mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2024-12-27 07:29:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

Uses _mm512_abs_pd for Packet8d pabs

Browse Source
This commit is contained in:
Steve Bronder 2021-03-18 15:47:52 +00:00 committed by Rasmus Munk Larsen
parent 0cc9b5eb40
commit f019b97aca
1 changed files with 6 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
Eigen/src/Core/arch/AVX512/PacketMath.h
View File

@ -139,6 +139,7 @@ template<> struct packet_traits<double> : default_packet_traits
size = 8, size = 8,
HasHalfPacket = 1, HasHalfPacket = 1,
#if EIGEN_GNUC_AT_LEAST(5, 3) || (!EIGEN_COMP_GNUC_STRICT) #if EIGEN_GNUC_AT_LEAST(5, 3) || (!EIGEN_COMP_GNUC_STRICT)
HasAbs = 1,
HasLog = 1, HasLog = 1,
HasSqrt = EIGEN_FAST_MATH, HasSqrt = EIGEN_FAST_MATH,
HasRsqrt = EIGEN_FAST_MATH, HasRsqrt = EIGEN_FAST_MATH,
@ -888,9 +889,7 @@ template<> EIGEN_STRONG_INLINE Packet16f pabs(const Packet16f& a)
} }
template <> template <>
EIGEN_STRONG_INLINE Packet8d pabs(const Packet8d& a) { EIGEN_STRONG_INLINE Packet8d pabs(const Packet8d& a) {
// _mm512_abs_ps intrinsic not found, so hack around it return _mm512_abs_pd(a);
return _mm512_castsi512_pd(_mm512_and_si512(_mm512_castpd_si512(a),
_mm512_set1_epi64(0x7fffffffffffffff)));
} }
template<> template<>
@ -900,7 +899,7 @@ EIGEN_STRONG_INLINE Packet16f pfrexp<Packet16f>(const Packet16f& a, Packet16f& e
// Extract exponent without existence of Packet8l. // Extract exponent without existence of Packet8l.
template<> template<>
EIGEN_STRONG_INLINE EIGEN_STRONG_INLINE
Packet8d pfrexp_generic_get_biased_exponent(const Packet8d& a) { Packet8d pfrexp_generic_get_biased_exponent(const Packet8d& a) {
const Packet8d cst_exp_mask = pset1frombits<Packet8d>(static_cast<uint64_t>(0x7ff0000000000000ull)); const Packet8d cst_exp_mask = pset1frombits<Packet8d>(static_cast<uint64_t>(0x7ff0000000000000ull));
#ifdef EIGEN_VECTORIZE_AVX512DQ #ifdef EIGEN_VECTORIZE_AVX512DQ
@ -923,18 +922,18 @@ template<> EIGEN_STRONG_INLINE Packet8d pldexp<Packet8d>(const Packet8d& a, cons
// Clamp exponent to [-2099, 2099] // Clamp exponent to [-2099, 2099]
const Packet8d max_exponent = pset1<Packet8d>(2099.0); const Packet8d max_exponent = pset1<Packet8d>(2099.0);
const Packet8i e = _mm512_cvtpd_epi32(pmin(pmax(exponent, pnegate(max_exponent)), max_exponent)); const Packet8i e = _mm512_cvtpd_epi32(pmin(pmax(exponent, pnegate(max_exponent)), max_exponent));
// Split 2^e into four factors and multiply. // Split 2^e into four factors and multiply.
const Packet8i bias = pset1<Packet8i>(1023); const Packet8i bias = pset1<Packet8i>(1023);
Packet8i b = parithmetic_shift_right<2>(e); // floor(e/4) Packet8i b = parithmetic_shift_right<2>(e); // floor(e/4)
// 2^b // 2^b
Packet8i hi = _mm256_shuffle_epi32(padd(b, bias), _MM_SHUFFLE(3, 1, 2, 0)); Packet8i hi = _mm256_shuffle_epi32(padd(b, bias), _MM_SHUFFLE(3, 1, 2, 0));
Packet8i lo = _mm256_slli_epi64(hi, 52); Packet8i lo = _mm256_slli_epi64(hi, 52);
hi = _mm256_slli_epi64(_mm256_srli_epi64(hi, 32), 52); hi = _mm256_slli_epi64(_mm256_srli_epi64(hi, 32), 52);
Packet8d c = _mm512_castsi512_pd(_mm512_inserti64x4(_mm512_castsi256_si512(lo), hi, 1)); Packet8d c = _mm512_castsi512_pd(_mm512_inserti64x4(_mm512_castsi256_si512(lo), hi, 1));
Packet8d out = pmul(pmul(pmul(a, c), c), c); // a * 2^(3b) Packet8d out = pmul(pmul(pmul(a, c), c), c); // a * 2^(3b)
// 2^(e - 3b) // 2^(e - 3b)
b = psub(psub(psub(e, b), b), b); // e - 3b b = psub(psub(psub(e, b), b), b); // e - 3b
hi = _mm256_shuffle_epi32(padd(b, bias), _MM_SHUFFLE(3, 1, 2, 0)); hi = _mm256_shuffle_epi32(padd(b, bias), _MM_SHUFFLE(3, 1, 2, 0));