1cd psqrt<Packet1cd>(const Packet1cd& a) {
return psqrt_complex<Packet1cd>(a);
}
template<> EIGEN_STRONG_INLINE Packet2cf psqrt<Packet2cf>(const Packet2cf& a) {
return psqrt_complex<Packet2cf>(a);
}
} // end namespace internal
} // end namespace Eigen
#endif // EIGEN_COMPLEX_SSE_H

4
Eigen/src/Core/arch/SSE/PacketMath.h
View File

@ -267,10 +267,6 @@ template<> EIGEN_STRONG_INLINE Packet16b pset1<Packet16b>(const bool& from) {
template<> EIGEN_STRONG_INLINE Packet4f pset1frombits<Packet4f>(unsigned int from) { return _mm_castsi128_ps(pset1<Packet4i>(from)); }
template<> EIGEN_STRONG_INLINE Packet2d pset1frombits<Packet2d>(uint64_t from) { return _mm_castsi128_pd(_mm_set1_epi64x(from)); }
template<> EIGEN_STRONG_INLINE Packet4f peven_mask(const Packet4f& /*a*/) {
return Packet4f(_mm_set_epi32(0, 0xffffffff, 0, 0xffffffff));
}
template<> EIGEN_STRONG_INLINE Packet4f pzero(const Packet4f& /*a*/) { return _mm_setzero_ps(); }
template<> EIGEN_STRONG_INLINE Packet2d pzero(const Packet2d& /*a*/) { return _mm_setzero_pd(); }
template<> EIGEN_STRONG_INLINE Packet4i pzero(const Packet4i& /*a*/) { return _mm_setzero_si128(); }

15
test/packetmath.cpp
View File

@ -473,6 +473,8 @@ void packetmath() {
CHECK_CWISE3_IF(true, internal::pselect, internal::pselect);
}
CHECK_CWISE1_IF(PacketTraits::HasSqrt, numext::sqrt, internal::psqrt);
for (int i = 0; i < size; ++i) {
data1[i] = internal::random<Scalar>();
}
@ -484,11 +486,6 @@ void packetmath() {
packetmath_boolean_mask_ops<Scalar, Packet>();
packetmath_pcast_ops_runner<Scalar, Packet>::run();
packetmath_minus_zero_add<Scalar, Packet>();
for (int i = 0; i < size; ++i) {
data1[i] = numext::abs(internal::random<Scalar>());
}
CHECK_CWISE1_IF(PacketTraits::HasSqrt, numext::sqrt, internal::psqrt);
}
template <typename Scalar, typename Packet>
@ -969,8 +966,6 @@ void test_conj_helper(Scalar* data1, Scalar* data2, Scalar* ref, Scalar* pval) {
template <typename Scalar, typename Packet>
void packetmath_complex() {
typedef internal::packet_traits<Scalar> PacketTraits;
typedef typename Scalar::value_type RealScalar;
const int PacketSize = internal::unpacket_traits<Packet>::size;
const int size = PacketSize * 4;
@ -989,17 +984,11 @@ void packetmath_complex() {
test_conj_helper<Scalar, Packet, true, false>(data1, data2, ref, pval);
test_conj_helper<Scalar, Packet, true, true>(data1, data2, ref, pval);
// Test pcplxflip.
{
for (int i = 0; i < PacketSize; ++i) ref[i] = Scalar(std::imag(data1[i]), std::real(data1[i]));
internal::pstore(pval, internal::pcplxflip(internal::pload<Packet>(data1)));
VERIFY(test::areApprox(ref, pval, PacketSize) && "pcplxflip");
}
for (int i = 0; i < size; ++i) {
data1[i] = Scalar(internal::random<RealScalar>(), internal::random<RealScalar>());
}
CHECK_CWISE1_IF(PacketTraits::HasSqrt, numext::sqrt, internal::psqrt);
}
template <typename Scalar, typename Packet>